Speed: Radiology, Errors, and Speed



by R Jasinski



This article discusses the effects of culturally conditioned speed norms on perception in the medical workplace, as it impacts image interpretation. The article draws examples from the non-medical world to make the point that the causes and effects of speed, as well as the way it is conditioned in the workplace, are not unique to medicine. The non-medical world has lessons and consolation to offer those concerned about perception in the medical milieu. The radiological literature concerning perception, speed, and factors affecting perception is also reviewed and commented on.

The reader will be repaid for slowing down to look in detail at the references provided here - this article being only a broad treatment of more complex matters, and serving as an annotated/referenced guide to more extensive material that treats in more detail the complexities of perception and speed. Page numbers are often provided to make such a survey of references more convenient, though not provided to encourage too speedy a glance through material that merits more time and attention. It might be a nice change of pace for imagers to read material whose half-life is longer than that of most imaging literature, which can sometimes speedily decay.

KEY WORDS - Medical/Radiological: Time, speed, time on task, accuracy, radiology, perception, perceptual, recognition, radiological accuracy, work speed, quality, time/motion studies, ROC curves, satisfaction of search, workload, false positives, false negatives, true positives, true negatives, time truncation, search truncation, double readings, structured reports, redemptive redundancies, imaging interpretation, radiographic interpretation, diagnostic radiology, diagnostic radiologists, diagnostic procedures, comparison films, mammography, cross sectional imaging, MRI, CT, ultrasound, radiologists' workload, centralized image reading, radiology workload, radiology groups' workload, detection of abnormalities, roentgen interpretation, Medical Image Perception Society, radiological reporting, interpretation of radiographs, lung nodules, breast cancer, lung cancer, lung tumor, breast tumor, radiologists' performance, radiographic examinations, interpretation, diagnostic accuracy. structured reporting, cine viewing, film viewing.

KEY WORDS - Error - disagreement - competence, errors, mistakes, medical error, medical errors, medical mistakes, radiology errors, interpretive errors, perceptual errors, missed diagnosis, missed diagnoses, alliterative errors, quality assessment, quality improvement, error in radiology, radiology quality assurance, context bias, variation, evaluation, reader error, expertise, quality improvements, measuring professional quality, analysis of errors, causes of error, malpractice, negligence, liability, medicolegal, perceptual errors, film reader error, viewing time, retrospectoscope, missed cancer, missed lesions, unreported cancers, retrospective, retrospectively, missed radiologic diagnosis, search errors, free lessons, near misses, reducing errors, retrospective error, prospective error, roentgen diagnosis, interpreting radiographs, radiological images, normative error, technical error.

KEY WORDS - Search, Imaging: aerial photographs, photoreconnaisance, search and rescue, search time, CAD, computer-aided diagnosis, mimicry, camouflage in nature, deceptions, perception research, visual search, visual expertise, human information processing, digital image processing, visual scanning, pattern recognition, use of the eyes, perception tests, comparison scans, search behavior, image perception research, vigilance, attention, humility, problems of perception, clue, image quality, observer performance, roentgen image perception, tumor detection, perception of visual information, receiver operating characteristic curve, ROC curve, image viewing habits, discipline, lesion conspicuity, detectability, threshold, detection, subtle, detectability, computerized detection, artificial neural networks, observer performance, optical illusion, optical puzzles, ambiguity, expertise, observer performance, interpreting, experience, master, double reading, worksheet, expert interpretation, specialist, medical imaging, visual psychophysics, consultation.

KEY WORDS - Organization process, Psychology / Sociology of Perception, Ethics: Pareto Principle, 80/20 rule, beginner's mind, attention deficit disorder, short attention span, accountability, cognitive dissonance, anonymity, accountability, work speed, vigilance, attention, humility, Abu Ghraib, Milgram, Zimbardo, prisoner, guards, Milgram experiment, Tuskegee syphilis study, Willowbrook hepatitis study, group pressures, Groupthink, consensus, premature consensus, whistleblower, whistleblowers, Titanic, Greenville, Andrea Doria, shuttle disasters, Columbia, Challenger, NASA, organizational culture, psychology, ethics, ethical, abuse, consultation.

KEY WORDS - Sources / Literary: Maimonides, The Preservation of Youth: Essays on Health, Barbara McClintock, Evelyn Fox Keller, A Feeling for the Organism, Slowness, Milan Kundera, Zen and the Art of Motorcycle Maintenance: an Inquiry into Values, Robert Pirsig, Fahrenheit 451, Ray Bradbury , Pilgrim at Tinker Creek, Annie Dillard, Zen in the Art of Archery, Zen in the Art of Flower Arrangement, Eugen Herrigel, Gustie Herrigel, Achilles in Vietnam, Odysseus in America: Combat Trauma and the Trials of Homecoming, Jonathan Shay, Marine Sniper, Carlos Hathcock, Charles Henderson, The Preservation of Youth, Maimonides, On The Internet, Hubert Dreyfus, distance learning, Candide, Voltaire, Pangloss, Forgive and Remember: Managing Medical Failure, Charles Bosk, Richard Feynman, The Pleasure of Finding Things Out, Groupthink, Irving Janis, House of God, Samuel Shem, The Mountains, Stewart Edward White, poetry, poetry and medicine, Sobel, Elata, Empathic Practitioner, More, Milligan, Connelly, Jane Goodall, Reason for Hope: A Spiritual Journey, Parabola, Web of Life, Louis Pasteur, Dirty Harry, Clint Eastwood, Jill Ker Conway, True North, Faster, Gleick J., Ochberg FM, Post-Traumatic Therapy, The Making of a Surgeon, Nolen WA, Serpico, The Far Side, Gary Larson, That Others May Live, Brehm J., para-rescue jumpers, Catch 22, Heller J., Harvard and the Unabomber, Alston Chase, Judith Herman, Trauma and Recovery. In Praise of Slowness, Carl Honore, Faster than the Speed of Light, Joao Maguiejo, Hilfiker D, Healing the Wounds: A Physician Looks at his Work.




Zen and the Art of Motorcycle Maintenance: An Inquiry into Values.

A Feeling For the Organism, The Empathic Practitioner, When Words Are Abandoned: Feeling the Forest.

Zen in the Art of Archery, Zen in the Art of Flower Arrangement

Pilgrim At Tinker Creek

"Seeing and Saying" - Sobel and Elata

Achilles in Vietnam, Odysseus in America, Marine Sniper

"On Seeing Deer", from The Mountains

The Preservation of Youth - Maimonides

On the Internet - Hubert Dreyfus

The U curve of Perception - Stuck in the Middle

Involvement and Mattering


Variability in the Workload/Time Relationship - a Clue to the Use of Speed ?

Increased Workloads Have Limits, Produce Error, Produce Stress


Radiologist Performance Frequently Brackets the 80/20 Benchmark

Various Practice and Interpretive Patterns Suggest an 80/20 Approach

The U curve of learning, 80/20 radiology, and Knowing Grass in Detail

80 / 20 and Human / Institutional Limitations


Using a Fixed Time per Case Privileges Speed

Not Extending Effort to Match the Challenge of the Image Builds Speed

Normative Practice Flaws can Build Speed

We do not Have an Unlimited Amount of Time


Orderly Looking and Deliberateness Aid Perception

Orderly Search Is Not Necessarily the Rule

Even Orderly Search Has It's Own Limitations - Perceptive Looking is Essential

Perception Requires Experience, Knowledge of Target and Background, and Time, to Disembed Lesion from Background, Using Visual Handles

Proper Attitudes Are Themselves Perceptual Resources

Humility, Beginner's Mind, Patience, Attention, and Time Are Required for Perceptive Looking

Attention is a Perceptual Resource

Adequate Time is Necessary, but Time Itself is Not Sufficient

Giving More Time to the Task of Perception Creates a Moment to Use Correct Attitudes and Methods

Just Sitting, and Searching, is Difficult



Biases For Normal, Bias for Speed

Dreams of Flash Diagnoses Promote Speed, and Promote 'Normal' Diagnoses

False Positive Fears Feed a Bias for Normal, and for Speed

False Positive Errors are Harder to Defend than are False Negatives.

But False Positives Aren't Always that Bad

Having Mostly Normal Patients Encourages a Bias for Normal, and for Speed

Prospective Population, Retrospective Performance, and Accuracy

Speed vs. Time On Task

Subspecialization Uses Time/Experience Accumulated in the Past to Minimize Time Spent in the Present and Future.

Double Readings, Time On Task, and Speed

Computer Assisted Diagnsos (CAD), and Speed

CT - A Speed Trap?

The Missing Body Part - Speed Trap II?


The Implicit Audience Effect may be Underemphasized in Medical Imaging



Discussing Errors is Depressing

Discussing Errors Feels Hazardous

Discussing Errors Meets Impediments

Accepting Error, Not Accepting Errors


Latent Errors, and the Slow Rot of Speed

Redemptive Redundancies, Outcomes Studies, Fig Leaves


Groupthink in Radiology








While reading Raymond Bradbury's Fahrenheit 451 (sk1), the RSNA News August 2002 (sk2) came, with Dr. Leonard Berlin's article discussing perceptual errors in radiology. Such errors are not surprising. The viewing conditions for watching Lord of the Rings are usually better in terms of ambient distractions, stray light, and noise levels than for much of the work done in radiology (sk64p255). Speakers at RSNA 2002 remarked that moving the radiologist to the other side of the planet generally resulted in better viewing conditions, because no one would distract them there.

Then Radiographics (sk3) came with a letter by Drs. Ania and Asenjo regarding a general societal and physician unwillingness to value time spent studying problems in depth, reminiscent of Milan Kundera's Slowness (sk4), which makes further comment on what gets lost by technology, and its offspring, speed. Perhaps Bradbury's Fahrenheit 451 (sk1p9) offers a clue to the long term persistence of perceptual error in radiology.

In Fahrenheit 451, Bradbury describes the expansion of billboards to 200 feet from a preceding 20 foot length, an enlargement needed to make sure that the drivers of that society, going 200mph, would still be able to read the billboard ads. Drivers moving at the extra-high speeds that were normative in the story would be sure to miss the details of what they were looking at, unless the signs were enlarged - seeing flowers and grass only as a colored blur, and never seeing or understanding what they were looking at in detail - losing that understanding because of speed. This slow grass analogy of Bradbury's is similar to the "What" system described by Livingstone (sk146p158,159 - see also p46-52,64,66,77,104,110), a slow detailed system that advertisers sometimes manipulate to force attention and comprehension of their ads. Bradbury's words are more eloquent than my description here, and Livingstone's images more intriguing. The reader might enjoy reviewing Fahrenheit 451 and Vision and Art: the Biology of Seeing. The slow grass versus fast blur examples of Bradbury's are reminiscent of other "big picture" (sk130p16) approaches that misconstrue neglect of detail as a virtue of efficiency required to maintain a presumed lofty "big picture" awareness - or perceptual failure.

Lately, upon reviewing studies from elsewhere, I frequently find missed abnormalities, or other errors, to the point that not finding errors of potential clinical significance seems less common than finding them.

Cases like:

-calcified gallstones unreported at abdominal/chest CT

-partly calcified anterior communicating artery aneurysm, unreported

-retroperitoneal hemorrhage, unreported

-paraaortic adenopathy, unreported

-ommitted erect spot films of the colonic flexures

-a nuclear thyroid scan recommending thyroid ultrasound, ignoring the ultrasound already performed, reported on, and in the film jacket, from the previous week, one or two studies ago.

-high probability lung scans that aren't high probability, but that do distract attention from the for-real mediastinal abscess

-normal and unchanged conjoined nerve roots misreported as new disc herniations - though reported accurately on multiple previous studies then ignored by both the referring doctor and radiologist in the new report.

-radiology residents told they don't have time to read their dictated reports prior to signing them off - this despite the standard that final reports should be proof read (sk89p942).

-large opaque renal calculi missed at nephrographic phase abdominal CT scans.

- and other examples either too numerous or sensitive to provide detail.

Such examples are not from Podunk General, but from moderate to large urban hospitals, some well known. 20 years ago, such things seemed less common.

Maybe I know people who are just unlucky, or some who may be unfortunate enough to live in a 'lunch bucket town' - as one doctor's wife put it - rather than a posh suburb, but I think the problem is more or less endemic - reading films at a rate too rapid to allow proper attention (sk94). I'm not the first one to notice these things - one-third of physicians report errors in their own or a family member's care (sk87). More recent accounts in radiology find (sk61) that premature truncation of search produces satisfaction of search errors more than previously held to be the case. Organizational pressures to cut corners have been observed in other medical disciplines (sk217,sk31p191). Not taking enough time with patients has been listed among the top four causes of medical errors (sk87), along with overwork and understaffing - themselves conditions likely to privilege the use of speed, and to diminish speed's denominator - time.

Others have noticed that radiologists can somehow manage to miss the obviously abnormal (sk48,sk49,sk100,sk8, sk58,sk65,sk214,sk19,sk106,sk107) with multiple experienced and motivated readers missing cancers visible in retrospect (sk69), with missed abnormalities sometimes not even small (sk37p119), and with misses sometimes having the odor of negligence (sk40p149). Some authors observe that radiology has not given much overt attention to one of its most important resources - time (sk14), and maintain that adequate time is important for accuracy (sk49p80).

Such implausible errors probably come, at least in part, from a mental set to read the film in a highly 'productive' fashion - adjusting the rate of interpretation to enable reading more and more cases in a fixed, or diminished, amount of time - producing the search time truncation effect - where abnormalities described are those seen in the first 20 seconds - Procrustean radiology. Anything after 30 seconds is a zebra - by definition - since it is rarely observed (sk220). This truncation tendency may be self perpetuating. Reports suggest that failing to detect abnormalities increases the future likelihood of not seeing them, (sk103p367) producing, and reproducing, the 'false zebra' phenomenon. Once inadequate search habits become automatic (sk51,sk52,sk223) it may be harder to improve them.

Perceptual error is particularly damaging because failing at the primary task of detection precludes further diagnostic accomplishment (sk108p251). An irreverent teacher, trying to emphasize the primacy of perception, commented - "There are guys who could talk forever about a finding if somebody would only show them the abnormality ". Wondering how radiologists could miss obvious findings, even after years of training, a patient's wife observed "that's how they know the words".

Many studies have demonstrated a bias for false negatives, and for normal. Missed diagnosis accounts for 70% of lawsuits against radiologists (sk24,sk80). Abnormalities that are overlooked can be flagrant (sk88p278), with errors enigmatic (sk80). At least half the errors made in clinical practice are perceptual (sk76), with that figure as high as 75% (sk2).

Multiple reports, over decades, often retrospective studies of cases with known abnormalities, identify perceptual misses occurring from 20% (sk27) to up to 60% of cases (sk24), with 30% a general estimate of the failure to perceive abnormalities present (sk27,sk16,sk16,sk32,sk18, Sk137,sk47).

Under readings surpass over reading by factors ranging from 3-10 times (sk233, sk58,sk16,sk37,sk233,sk32,sk40, sk58,sk63,sk20). Others have found that radiologists quit searching prematurely (sk88,p287,sk49p85), even while detecting more true positives than false positives (sk23,sk14), quitting with some of the work of distinguishing between normal and subtle abnormal remaining undone.

The 'satisfaction of search effect' (sk61) may in part be due to early cessation of search - time truncation - ignoring the possibility of additional abnormalities. One of the favorite measures of perceptual performance, the ROC curve, emphasizes accomplishment - the area under the curve of perceptual work performed - rather than the area above the curve, perceptual work left undone, where one would think attention might be focused for further improvement (sk233).


The problems of looking, and not seeing, are not unique to medicine (sk79). In the non medical world, speed has a known negative effect on perception, like the small print effect in contract analysis (sk79) - where a correct understanding occurs only if you look slowly - demonstrating the obvious difference between serious reading and the skimming pattern more appropriate for the reading of pulp fiction.

Other fields share our problems with visual tasks. Roofers use brooms to remove the structural noise of twigs, and other debris - to better see the small defects that are clues to leaks (sk199). Puzzle enthusiasts locate a specific piece based on its shape, color, border sharpness, borders interface, and structures displayed.

The military analyzes aerial photographs to find its own version of cancer (sk18,sk65). Their camouflage patterns bear a similarity to the diseased lung that hides early tumors, and military work in radar signal detection has contributed the ROC curve to medical image analysis (sk119).

In search and rescue work on the open ocean, searchers doing the visual scanning to find people in the water need 20 minutes of rest every hour, because the work of perception is so demanding (sk260) . While the task of searching a broken, grey, and white foaming ocean for a small human dot would seem to demand the same perceptual effort required when searching for microcalcifications in the grey/white visual tumult of a mammogram, unlike search and rescue, there are no mandatory rest break standards in medical imaging. Perhaps taking longer rest breaks in medical imaging would have the effect of setting off and emphasizing the importance of those moments that are spent in the search and rescue work of medical imaging, somewhat like the aesthetic effect of negative space in visual art.

Microscopists depend heavily on vision, as do archers, snipers, air rescue teams, and deer hunters. Naturalists use honed visual perception, and describe various forms of camouflage that occur in nature (sk38,sk195).

Psychological factors such as a bias for benign diagnoses, low motivation, boredom, alertness, medication, fatigue, and unwarranted confidence, may influence visual perception and perceptual accuracy (sk79).

Because perception is central to the radiologic task, anything, like speed, that interferes with perception, especially threatens radiologic work. Looking at perception and speed in non-medical literature offers psychological distance, a little consolation, and allows disengaging from the usual and customary habits. Non-medical works like those discussed below emphasize that perception is enhanced by thoughtfully spending more, rather than less, time on task, and that issues of speed, noticing, and attention are not unique to busy doctors, but are larger human issues, like quality, and caring, in general.

Zen and the Art of Motorcycle Maintenance: An Inquiry into Values

Robert Pirsig, in Zen and the Art of Motorcycle Maintenance (ZAMM), discusses the problem of perception as a major element in constructing Quality (sk7p246-288). In ZAMM, Pirsig (sk7pp10-25,44-72,148-150,247-269,281,284-288) describes two modes of engaging the world, a 'groovy' mode that is fast, easy, and feels good, and another more tedious, technical/analytic way of approaching things, necessary for difficult problems - modes that roughly correspond to the fast and slow phases of radiologic interpretation described by Christensen (sk14), the automatic vs controlled processing of Shiffrin and Schneider (sk51,sk52) in detection, search, and attention phenomena, the "synthetic" versus the "analytic" approaches described by Smith (sk49p96), the"global" versus "discovery scanning" described by Samuel/Kundel/Nodine (sk47,sk103), "survey sampling" versus "examination sampling", the 'declarative' versus the 'compiled' knowledge of Lesgold (sk223), the "Where" versus the "What"

(sk146) systems of Livingstone, and the "glancing" versus scanning (sk106p1245) modes.

The first groovy/fast/automatic mode probably has a greater attraction to managers, delivering the benefits of a ride on the machine without needing to know too much about it, and depending on others to make it run, whereas the technical/slow/controlled mode is more in the realm of the physician who actually does, and enjoys, the work. The groovy approach, focusing on the easy and obvious, fails at the point where something new, difficult, or different is encountered, and where better method is needed. The technical, classical approach that Pirsig describes, the willingness to do the tedious, unattractive parts, requires not only the master's careful habits, but the humility of beginner's mind, as if the film might have something to teach. The care and feeling for the mechanical work that Pirsig describes is similar in tone to the microsopist's work described in Evelyn Fox Keller's A Feeling for the Organism, and similar to the Dreyfus model of competence and mastery (sk6,sk6c,sk6b). Ultimately, reluctance to do the slow work of observation and analysis subverts having the skill necessary for a technology that really works.

A Feeling For the Organism, The Empathic Practitioner, When Words Are Abandoned:

Feeling the Forest.

A Feeling for the Organism (sk13), by Ellen Fox Keller, describes Barbara McClintock's keen observation, and prescient work with corn genetics. As a microscopist, depending on qualitative and careful observation, she is a kinswoman to radiologists (sk13p148-151). Her observations on the importance of perception (sk13pp198,xxi), the need to take the time to look carefully, the value of doing routine tasks, (sk13pp69,80,97,101,102,103,115-119,147-151,180,206-207) and the importance of motivation (sk13p200) are particularly pertinent to radiologists (sk44). As with Pirsig (ZAMM,sk7p281), she describes the importance of observing without imagining ahead of time that you already know what is there to be seen (sk13-p179,sk79), a temptation that radiologists face when reading to find abnormalities in populations that are statistically most likely to be normal - the downside of a limited perceptual set (sk79,sk63,sk49,sk119). McClintock observes how it is not always easy, but important, to resist the seductions and sweet nothings of speed (sk13pp180-181,198,200,206), and like Kundera, she describes, how this is getting more difficult (sk13p206) in a technological society.

Other social trends seem to confirm McLintock (sk249, sk251). Radio commercials use speed narrators to drape defensive legal lingo over commercials, talking fast enough to say they said something, but too fast for anyone to really understand it all. Fast food continues to make tradeoffs of time for quality (sk249p149). In the 1970s, when oil shortages and embargoes skyrocketed fuel pump prices, the country was able to slow down to 55mph, but now with even higher gas prices, the will to slow down seems lacking. 20-30 years ago, before the era of computerized spell checkers, paradoxically, spelling errors seemed less frequent at RSNA talks than they do now.

This feeling for the organism that Keller describes means a deep familiarity with the object of study in such detail that a kind of empathy develops, like the oneness between subject and object that some of the Zen literature discusses. This sympathy or empathy is like the deep technical knowledge of Pirsig's analytical mechanic, rather than the fast and superficial approach of those who only want to ride the machine. Pirsig's technical approach might seem antithetical to empathy, but being embedded in the experience of details and depth, in fact has much in common with empathy. Pirsig's description of the alienation of workers from their own work can similarly be applied to alienation of physicians from patients.

In medicine, the power of this self/other boundary permeability is invoked by the canny patient who asks a physician what would be his/her own choice if making the same decisions for the physician's own family, and also employed by the canny researchers who asked anesthesiology program directors if they themselves would feel comfortable undergoing anesthesia under the direction of their own graduating residents - with rather disheartening results (sk73). Somewhat similar trends have been obtained in radiology (sk71).

ER docs in the habit of interpreting head CTs on patients they've seen clinically - the habit of involvement - perform better than those not in the habit of reading head CTs for hemorrhage (sk225).

The multiple authors in Empathic Practitioner (sk31pp3,5,7,9,12,78,158,163,173, 189-202) similarly emphasize how subject / object separation can lead to poor perception and care, confirming Dreyfus (below), who also found that detachment detracts from care. Detachment and ingroup/outgroup effects have resulted in empathic blocks and ethical failures in well publicized and less publicized studies (sk246,sk235).

Jane Goodall indicates that empathy with the subject aids the collection of good data at the most basic level (sk195-pp54,59). In "When Words Are Abandoned: Feeling the Forest", excerpted in the quarterly Parabola, with an attached interview, detachment and empathic failure acts to support the service of economic agendas other than the well being of the subjects of scientific work. Empathic failure and detachment in the service of economics might similarly occur in highly manipulated medical reimbursement systems. Goodall's description of the separation of subject and object in the study of animals, together with the role of self-interest (sk195p59), has a flavor similar to that of the ingroup/outgroup dynamics found in various prison and medical situations (sk242,sk246,sk247,sk235), where the interests of animals, prisoners, or patients go invisible and unvoiced.

Zen in the Art of Archery, Zen in the Art of Flower Arrangement

Eugen Herrigel's work, Zen in the Art of Archery, (sk15) emphasizes the value of performing archery in a non-hurried, unattached way, paying close attention to each component of the act, like the developing Master level described by Hubert Dreyfus (sk6,sk6c). Focused attention to important details, and lack of attachment to the ending, or to distractions, were all important in consistently shooting well. Non attachment might prevent radiological, speed-related, "satisfaction of search"/time truncation errors, tunnel vision, and all the other mental traps of radiologic perception that hurry us on to hearing the thunk in the target - the "end, next case" - as the goal, rather than having discovery and practice as the goal. Loss of boundary between self and activity probably prevents the psychological distance between self and other that enables time truncation methods. This same loss of self/activity boundary, remarked on by ZAMM, and by McClintock/Keller/Goodall, helps make the Master's efforts appear effortless, like a natural reflex, operating without conscious control.

The patience needed to become a master archer is probably required for anyone to become a Master Anything. Herrigel's wife Gustie, in her book Zen in the Art of Flower Arrangement (sk15b) observes that the master arranger first lays out the basic tools and materials for the task of proper flower arrangement (sk15bpp18-19,28-30) comments reminiscent of Dr. Beverly Wood's writing in Radiology (sk44) who finds that experts first lay out the basic problem elements and observations before searching for diagnoses, an observation also shared by others (sk63,sk79,sk31pp173-176).

Goodall, as a naturalist, comes close to the imaging interpretive situation when she advocates letting the words come later, after the observational work is done (sk195p55) - suspending verbal expression and labeling in order to attain a heightened receptivity and a more complete data set, prior to putting forth answers. This is very similar to what happens in film interpretation-those who want to quickly put words on the page may not give enough time to developing a basic representation of the problem, may not give time to perception and the receptive silence that precedes the useful production of words. It is probably better to be perceptive first, then productive.

Lesgold (sk223pp36-37,51) makes the same point, observing that experts have a deeper fund of detailed supporting evidence (knowing the details of grass) than others, and recommends a two step process in problem solving - first the representation of the problem - where the elements of it are laid out in detail, and then the solution. Performance pressure and anxiety about pumping out the work, may encourage too perfunctory an observation and analysis prior to dictating a short, terse report - a form of linguistic premature ejaculation.

In Empathic Practitioner, (sk31pp173-176), Connelly similarly describes the special kind of attention that fosters perception, quite similar to the recommendation of others to first obtain a basic and complete representation of what you are looking at.

Pilgrim At Tinker Creek

Annie Dillard in her Chapter "Seeing", from Pilgrim at Tinker Creek (sk11) , discusses how looking without observational discipline results in not seeing the plainly visible. She repeatedly overlooks whole flocks of birds, and discusses the subtle clues to the location of caterpillars, and the meals of mice - not unlike subtle nodules, infiltrates, and hilar enlargements. Shunning slam bam observation, she describes the advantage that the 'lover', the 'knowledgeable', and the 'specialist' have in seeing - being willing to give more time and attention - and she stresses the need for both caring and competence in effective observation. (sk11pp14-19). Similar trends are found in radiology, with experts and subspecialists generally performing better than generalists and non-experts for a variety of tasks including abdominal CT, chest CT, head CT, magnetic resonance imaging, and chest radiographs (sk20,sk105,sk76,sk89,sk223,sk224,sk225,sk240,sk248). This is to be expected, since those with a large amount of specific knowledge in a given field will have better recall of pertinent details, and will generally process information more effectively (sk223pp33,34). Such detailed knowledge is similar to knowing grass in detail, rather than as a blur, as in Fahrenheit 451.

"Seeing and Saying" - Sobel and Elata

Like Feynman's book, The Pleasure of Finding Things Out (sk29) describes for science, Sobel and Elata (sk22) regard poetry and medicine as all about seeing, discovering, and saying. Poetry, concerned as it is with going from imagery to the written word, and then back to imagery, bears some relation to the process of medical image interpretation, reporting, and report reading (see also sk31,pp205-212 regarding reading and medicine). Sobel and Elata, in their article, "Seeing and Saying", like Feynman (sk29), Pirsig, Goodall (sk195), or Keller's life of Barbara McClintock, argue for involvement and non-anonymity, discuss the joys of discovery, and how the work of seeing and discovery is demanding and requires attention, and care - the antithesis of neglect. They recommend poetry as an exercise in slowing down to perceive, observing that poetry reading (sk22p97) can restrain the usual and customary haste to get to the bottom line, and in the rush missing meanings, and findings unfound - essentially time truncation. Reading poetry is about making time, or giving time - and is about the search for meaning - not too dissimilar from the search for meaning implicated in satisfaction of search errors. Image interpretation, like poetry, is ineffective when just saying or knowing the words and rhyme passes for, or takes precedence over, knowing/ grasping/ having a feeling for the imagery. Poetry, like imaging, can supply multiple nuanced meanings from one shadow/word element of its lexicon - depending significantly, like medical imaging, on current context, and past history.

While a variety of formal verbal and written assurances of quality care and quality orientation are suggested as pre-emptive aids to the legal defense of malpractice cases (sk211), one of the basic tenets of quality literature and poetry may have important application for health care practice - - - - Don't tell them --- show them (sk148p122-124).

Achilles in Vietnam, Odysseus in America, Marine Sniper

Like the military (sk226,sk106,sk65), radiology is also concerned with targets, hits and misses. Lives matter, and not surprisingly, similar problems of dangerous and fast, versus safe and slow, have been described in Jonathan Shay's book, Achilles in Vietnam, (sk9pp14-17), discussing how doing the job can get betrayed by speed and institutional priorities (sk9, sk130) of appearing to have done the job. Shay gives similar episodes in Odysseus in America (sk130esp166-167), describing the glib errors of "big picture" mentalities (sk130p16), the tendency of technology to be overvalued compared to skill (sk130p217,p223), and the habitual tendency of ethical discussions to ignore the diffusely relevant ethical aspects of routine competence (sk9p14-19,sk130p224), while being distracted by, focusing on, and emphasizing, the esoteric and unusual.

Another military text, Marine Sniper (sk10pp209-219,252-254,273-275) describes a famous American sniper, Carlos Hathcock, able to use honed perceptive powers to save others from ambushes and booby traps, and to use measured slowness to sneak into a North Vietnamese camp to kill a key general. When teamed with regular troops using speedier, less careful methods, however, he is nearly killed, suffering severe injuries (sk10).

"On Seeing Deer", from The Mountains

The chapter "On Seeing Deer", from The Mountains, (sk12), by Stewart Edward White, will feel familiar to radiologists. Writing in 1904, at the dawn of the radiological era, White is amazingly predictive of the perceptual issues and observations that radiologist would deal with over the century to follow, and describes how experienced woodsmen find deer.

The deer hunter's "skillful elimination of the obvious" (sk12p121) precedes Gale et al (sk214,sk226p243 ) who use the term "anticipatory schema" to describe how radiologists rapidly disregard normal structures in order to focus on those that are likely to be abnormal.

Managing "to see everything" (sk12p120), "constantly searching the prospect with keen understanding eyes" (sk12p122), all take time and don't happen with a grand sweep of the eye, as Christensen later observed (sk14). The ability of deer to hide in plain sight is like some abnormalities that radiologists miss, and lawyers find, the ones that "you can look straight at them and not see them at all" (sk12p123) - in both deer hunting and radiology often due to overlying obscuring structures (sk37) or distractions, and very similar to the effects of temporal lobe lesions on the "What" visual system described by Livingstone (sk146p64)

White's detailed description of how to look for deer as a function of their age and sex will be familiar to the radiologist (sk12p125) - playing the odds of demographics. The radiologist's observation that perception is improved by knowing where to look for what (sk21,sk44,sk107,sk88p287) is affirmed by the woodsman's experience. The deer hunter's advice - look closely at regions of brown - is an example of the observational discipline of "visual isolation" (sk49p79), or disembedding, that expert radiologists apply in searching films - this visual isolation using selective, conspicuous, characteristic features as a mental 'handle' to focus productively on clues to quarry.

White's comment "When the number is filled out he sees nothing more" (sk12p122) precedes radiology's 'satisfaction of search' concept by nearly a half century.

The less apt performance of the tenderfoot, who pays attention only to the " striking features" (sk12p122) , is similar to the way in which the most visually impressive or unusual parts of images become figure (sk88p285, sk47), and is contrasted with the quiet attention of the experienced woodsman who looks carefully. White finds (sk12p126) that "Perversely enough, the times that you did not see deer are more apt to remain vivid in your memory than the times when you did".

I can believe this, recalling how, ill as a tenderfoot intern on a medicine service, I wet read my own 'normal' chest film. This was later followed by a call from a radiology attending.

"Do you know you have infiltrates on your chest film?"

"No - - - really? . . . which side ?"

"Well .. . . . there are more than one... . . actually."

Revealing a lack of perceptive discipline, experience - and a lack of previous films - this was probably remediable, given illness, and being more immortal then. Cases like this demonstrate how it is easy to see the ways in which unconscious factors lead to error (sk131), but easy to see only in retrospect.

Like deer, abnormalities in medical imaging may not announce themselves, and their detection requires an active, informed search, rather than passively responding to the obvious (sk94). Like the foreleg of a deer in a thicket, we may only see part of a lesion, and have to be able to synthesize the mental image of the whole deer, or tumor, from the visual scraps and clues that analysis, or memory, provide (sk49).

Deer hunting, also concerned with targets, hits, and misses, is probably easier than radiology, because radiologists hunt many unknown targets simultaneously, which requires setting, and resetting, the mind's view of what is figure, and what is ground (sk14,sk222pp105,sk64p254), for each anatomic region - all of which necessarily takes time. Giving time means avoiding the rapid fire misses that are the radiological equivalent of the trigger jerking deer hunter's 'buck fever'.

The Preservation of Youth - Maimonides

Maimonides shows that all this is an old lesson which must be learned and relearned, writing:

"To understand all this, as it is described in books, is very easy .... But its practical application...., in a case involving a particular patient, is quite difficult for the wise man .... But laymen who are ignorant of the foundations of this science, yet talk with certainty and without restraint, consider the problem an easy one and do not believe there is an illness that requires much observation." (sk42bp78)

".... while Hippocrates talks about its difficulty, crafty people consider this profession light and easy."

"The more perfect a person becomes in one of the sciences, the more cautious he grows, developing doubts, questions, and problems .... the more deficient one is in science, the easier it will be for him to understand every difficulty, making the improbable probable ...and increasing the false claims which he represents as certain knowledge, and is eager to explain things that he does not understand himself." (Which seems very reminiscent of Feynman's comments.)

The task of visual perception, often performed unconsciously, becomes by this unconsciousness more difficult to understand and control. Maimonides quotes Galen - " It is easy to say that it is proper to anoint old people with oil and rub it in, but to do it properly is one of the most difficult tasks."

Laymen, the rapid radiologist, and the uninvolved observer can oversimplify the difficulty of professional tasks. I used to think diving was difficult until I saw all those people at the Olympics doing it. This distorted laymen's view of requisite task rigor might be partly cured, for the field of radiology, by demonstrations of optical puzzles, like Brooks triangles, and other illusions or demonstrations published in radiology or other vision dependent fields (sk2,sk58,sk57,sk65, sk103,sk106,sk107,sk212,sk254).

Jurors might be impressed with the difficulty of perception, if given the opportunity to struggle with these things themselves, producing the 'where's Waldo' defense (sk39,sk80) strategy. Some visual puzzles, like the two faces vs. a goblet figure (sk107), demonstrate how difficult it is to keep in mind two visual possibilities at the same time (sk200,sk254). Costochondral calcification - or subtle lung mass? Goblet - or two profiles (sk107)?

On the Internet - Hubert Dreyfus

Hubert Dreyfus' apprenticeship model for skill acquisition, like medical learning, defines stages through novice, advanced beginner, competence, proficiency, expertise, and mastery, requiring intense personal contact with the material to be learned (sk6c-pp35-48;,sk6). Both radiologists (sk71), and Dreyfus recommend guided learning or mentoring to improve performance. The master physician, in the Dreyfus model (sk6,sk6b,sk6c), develops by intensely engaging with the work at hand, and by retaining mental flexibility (sk44) - the beginner's openness to the possibility of error - as a master chest radiologist once said - "Well, if I've seen it before, this is the first time I've recognized it. (sk205)"

The skills of a discerning radiologist are not acquired in 4-6 years of post graduate training, but with attention over time, as Nodine, Kundel,and others (sk71,sk70,sk1,sk51,sk52,sk44) have observed, describing how experts bother to routinely look at the problem areas (sk35, sk270).

Both speed and accuracy are known characteristics of experts (sk39,sk6c-p42,sk6-p15,sk6b), like chess masters (sk44,sk6c-p42), with speed added to accuracy over time, based on accumulated perceptual experience (sk71). Experts have more accurately honed visual perception (sk71), probably somewhat like knowing the difference between a green blur, and grass (sk1).

The payoff of making the time to develop good visual search habits comes when speed is added to a foundation of accuracy. Similar to the Dreyfus model (sk6,sk6c), to the groovy vs. technical methods of Pirsig, or to Herrigel's archer, the speedy 'automatic processing' described by Schneider and Shiffrin (sk51,sk52) develops once the 'controlled processing' of radiologic perception - taking time to know that grass is more than a green haze (sk1) - is carefully repeated enough times to become automatic. Like Herrigel's archer - " it shoots" - (sk15pp52,51,59,61,76), the expert performer first achieves unself conscious action, and then the speed that comes with unself-conscious action.

If accuracy is sacrificed to speed, before the tasks of accuracy and controlled processing (sk51,sk52) are learned, discernment may also be sacrificed, since learning discernment takes time (sk6c; sk6p16). Speed precludes taking in details and complexity - precludes knowing grass as more than a green blur. Perhaps extra training makes time for, and fosters the attitude that one is still a learner, allowing time enough for learning the details of perception , more so than will be the case later on, when as a practicing radiologist, demands for speed compete with time for perception. Like a woodsman sharpening his axe - attending to the small amounts of metal that make the difference between a razor sharp tool and a bludgeon - paying attention means making time to notice small differences, and retain efficient performance, rather than build the habits and skills of rapid but mediocre performance.

The U curve of Perception - Stuck in the Middle

A U curve of perceptual performance as a function of experience has been described in radiology (sk223), perhaps reflecting the beginner's more alert involvement and attention in detail to new material - paying attention to grass in detail, and reflecting the expert's more facile familiarity with new material, each resulting in enhanced performance, with the middle levels of developed skill being familiar enough with the task to be bored, but not familiar enough to be excellent.

In Zen in the Art of Archery (sk15pp71-78), Herrigel describes how a student swordsman loses a bit of his edge as he acquires some of the knowledge of swordsmanship, and only later reclaims some of his original selfless attention and skill as he progresses, reminiscent of the U curve Lesgold describes (sk223) . Some studies seem to confirm just such a signficant beginner's mind effect. In one study (Sk245), a resident - R2, had true positive and false positive scores superior to staff radiologists, by giving more time to make more comparison scans, and performing more second looks in difficult cases. The other observers did not increase their visual efforts, even when having diminished confidence. This doing what needs to be done, even if unsupervised, as if it mattered, is a variation on the military concept of the strategic corporal (sk130p223) - fully aware of the needs of the mission, and acting to accomplish the mission, if necessary even without the supervision and guidance of officers.

Some studies find that radiologic technologists may screen chest radiographs with fewer ommissions than staff (sk233), or that a resident with one year of training (sk233) may make the fewest number of significant errors, even compared to staff radiologists. Interpretive performance of reading head CTs for hemorrhage by neurologists and radiologists was mildly, but inversely proportional to years in practice (Sk225), perhaps because the older cohort lacked the benefit of learning these skills from experts during their own residencies (learning to see grass in detail), or perhaps they lacked the openness to learning and paying attention that is characteristic of a beginner's mind. At the other end of the U-curve, advanced track radiologists (Sk225), with a higher score on detecting hemorrhage, were consistently better than those in a standard track, perhaps operating on the second, upward, limb of the U-curve - operating as experts.

Variable correlation of radiologic performance with increasing experience (sk233) is consistent with the U curve phenomenon, the nadir of performance occurring when one is experienced enough to be bored, but not experienced enough to have the unconscious perceptual edge of the specialist. Herman and Hessel (Sk233p62,67) explain these results by suggesting that, given fundamental training, an individual's personal character and attitude may more strongly affect perceptual accuracy than length of formal training or experience, echoing Shay's (sk130p167) observations in Odysseus in America. This may explain Bosk's (sk5) observations that normative errors, those from failure to bother to apply knowledge, can be more important, persistent, and impervious to the effects of training, than are technical errors, those due to a deficiency of knowledge.

Involvement and Mattering

The U-curve of performance versus experience (sk223), and an attitude of involvement, may explain much of the precocious high performance of anomalies like Reader B and R2 above (sk233,sk245). Dreyfus also observes how emotional involvement (sk6c-p37) is a key factor in the development of excellence, (sk6c-p42 ) with experts willing to spend time on task as if the task mattered - doing it like they meant it.

Involvement and mattering (sk79p413), are important in the development of mastery (sk6c). In the radiologic literature, Smith (sk4p96) and Garland (sk63) both observe how the approach to life in general, and philosophy of the interpreter, are important factors in improving film interpretation. Like Pirsig (sk7) and like Keller (sk13), Dreyfus finds that development beyond competence requires emotional involvement with the goal of the task, a non-anonymity, so that the report is more than a result, and says as much about the author, as about the patient. With minimal emotional engagement, there are fewer of the questions that make film viewing interesting enough to give time to, and fewer questions that make searching productive. Thomas (sk79p406) relates inadequacy of image search completeness and time spent in perception, commenting on how reading style reflects factors such as "interest, involvement, and self deception".

If lack of positive feedback directly from patients is a deprivation for radiologists (sk89p943), the anonymity structured into the radiologic task may be a risk factor for non-involvement, and thus a risk factor for error (sk89,sk92).

Attempting to develop beyond competence, in a context of isolation, and with disembodied involvement, as described by Dreyfus in his description of the distance learner (sk6c) - is disconcertingly similar to a detached radiologist working in the distributed information reality of PACS. Split off virtual reality, rather than personally felt reality, may be a risk factor for the detachment that ultimately leads to perceptual and ethical lapses, a point made by Sobel Elata (sk22) and Goodall (sk195), Split off virtual reality may be incompatible with the attitude and emotional valence that are essential to vigilance (sk47) and perception.

Emotional involvement, necessary for development beyond competence, is probably also incompatible with Procrustean, time truncation technique. Dreyfus' relational view not only relates scientific theory to actual practice, but is about the relationship of doctor to patient. Not surprisingly, lawsuits often derive from absence of this felt relationship, making uninvolved anonymity both a predisposition to error, and an added risk factor for lawsuit when errors occur.


Radiologic workload has increased (sk47a), potentially forcing rapid image interpretation, time truncation methods, and errors.

Various sources, over the time period of 1986 to 1996 have placed radiologic procedures per FTE at between 10,000 to 13000 per year (sk115,sk117,sk96). Some sources (sk117) claim that per physician productivity was not changed much from 1986 to 1989, but in general others have found opposite trends.

Workload in procedures per FTE, and in RVUs per FTE is reported also to have increased over the period 1996-1997 (sk43p821). Others have found that (sk139) workload difficulty has increased by 18% over a longer period from 1973 to 1989, and with an 18% increase in RVUs from 1995 to 1998 (sk47a). Others find increased RVUs accounts for an increased workload from 1991-1996, even if procedures hadn't increased (sk115), with radiologic manpower falling behind workload over the past 20 years (sk93).

While stated workload varies according to the type of imaging done (sk116), published workload figures may underestimate the workload. Headhunters have informally stated that 16K, or 20-22K, per year per FTE, is the usual workload, with extremes ranging as high as 60K cases per year (sk218). Thus, the comparatively lower workloads published in radiology journals may be falsely reassuring. The ommission of outliers may render invisible those practices reading at three times the median rate (sk47a-p1124).

Case numbers underestimate the real increase in workload, since more cross sectional imaging produces more images per case (sk139). Thin slices used to enhance accuracy also increase the amount of visual work (sk132). With new technology, and image interpretation time often exceeding acquisition time (sk119), the human factor - the radiologist - becomes the rate limiting step to rapid throughput and fiscal accomplishment. Technical advances worsen, or at least accentuate and emphasize the failures of human perception to match the capabilities of ever more powerful equipment, as both Berlin (sk239p270), and others have observed (sk240,p154; sk248,sk65,sk230,sk263).

Increased workload can adversely affect perception (sk49p97). Heavier workloads not only encourage the use of speed, but (sk49p97 ) the fatigue of a heavier workload itself might blunt perceptive skills, even if appropriate time is used per case. With a greater workload, the distraction of doing four things at once may impair careful evaluation (sk77). Negative effects of workload and speed have been described for other physicians as well (sk274p45).

Latent errors, such as newly implemented technical advances that make work more cumbersome, can introduce inefficiencies that force the radiologist to squeeze from somewhere the extra minutes needed to keep up with a higher workload (sk114) . One always available source of such extra minutes, especially for staffing conditions already as spare as possible, is applying time truncation methods to interpretive time - the down side of speed.

Berlin (sk211) describes a malpractice case where allegations regarding excessively high workload and work rate increased the liability of the radiologist, with the new and threatening twist of punitive damages. While in this case the defendant radiologist denied that his accuracy diminished as the day went on, in fact other papers suggest that accuracy decreases as the day goes on, especially after lunch, sometimes with the observers unconscious of this decrement in accuracy (sk213). Interpreting more than 20 abdominal CTs per day is reported to double the error rate (sk89p942,sk77). One paper (sk138) even suggests that fatigue occurring during the course of interpreting a single exam might explain diminished accuracy in interpreting the lower slices on a CT scan.

With cost cutting, and demands for economic production, a long look becomes harder to support (sk28), with a potential for diminished quality (sk92,sk94), and for (sk79) looking without seeing. Such pressures are only likely to get worse (sk246).

The detachment from the objects of their work that Goodall describes (sk195p59) in some scientists, may serve economic agendas generally inimical to the interests of the animals studied. This tendency may generalize to all primates, not just non-human primates, particularly if the interests of the animals, prisoners, or patients are unvoiced and invisible.

Variability in the Workload/time Relationship - a Clue to the Use of Speed ?

Poor correlation (sk43p821) between hours worked and workload, with high variability of RVUs per clinical work hour (sk43), and with high variability in work/FTE among various practices (sk43) implies significant variability in interpretive rates. Other studies confirm a high variability in the workload per FTE, with (sk96) the 25th percentile at 7000 procedures/ FTEyr and the 75th at 14,100 procedures/FTEyr. Even authors investigating workloads seem unable or reluctant to define acceptable or optimal throughput rates (sk116).

High variability in workload does not preclude evaluation of the appropriateness of that variability, but rather invites such evaluation when the causes go unaccounted for. Varying organizational cultures, and acceptance of time truncation methods, may explain part of this inconstant and variable relationship between workload and time worked. High variability in workload per radiologist may simply reflect that some radiologists are unduly optimistic about their ability to read cases accurately at a high rate of speed. Garland (sk63) quotes radiologists who claim to read over 400 "films" per hour. Attempting to keep practice costs low predicts that radiologist staffing will be minimized to the extent feasible (sk275p271), which increases the likelihood that under staffing, rather than over-staffing, will be more likely.

Groups reading large numbers of cases might not be reading too many if their practice employs equipment and personnel that allow the radiologists to maximize the time they spend looking at images, and minimize the time spent handling films or paper. Groups reading a middling number of cases might be reading too many if their work environment is cumbersome and disruptive. While some claim that it is difficult to define or predict a normative workload (sk211,sk45,sk89p942), attempts have been made in this regard (sk89,sk92), and most radiology groups do in fact routinely make such determinations when setting the staffing level of their departments (sk275). Because the work facilitating factors that might account for high variability in workloads across practices (sk211) tend to be of the type that require additional resources, and may therefore be more likely to be lacking than present, a cautious approach would be in general to favor lower numbers of cases per FTE as being more desirable.

Increased Workloads Have Limits, Produce Error, Produce Stress

One half (sk46) to three fourths (sk92) of radiologists report too much work (sk45), and resultant stress. Physicians and radiologists themselves have related (sk87,sk89) errors to understaffing, overwork or fatigue, all implying the possibility of insufficient time on a task. Nearly one-third of radiologists reported that being concerned about the effect of diagnostic errors on patients was a major source of stress (sk93).

Increased (sk82p849) interpersonal conflict among individuals in group practices due to emphasis on relative value units and individual productivity suggests that economic pressures are easily strong enough to disrupt relations between colleagues. The physician patient relationship - already a power imbalance, and in radiology, especially attenuated by minimal physician-patient interaction - is a relationship similarly, and perhaps even more, susceptible to degradation by the use of speed.


The persistent level of radiologic error may be related to the use of the business world's 80/20 rule, also known as the Pareto principle. The 80/20 rule holds that 80% of the benefit comes from the initial 20% of effort, and recommends deferring, in the name of efficiency, the last 20% of possible benefit that would require 80% of a potential total effort involved. To the extent that medicine is financed as most businesses are, it will tend to use proven business methods.

One explanation of the unchanging radiologic error rate is that benefits of new technology are exploited first with an 80/20 eye to previously unmet diagnostic needs and to anticipated economic gains, capturing easy, large diagnostic gains and revenue first, while putting off fine points, or optimizing accuracy.

While some claim that radiology has in the past been exempt from the 80/20 model (sk98), the traditional error rate of 20-30% suggests otherwise. A 20% deferred search is not too far from the 20-30% figure for radiologic error in retrospective studies (sk137). The 80/20 binary, being right on the easy ones, and willing to miss the hard ones, seems similar to Christensen's split rapid phase/slow phase of perception (sk14,sk79), where attention was first given to areas of highest yield (sk14,p364). This recognition that more accurate and detailed work takes longer, with speed and quality of information a trade off, is also explicitly recognized in military photoreconaissance work that subdivides reports rendered into first, second and third phase categories, these designations depending on the promptness of the report, and the quantity of information desired (sk222p91), a procedure probably similar to preliminary versus final reports, but with a more explicit connection of potential inaccuracy to rapidity of reporting.

An 80/20 strategy avoids ludicrous, or clearly objectionable, mistakes, only missing lesions subtle enough to be excusable. Demands to cheapen the cost of providing health care, a corollary to health care as a right, or source of profit, will tend to defer the final 20% of potential diagnostic gains that require disproportionately more resources. Tougher increases in diagnostic acumen beyond a basic level may garner few rewards other than personal satisfaction, particularly if referrals are handled by impersonal contracting institutions.

Radiologist Performance Frequently Brackets the 80/20 Benchmark

Many studies of imaging interpretation show that actual radiologist performance often closely brackets the 80% Pareto benchmark, even for the limited research task of finding one type of abnormality in a limited number of organs. Lower performance is to be expected when the task becomes as complex as real world tasks are.

The performance of radiologists in detecting (sk210) rib fracture was AZ=0.80 when asked to look for rib fracture only, but dropped to AZs = 0.73 when multiple abnormalities were to be considered. (A perfect score would be AZ = 1.00)

For a given test set of chest radiographs with confirmed abnormalities, radiologist performance as measured by ROC curves varied from .79 to .95, according to a ranking schema of expertise, with the category 'all radiologists' producing an ROC curve area of .86 (sk42).

In evaluating head CT scans for the presence of stroke related disease, non specialist radiologists were correct in 83% of instances (sk224, sk225), with only 52% of radiologists 100% sensitive for detecting hemorrhage(sk225). For difficult hemorrhages, radiologists correctly read 80%.

At a past RSNA exhibit (sk8,sk196), with computerized evaluation of radiologist accuracy in detecting nodules on digitized chest radiographs, without CAD, reading chest images for lung nodules, the area under the ROC curve was 77% for radiology residents, 81% for 54 general radiologists, and 82.5% for 23 chest radiologists.

Mammographic studies report an ROC curve area of .74 for radiology residents, and .84 for radiologists, raising concerns that in the future, diagnostic performance might worsen (sk71).

In a study concerned with detecting lung lesions (sk142), ROC curve areas, or Az's, ranged from .73 to .80 (out of a possible 1.0), with the authors observing that the accuracy of the radiologist reading is sensitive to even the order in which abnormalities of concern appear on a checklist (sk142). This suggests they are responding mostly to what is specifically asked by the history, not necessarily responding to the challenge of the film.

That 80/20 is not merely an accident of human error, but that it is regarded as a reasonable achievement and goal is evidenced by the ROC curve areas of .79, or otherwise in the .70 to .80 range, achieved by artifical neural networks used to identify nodules using CAD on chest radiographs (sk207,sk196).

In another study (sk232p590), arbitration accuracy was assumed to be, and was set at 80%.

In a mammography series, about 25% of cancers were detected by only one observer, suggesting that diagnoses requiring consensus readings might damp performance down to near the 75% level (sk234).

80/20 radiology may be a balance between flash viewing, with reported 70% accuracy, and free unlimited viewing, as if one had all the time in the world, with 97% accuracy (sk106p1246), consistent with the finding that when search time is limited, (sk79) concentration is focused on high probability areas - the 80/20 payoff.

Various Practice and Interpretive Patterns Suggest an 80/20 Approach

Numerous studies (sk89p941) imply a rule of thumb, time efficient, 80/20 performance pattern. Satisfaction of search errors are about truncating search after easy and obvious abnormalities absorb attention and truncate what might have been a vigilant search for subtle abnormalities (sk47) - stopping the search when searching gets difficult - stopping at 20% of the effort, for 80% of the gain.

The sampling strategy described in detail by Kundel and Nodine (sk50p321) is essentially an 80/20 strategy, as is the pattern found by Christensen (sk14 ) regarding the tendency to focus on the rapid phase of image interpretation, and truncate the slow phase of interpretation. This might explain the U shaped performance curve of radiologic expertise (sk223), with the nadir of the curve when adequate skill is achieved but with even more effort, experience, and time needed to advance to expertise and excellence.

A study evaluating observer performance reading CT scans of ovarian carcinoma found little value in checklists and consensus readings, but also observed that a different conclusion might have been obtained if only the difficult cases were evaluated (sk230,p409) - in other words, for the population of cases including easy cases, acceptable performance - or 80% of the gain - may be relatively simple to attain with the initial effort or interpretation. Better performance on the harder final 20% of cases might require more exacting interpretive methods, including checklists.

The finding that large areas of images go unexamined by competent radiologists (sk112) suggests that 80/20 decisions are made, perhaps unconsciously, in the process of image interpretation. 80/20 results should not be surprising if a significant amount of the information on diagnostic images is not evaluated.

Deterioration of quality (sk88) is said to occur in the hands of less than careful radiologists seeking to minimize the work involved, using the criterion of "adequate for diagnosis" (sk99) - in other words maybe 80%, not 90%, or 95%, or 100%, accuracy.

Some find that acceptable interpretive performance can be obtained with only moderate training (sk119p1093) - consistent with 80% rather than 100% effort.

An inverse relationship claimed between quality of radiology training and radiologist salary (sk95) suggests that 80/20 interpretive approaches are well rewarded, perhaps explaining the traditional error rate of 20-30%. Well trained graduates, if able to bring more search skills and questions to a film, might become obligated to search longer, and take longer to answer the questions a heightened awareness raises, thus potentially lowering billings.

Low fees for screening mammography might (sk28) encourage less than the best diagnostic performance needed to detect subtle findings.

With time limited, visual search focuses on locations that past experience has identified as high yield (sk214),

a trait that may reinforce the continued missing of those lesions that haven't been found in the past either (sk103p367) - the false zebra effect.

Recommendations to trust only early confident diagnostic decisions and then to quit searching (sk231) may approximate the 80/20 approach, because it is likely that the tougher last 20% of diagnostic calls will mean more work and effort.

Practice methods that tend to distance the physician (sk271) or radiologist from the patient, using 80/20 economically efficient technologist intermediates might explain a lower sensitivity of obstetric ultrasound in the United States compared with Europe (sk82). The economics of health care structure in the United States, said to promote more non-specialist imaging interpretation, has been suggested as a cause for large discrepancies between specialist oncology CT and MRI re-interpretations and initial interpretations (sk240, sk248).

While clinical information can be used to focus attention, and increase sensitivity for findings of explicit clinical concern, it might also be (mistakenly) employed to select search strategies that avoid the inefficiencies of searching for findings prejudged (correctly or incorrectly) irrelevant in a given clinical context (Sk119,sk37)- essentially an 80/20 strategy.

The wisdom of the 80/20 rule, generally used to described efficiencies of customer/commodity transactions, seems inappropriate for health care. Even Russian roulette, having only one poor result out of six, compares favorably to 80/20 methods, offering slightly better odds at 83/17. 80/20 medicine may be distasteful to patients, producing mixed feelings about examinations which, if they really do have disease, and applying a "Dirty Harry" (sk215) standard of care, they must first ask themselves if they are feeling lucky.

The U curve of learning, 80/20 Radiology, and Knowing Grass in Detail

Bradbury's metaphor about speed and knowing grass in detail may help relate the U-curve of radiologic expertise to the 80/20 law. The easy initial 20% of effort to attain 80% of the results can be done with an intermediate level of expertise that mostly uses an automatic/proceduralized processing (sk223). This is likely the nadir of the U curve of performance versus training/experience.

Lesgold's work seems to confirm this (sk223). Initially, slowly, beginners pay attention to everything, and begin to acquire a memory collection of normal and abnormal appearances. As time goes on, they may sort appearances into categories of findings, noting more commonalities, and doing more lumping than splitting of findings and appearances. They move from the slow, analytical declarative or controlled level of processing (sk223p51) to the faster proceduralized or automatic level of processing. The "automated schemas" that take control in the intermediate state of experience (sk223p52) are not as accurate as the more discriminating schemas that develop with subsequent years of paying attention to detail. With more experience and expertise, if the intermediate performer will again go slowly enough to pay attention to details, to see grass at high power, and not as a blur, the intermediate will be able to make the fine distinctions that comes with expertise (sk223pp53,54,55,56). Once past the beginner stage, if intermediates lack the time to go slow enough to pay attention to details, they may not progress beyond the intermediate stage of rapid 80% processing, to the expert stage of rapid near 100% processing.

80/20 and Human / Institutional Limitations

80/20 radiology, and the traditional 20-30% error rate, may be the best accommodation radiologists can make in trying to balance patient care, economic, and scheduling demands. Poor top-down decisions or deficits in organizational culture pass latent errors downward, creating 80/20 conditions that facilitate execution errors (sk120,sk79p414, sk274p37), at the sharp end of the organization.

Examples of latent errors would be equipment purchases that make work more cumbersome and that divert limited perceptual resources from image interpretation. Less than optimal training for staff in general produces more marginal performance. Physical layouts that waste time and effort take limited resources from patient care and diagnosis. Finally, throughput demands may use up attentional and perceptual resources achieving quantity and volume goals, rather than the-harder-to-nail-down quality and accuracy.

80% accuracy, with 20-30% errors, representing the execution errors of a particular radiologic physician, might be 100% of the performance that can be reasonably expected given the history of latent management and regulatory decisions and errors passed into the clinical arena (sk89).

Rapid technological change (sk239,sk198p61) punishes taking the time to completely understand the intricacies of new technology at the 100% level. 80/20 radiology, might, or might not, be the best performance attainable in a given health care environment. With rapid technical change, by the time anyone's learning curve has conquered the 80% of potential gain on an old technology, the technology is updated or replaced, and few will master the final 20% of diagnostic gains, much like new software replacing the old stuff just when you've finally got it figured out.

80/20 radiology should not necessarily be accepted as the most reasonable error/effort tradeoff however. Small gains in accuracy, when put into a larger context of overall outcomes, might be well worth the effort, yielding significant improvements in cost per quality of life year, as some authors have suggested (sk78p844). The full costs of false negatives, like the long-neglected uncounted costs of industrial pollution, are not as readily tallied in dollars, or in ledgers, not as readily balanced against the financial costs of false positives (sk232p593), and become thereby easier to ignore.


Even older reports hint at time truncation, with time per case not correlated to the number of abnormalities present, with lesion detection diminished after the first abnormality found (sk26), and with search time fixed for most cases (sk26).

Truncating image inspection before the detection ratio of true and false positives becomes unfavorable, but with abnormalities remaining to be found (sk23, sk14p363), seems to be an example of premature search truncation.

Using a Fixed Time per Case Privileges Speed

While a fixed time per case might imply a well-honed and complete search pattern, the observed failure to detect abnormalities after finding an initial one suggests that searches are often incomplete. The notion of 'satisfaction of search' - that after one abnormality is found, then others are not found - may simply mean that each study is given about the same amount of limited fixed time, allowing for only a rapid phase search , as described by Christensen (sk14), adequate to find obvious lesions (sk17), but with searching then truncated prior to using the observational disciplines, rigor, and time necessary to find subtle lesions (sk47).

Optimally, search time and habits should vary in proportion to the complexity and demands of an image (sk79p409), meaning that a limited or fixed search time negatively affects search pattern, work habits, and judgment (sk79). Terminating search after a set time has passed is probably not the best criterion for when to break off search (sk61), thought it might, if long enough (sk127), be an improvement over time truncation methods.

The observation (sk26) that a relatively fixed search interval of 30-40 seconds, is constant, for "strategic reasons", with " a lack of perceptual resources" seems too broad to suggest specific remedies. "Strategic reasons" might mean reasonably ceasing to identify possible abnormalities before the penalty of false positives becomes too great, or it might mean reading films for maximal productivity, according to the 80/20 Pareto law, writing off the 20% of findings unfound as normal and customary error - tolerated to keep false positives low, and work 'efficient' - thus maintaining the time-honored 20-30% retrospective error rate. Reading strategically may mean calling abnormal only high confidence findings, which implies writing off the tougher calls. Limited "perceptual resources" might include search habits, clinical history, knowledge, experience, labor saving devices, solitude, attention and, especially - time (sk61).

Attempting to be prudent, a radiologist might adopt the strategy of terminating search when findings are of such a high confidence level that they are certain to be followed by fruitful action (sk23,sk60). Search truncation at a point when true positives outnumber false positives does cut the costs of wild goose chases, but the counterbalancing costs of missing subtleties often go uncounted (sk132). Outcomes assessment terms like 'efficacy' describe right actions that follow right thought, but like Sherlock Holmes' dog that didn't bark in the night (sk94), a term or measure for the non-efficacy of missing subtle findings usually detected by slow searching (sk14) has not been popularized like 'efficacy' has.

Not Extending Effort to Match the Challenge of the Image Builds Speed

Deciding to limit the effort one gives to difficult, or error prone, tasks can limit performance. With images rotated to alter the usual gestalt of viewing, radiologists did not significantly increase the amount of unlimited viewing time to compensate for this (sk144), resulting in poorer lesion detection, all this suggesting that there is some resistance to increasing the time allotted per case.

A reported tendency (sk49p83) to miss lesions outside the stated region of interest implies a search truncated to the extent such limitation will be tolerated, offering no more than what is explicitly asked for (sk32). The tendency of observers to describe what they expect to find also implies a perceptual pre-set, and a motivation apart from searching to find any and all abnormalities (sk49p99), and increases the likelihood of the false zebra effect.

Interpreters may become distracted by a positive finding (sk17,sk65), and truncate search prematurely, somehow believing that disease processes take turns, or are mutually exclusive - with the presence of one disease precluding the possibility of others - much like a vaccination. This is the fallacy of the two bombs - where a man smuggled his own bomb onto the plane with him, supposing he was thereby safer because the statistical likelihood of two bombs on the same plane was so remote (sk273). Perhaps patients with two or more abnormalities are beyond a normative quota of one lesion per patient, using up an excess of scarce perceptual resources, and violating the norm that everyone is more or less entitled to an equal number of abnormalities.

Normative Practice Flaws Can Build Speed

With the use of CAD, total reading time is held constant, or decreased (sk35), with routine non-CAD visual search performed less rigorously in the truncated total time available for routine viewing. Perhaps in highly technological fields, powerful equipment fosters the view that the human factor is relatively unimportant, but radiologists already have remarked on how critical the human factor is (sk42, sk65,sk119).

Skilled readers under time constraints shift focus to high yield areas (sk79), going for the 80%. Others observe that adequacy of coverage, and confidence in that adequacy, are related to time spent in perception, but with lesions nonetheless missed even in regions where they are known to be most likely (sk79).

Vascular abnormalities may be missed at CT (sk77), occurring in inconspicuous, but sometimes physiologically critical, structures. Lung windows are sometimes omitted from CT series done for oncology patients (sk89,sk240). Musculoskeletal abnormalities, one of the most common abdominal CT misses, might not be missed if bone display settings were used more often (sk77), a situation similar to Goodall's experience with high school students who had stopped their observation of animals to play cards, thinking they were done with the job, and declining to use the binoculars they had with them to look harder, the technical capacity of the equipment outstripping the interest and motivation of the observer (sk195) - essentially normative error, as described by Bosk (sk5).

A substantial variance between specialist and outside interpretation of oncology imaging (sk240,sk248) was suspected in part due to a mismatch between generalist interpretive expertise and the high-technical quality of the equipment.

Busy schedules may predispose to speed and search time truncation, producing an increased number of errors on cases interpreted on Mondays (sk49p72), with lack of time perhaps limiting standard observational maneuvers like hot lighting overexposed films (sk49p93). Sensitivity has (sk112) decreased between morning and post lunch sessions, with less time available in the afternoon prior to quitting time.

Even comparison with previous images, standard to detect changes in position and size of normal structures (sk118, sk18,sk24,sk53) may be neglected to save time (sk37,sk49), accounting for false negative exams, with half of (sk78) potentially resectable lung tumors overlooked because previous studies were not reviewed. Sometimes omitted, comparison with films older than the most recent radiograph might aid detecting tumors, since short interval comparisons often show no detectable change, but a longer interval comparison might allow a diagnosis Aunt Mabel could make - "It's been such a long time, my how you've grown".

We do not Have an Unlimited Amount of Time

Unfortunately, few studies detail the effect of search time, and search time truncation, on ROC curves of radiologists. One study doing so (sk127) found that short viewing times result in the greatest differences in lesion detectability - in other words, short viewing times results in the greatest rate of misses of subtleties. While this study reported that "unlimited viewing time" resulted in a false negative rate of 26% for subtle lesions, since the films were not viewed for an infinite amount of time, this 26% miss rate must correspond to some realistically finite amount of time, perhaps this time also inadequate to find subtleties. Another study allowing an unlimited amount of available viewing time, had a calculated median of about 2minutes per 2view mammogram (sk135).

The term "unlimited" viewing time can be misleading, creating the impression that even an infinite amount of search time is still inadequate to find all abnormalities. This can then lead to underestimating the value of longer search times, and obscures the possibility that longer viewing might diminish false negative rates. Some authors seem aware of this reluctance to spend time on a film, making the suggestion that a longer "enforced" viewing time might be explored as a remedy (sk127) for missing cancers.


The degree to which perceptual error occurs in medical imaging is substantial (sk102). A total miss rate of 20-30% (sk112) of confirmed abnormalities is a generally accepted figure, with some studies reporting higher rates of error (sk131), ranging up to 90%, for subtle lung carcinomas (sk138), seen in retrospect (sk105), and with perception error contributing to 84% of errors in the detection of carcinomas on double contrast colon exams (sk105).

Errors may be due to search defects - looking at only a portion of the image, or due to recognition error - looking at the abnormality, but not seeing/recognizing/perceiving it (sk108). Recognition error and limitation of perceptual skill are generally the most important factor in the quality of image interpretation (sk45b,sk42,sk65,sk18,sk33,sk17,sk37,sk131,sk137). Perceptual errors are common, even in the presumably more vigilant context and focused setting of a research (sk84) study.

Especially in plain film diagnosis, not only must each portion of an image be viewed for abnormality, but one section must be compared to another, to detect subtle abnormalities of proportion, position, curvature, size, and density that can only be seen in the context of the appearance of the rest of the film.

The effects of technique, rotation, magnification, and the effects of pathological changes, such as atelectasis, on otherwise normal structures must then be factored in. Overexposed film regions might need a bright light. Changes in size, or position, of a normal structure might be the only sign of underlying disease, not detectable on any given exam, but detectable only by differences between exams. Comparison with old films involves comparing each region of each film, and allowing for technical, projectional, and other differences between the films. Known regions of frequent error may get a second look, as might structures pertinent to the clinical history. All of this is a complex task, and often a difficult one, even in the hands of experts (sk37), and not accomplished in milliseconds.

As imaging techniques get better at clearly displaying abnormalities, observer error becomes the performance limiting step, and therefore of more concern (sk119,sk84). Many missed findings are adequately imaged (sk8,sk88,sk102,sk65), and are obvious in retrospect - similar to Dillard's observations in natural science (sk11). Awareness of perceptual errors is difficult, because much of perception is unconscious (sk21,sk65,sk113).

Orderly Looking and Deliberateness Aid Perception

Orderly looking precedes orderly perception. An orderly approach to looking at images is held to be important in precluding radiologic errors (sk119,sk18,sk22,sk70,sk64,sk67,sk49p85). Systematic analysis of all anatomic compartments is recommended (sk90), a heuristic method particularly useful in CT (sk204). Radiologist search of a normal chest film is said to show a consistent pattern, (sk50,p321), but this is altered by clinical history (sk50p322). Better sensitivity in interpretations is said to demand a more careful review of images (sk105), and so experienced radiologists examine the chest periphery more than beginners do (sk112,sk214).

Orderly Search Is Not Necessarily the Rule

Despite calls for orderly search coverage, it is uncommon (sk50,sk65,sk18,sk88, sk17,sk79,sk70), even among experts. 83 percent of scanning patterns deviate from a systematic search (sk70). The usual visual search may omit up to half of the image (sk79p47,sk106,sk112 ), with observers imagining incorrectly that their search is orderly (sk79). Defects in search routines, such as ignoring the abdomen when viewing chest radiographs, result in misses of abdominal masses (sk32).

Right upper lung cancers are underdiagnosed compared to the left (sk37), perhaps because the visually prominent aortic knob leads the eye out into the peripheral left apex, but looking that far out on the right demands a specific effort to do so. Maybe left apical masses in patients with a right aortic arch or azygous continuation of the inferior vena cava would go similarly underdiagnosed.

For chest images, judging from the size of fixation clusters (sk106), perhaps scanning along the ribs would not only systematically examine the ribs, but also the lung fields, with the radius of foveal vision overlapping from rib to rib, offering more complete coverage of lungs and ribs simultaneously.

Kundel and Nodine (sk50p319) found radiologic viewing neither complete nor methodical, with errors occurring in the first 10 seconds of a search that fails to evaluate the entire film, but merely samples it (sk50p321), rather than evaluating each region in detail. Each visual fixation is given a limited time - which then limits the information that can be perceived, a form of time truncation, and a version of the 80/20 rule- occurring even for areas looked at. This failure to focus on lesions contributes to such error (sk8,sk18,sk112). A longer detailed look can significantly improve results (sk135). Target detection rates can be improved (sk245p222) by using a restricted display size to force looking at all the image, and thus increase scanning coverage.

The time that radiologists devote to portions of the chest can be heavily skewed by history (sk50p321) producing a focused, but skewed, and therefore disorderly search.

A mistaken focus on large, visually impressive findings (sk88p285) that become figure, may distract from noticing findings that are more significant for diagnosis. Visually small clots, or collateral blood vessels, on CT may have more clinical significance than large renal cysts.

That the critical flaw in film reading is the temptation to stop searching before looking at the entire image (sk65, sk88), basically says that search truncation/time truncation is a major perceptual problem.

Even Orderly Search Has It's Own Limitations - Perceptive Looking is Essential

Orderly search in itself does not insure perception (sk64,sk88p287). The many examples of optical illusions show how much perception can differ from merely looking. Some significant findings go unfound, even when looked at, suggesting that merely looking does not insure diagnosis (sk65,sk103,sk125), since perception requires a look that is discriminating.

The ability of normal lung structures to mimic nodules of small size is an intrinsic limit to radiologic accuracy, even with systematic search (sk125,sk137). Even orderly, segment by segment, image insepection (sk107) may fail when the object is similar to background, since there is less other normal lung to compare appearances to (sk245). Orderly comparison (sk125,sk70,sk112) scanning, looking side to side for subtle asymmetries between comparable parts of the lungs, may help.

Recognition error could be considered a form of search error, one related to search method, rather than search coverage. Neglecting portions of an image is an error of search coverage, but looking with inadequate time (sk14p79), sk44,sk47,sk79) or with inadequate observational maneuvers, constitutes an error of search technique, an error of underestimating the effort needed to find abnormalities (sk17). An orderly search may in this way be inadequate, and instill false confidence, or instill a sense of haste to get through a long orderly checklist, some studies describing more, not fewer, errors on the part of radiologists thought to use an orderly search (sk49p85,sk65p695). Merely using more time to look, without discriminating use of the features of normal and abnormal that enables perception, can also increase false positives (sk71,sk88).

Perceptive looking probably requires knowledge, experience, and discrimination regarding the features of normal and abnormal, time to do the work of perception, a method of looking to insure search completeness, and the proper attitudes, including humility (sk132,sk49p86) and attention, that make for looking as if looking counted.

Perception Requires Experience, Knowledge of Target and Background, and Time, to Disembed Lesion from Background, Using Visual Handles

Inability to distinguish abnormality amid background, to disembed it, is an important and common source of recognition error (sk64, sk65,sk106).

Perception occurs when an abnormality is distinguished from distracting background features, or noise (sk27,sk56), because its essential features are known, and are then visually separated from the background (sk67). The expert paleontologist visually disembeds fossils from other rocks in strata, or gravel piles, and gem hunters pick out mineral bearing strata and rocks that they know are signs of paydirt. This (sk107,sk44,sk21,sk88p287) requisite familiarity with what normal and abnormal look like, and where to look, give the expert the power to lift subtle lesions out from among a sea (sk64p255) of irrelevant and obscuring image features. "Chance favors the prepared mind" (sk265).

Disembedding takes time (sk226pp252,254), with sampling longer for hits than for misses. Deeply obscured visual targets require more time to be discovered (sk106p1247).

Subtleties can be made conspicuous by technical aids that simplify search by increasing visual signal - like contrast material, or with aids that minimize visual complexity and noise (sk88,sk131,sk126,sk132,sk134,sk137), such as conventional tomography (sk126), computed tomography, image subtraction (sk126), stereo radiography (sk88,sk112), dual energy digital chest radiography (sk134), and automated vascular image extraction (sk132).

Visual handles, such as 'signs' in radiology, or "second order relationships" in an image (sk99p271), are cognitive tools that similarly simplify visual search (sk102)- key features that enable the disembedding, or recognition and mental grasping of part of an abnormality, and by this handle of recognition, controlling all of it. Handles are the "artificially obvious" keys to perception that White describes in seeing deer (sk12). Seeing brown amid green, you suspect deer. Seeing a straight edge in a natural setting, you recognize human activity. Seeing barium drop at fluoroscopy, you suspect a gastric polyp. For example, high density amid vessels on a chest radiograph is a handle (sk134), or perceptual tool, that disembeds lung nodules from adjacent pulmonary vessels. The 'A' in Nina for example, may function as the 'handle' that helps Ninamaniacs find the word Nina in Hirschfeld's New York Times cartoons (sk226,p257,sk106).

Searches for visual handles often require a close look because slight changes in the background/target spatial relationships may significantly alter detectability (sk134). This (sk49p79) isolation of an object amid background, however, takes time, at the very least takes time in learning how to do it.

Experience with visual handles - learning target features - lasts for years (sk103), suggesting that the false zebra phenomenon is perpetuated by never learning to see signs of abnormality in the first place. If facility with visual handles expresses the effects of rigorous training of the "What" system (sk146p64), failure to recognize lesions may simply mean that detailed visual training of the temporal lobe has never taken place.

Visual perception, like hunting bear or grouse, requires tools, insight, and methods specific to the quarry sought (sk71,sk67), but the radiological problem is generally tougher than the hunter's, needing to look for not just one disease, as in screening studies (sk233,sk232), but all possible quarry, and never with certainty knowing, or being able to choose, whether we're hunting quail, or bear (sk222p105).

Proper Attitudes Themselves Are Perceptual Resources

In addition to time and experience, proper attitudes are a critical perceptual resource, as Pirsig observes in ZAMM, and as Herrigel's master archer emphasized. Both Smith (sk49p96) and Garland (sk63) remark that basic life perspective and philosophy of the interpreter, are important factors in improving film interpretation.

The first of these attitudes is humility, the beginner's approach of taking the attitude that "I don't know what I'm looking for" - building vigilance for any deviations from normal. The second attitude, that of attention - "I do know what I'm looking for" promotes doing an informed, careful search.

This paradox - valuing both "I don't know what I'm looking for" and " I do know what I'm looking for" - probably explains why free search for abnormalities is superior in some instances, and why clinical information nonetheless improves radiologic interpretation (sk119).

Humility, Beginner's Mind, Patience, Attention, and Time Are Required for Perceptive Looking

Deliberateness and perception require the humility (sk49p86) that respects the film as a potential source of knowledge and challenge, rather than as a routine chore to be dispensed with as quickly as possible prior to doing other more interesting things - and behaving like Goodall's high school students (sk195). Some authors, by suggesting that initial film review be made with ignorance of clinical history (sk79,sk88), cannily manage to use even ignorance as a tool - by using it to enforce the kind of humility that fosters perception. Humility avoids the feeling of having found something valuable which prematurely ends the search for meaning, producing satisfaction of search error.

The strength of humility is the strength of free search, done without clues (sk70,sk119,sk222p106) - done before clinical history fixes perceptual set, and makes you think that you know ahead of time what is probably on, and not on, the image (sk79,sk226,sk60). Humility, interpreting without clinical history, means knowing that failing to consider a diagnosis does not diminish its likelihood, and that one needs to visually confirm structures as normal, before presuming to exclude abnormality.

Several studies have show that in some instances, (sk31pp173-174,sk195) a humble beginner's mind effect, as with radiologic technologists, reader B (sk233), and R2 (sk245) results in less experienced, but perhaps more motivated and involved, observers having superior performance (sk224,sk225). It may be this humility and attitude of the beginner that is one of the individual traits that is more predictive than experience of accurate interpretation. Attitude may explain the results of studies where the best performance was found where it was least expected, in persons with less, rather than more, training. These precocious performers probably bypassed the normative error and limitations of non-involvement. Perhaps the most effective experienced interpreters work their way back up the U curve (sk223) since, with more cumulative years of errors to remember - errors made despite experience - they rediscover and retain humility.

Satisfaction of search, or truncation of search, is essentially a humility deficiency state, occurring when obvious abnormalities capture visual attention (sk65,sk47), and make us feel we've got the answer, or the "quest for meaning (sk88)", in the bag - diminishing vigilance for more subtle abnormalities.

Berlin (sk24) essentially recommends humility by asking radiologists to give time and attention to possibilities not mentioned in previous reports, rather than speed to a conclusion by reflexly repeating previous opinions - but doing this requires patience, and an independent mind. The haste to consensus, conclusion, and rapid interpretation, can (sk47) decrease the probability of finding other abnormalities.

Carl and Nodine (sk50p328), similarly suggest consciously delaying a look at the more engaging parts of the chest, and instead doing the less exciting, humbler parts first, like soft tissues and ribs, and then lastly look at the higher yield features, like the lungs and mediastinum. Schneider and Shiffrin (sk52p158) observe that those areas that receive attention first usually are evaluated more quickly and accurately, a tendency that inserts a bias when many tasks must be carried out, as in medical imaging. Saving the best for last retains the edge of attention and vigilance, and minimizes the distraction of quick and easy findings - those that produce satisfaction, and premature truncation, of search.

Like the Zen archer, or flower arranger (sk15, sk15b), experts (sk44) give more time to establishing the fundamental aspects of the problem before looking for the solution (sk31pp173-176; sk223,p51paragraph 2;sk195)- in other words, first doing the basic perception work (sk79,sk47,sk39), while having the discipline and self control to leave conclusions wait until all the data is in (sk63) - separating the work of seeing and judging. Such an approach avoids the pitfalls of anchoring bias, availability bias, regret bias, and framing bias (sk89p941) - biases which also happen to produce convenience, and time truncation. Good readers probably restrain their desire to come up with answers quickly, like Herrigel's archer - remaining humble enough to restrain distracting thoughts about hitting the bull's eye, and just doing the archery right.

Attention is a Perceptual Resource

After a humble free search, a directed, systematic search using history or patient demographics to prompt attention, and to ask specific questions of the film and of oneself (sk88), becomes a followup antidote to the prior use of humility, with attention built from the sense of accountability (sk6c-p39) essential for a perceptive, complete film review. Attention - the opposite of humility - corresponds to feeling obligated to read as an expert and knowing searcher, made possible by experience and clinical history. Attention , or vigilance (sk47) is a critical perceptual resource, (sk58,sk44) necessary for the disciplined "discovery scanning" needed to find subtleties (sk47) - and required to turn orderly looking into orderly perception.

Detailed focusing of attention is known to increase accuracy(sk64, sk39,sk49p78). By giving time to study an image region by region, nodule detection is enhanced (sk59). Comparison scanning, close side to side comparison of symmetry, is a recommended method to detect subtlety, though said to be not practiced much beyond training (sk70).

The threshold for detecting a specific abnormality is lowered by a high index of suspicion (Sk119). Radiologists expecting to find something have a better chance of doing so than those expecting to find normal. (sk49pp93,103,sk103p367,sk63). Radiologists not expecting to find cancers in women as frequently as in males, manage to find them less often, even when they are there (sk37). Assuring viewers of the likelihood of lesions can triple detection (sk79p413), perhaps by increasing vigilance and openness of the viewer to imagine, like Pirsig's mechanic with the headbolt, that some humble shadow out there might mean something significant.

Studies reporting that readers excel at perceiving different types of lesions (sk60) imply a selective inattention for those findings usually missed, perhaps reflecting the attitude that some classes of missed lesion are unworthy of attention, a privileging of one's own intellectual interests over the interests of the patient (sk49p19,sk63p326). In these instances, reading slowly did not preclude errors.

Attention uses varying methods for varying situations. Each region of an image has unique characteristics. The mediastinum, chest wall, neck, and upper abdomen, lacking the natural air contrast of the lungs, need a closer, slow phase inspection (sk17). Rib fractures might be better detected by reorienting the film (sk75).

Visual search requires looking long and hard enough to perceive what the eye is looking at. The rapid fall off of accuracy only a few degrees off the line of vision means that looking directly at one region, while using only the peripheral vision of a glance to cover others, impairs the perception of those peripheral areas (sk47, sk18,sk76,sk8), which may explain why viewing a chest film from farther away can improve detection of abnormalities (sk205).

Adequate Time is Necessary, but Time Itself is Not Sufficient

Quantity of time alone, without the observational disciplines that enough time allows, is probably not adequate to improve quality of perception. Smith described how time is important for specific tasks (sk49pp78-80), but could not state if differences in radiologist error were due to differences in interpretation speed (sk49p97). It may be, consistent with Dreyfus' observations with chess masters (sk71,sk6c,sk6), that highly accurate, but fast, radiologists had in the past given more time to learning from the beginning to read accurately, and over the years added speed to accuracy, providing rapid accurate readings that take only 30 seconds - and 30 years (sk44,sk51p2). A sufficient quantity of quality time, with practice of adequate observational disciplines, may be required for quality perception , when those who are to become expert relearn the skill and habit of routinely seeing grass in detail, as found with swordsmanship (sk15pp71-78), and acquiring expertise in anything (sk223).

Lesgold's paper (sk223) may offer some guidance. His slower, declarative knowledge (sk223p45) seems to correspond to the controlled process of Schneider/Shiffrin (sk51,sk52), and his faster compiled knowledge and routines to their automatic processing. Declarative knowledge requires more attention and time (Sk223p47), but with practice less time is required (Sk223p48). Time required increases proportionally to the complexity of the task. This creates an opportunity for satisfaction of search and time truncation. If the nature of the search task is prematurely oversimplifed, the faster compiled search routines can be developed and soon used, even with complex tasks, rather than using the more laborious declarative approach. Time truncation thus is about deciding a priori that one only needs to do the simple and easy things - a glance, not a search.

At this point, developing a consistent method of looking in some ways for some things, and neglecting looking in other ways for other things allows the premature use of rapid compiled knowledge routines, giving speed. This means looking for obvious hilar enlargement, not subtle hilar enlargement, or looking in the lung fields for nodules, but not behind the heart or through the diaphgragm which might require the use of slower declarative knowledge because many overlapping shadows must be processed in a more laborious fashion. If the perceptual tasks undertaken are limited by the 80/20 rule to those which can be done under speedy, "productive" compiled knowledge, subtleties - the details of grass - will be written off as too hazy or blurry to be bothered with.

The negative pairing, negative transfer effects that Lesgold refers to (Sk223pp48-49) is a problem that radiologists often face when image appearances, being the final common pathway of a multitude of processes, signify many potential meanings. The tendency of negative transfer to be more important the more difficult the task makes this effect particularly important in medical imaging. In imaging, the potential for error is great, since the inability to get rapid, or any, feedback on such errors means error will often persist unchanged. When faced with such complexity, and with lack of feedback, error evaluation again requires slower declarative methods, rather than time truncating compiled knowledge. 80/20 environments may not have time for declarative knowledge, especially since it is hard to get truth data on the final, most difficult 20% of the total task.

Given the U shaped relationship between transfer and the amount of original training (Sk223p49), this suggests that for complex tasks, like medical imaging, longer and longer amounts of initial training might be needed to get beyond the nadir of the U shaped curve. Positive transfer occurs if the amount of the original training is sufficiently large. Attempting to apply this to medical imaging, it may be that giving time to get the basics securely understood - the advantage of slowness (sk1pp9-11), makes subsequent learning easier and lasting. Satisfaction of search probably occurs most for those persons who did not devote enough initial training to the task of knowing grass in detail - as more than a green blur.

The production of accurate automatic compiled processes (Sk223p50) may be the result of learning initially about grass in detail (sk1), and after that, knowing it forever. Then using this knowledge becomes not only fast, but remains accurate, much like knowing phonics makes reading easier, once sufficient attention has been given to knowing the alphabet and its sounds in detail, like knowing grass in detail. Once the basics are securely understood and automatic, subsequent knowledge is more easily assimilated and usable.

This type of learning in detail may be the crucial thing that happens when enough time is given to learning the methods and rudiments of perception.

Giving More Time to the Task of Perception Creates a Moment to Use Correct Attitudes and Methods

One approach to minimize perceptual error is not to take more time, but to give more time to each image. Significantly, the term - 'take more time' - contains the curious, and obscuring, language inversion that if one does not 'take' more time, it sounds almost like an act of generosity. That CAD improves performance of inexperienced readers implies that looking longer, if only to look differently, might improve performance (sk8,sk55,sk35).

Using an adequate amount of viewing time is (sk92,sk90) often critical for quality radiology. Bothering to take time to look harder can, on occasion, overcome even lack of experience(Sk245). Many authors recommend giving more time to look (sk49p84,sk65) and distinguish lesions among other structures (sk49p79,sk18), since limited fixation time limits ability to detect abnormalities hidden among structural noise and obscuring features (sk50p323). Others find (sk66) performance better with unlimited viewing time. Schneider and Shiffrin (sk52p159) essentially identify satisfaction of search errors when they observe that perceptual failure occurs when the subject decides in advance that further work is unlikely to be profitable, or that there is some arbitrary time limit that precludes the possibility of further effort - essentially the 80/20 approach.

In one study ( see both sk127,sk119), more lung lesions were missed with less than four seconds viewing time. Other studies have found longer viewing to occur with both false negative and false positives (sk200), these errors correlating with the inherent task difficulty and greater attention needed to discern between normal variant and the subtly abnormal.

The relationship between time spent, and errors, is suggested by one group (sk68) whose initially preferred measure of performance - time spent, rather than errors - was confounded by residents who, confusing speed with performance, chose to finish faster, while making more errors, producing the paradox that the less perceptive feel free to quit sooner than the more capable.

Giving more time creates a place for developing the deliberate observational habits needed to detect subtleties - to see more than a green blur. Subtle findings, like deer legs in a thicket, or upper lung lesions (sk37), require inspection, rather than a glance. (sk49-p102p79) Slowness makes time for (sk49p80) search disciplines analogous to the deer hunter's looking especially closely at patches of brown (sk12p120), a form of disembedding (sk65) or "visual isolation" that Smith (sk49p79) describes. Useful speed comes only after this step. Generating error, and doing so quickly, takes speed that could be a virtue, and merely uses it to amplify error. Quality perception probably cannot be done at 200 mi. per hr - not surprising - even to perceptive laymen like Bradbury.

Just Sitting, and Searching, is Difficult

Giving sufficient time to image interpretation may be difficult because it can be hard to just sit with the film, like Eugen Herrigel's (sk15) painfully difficult and deliberately slowed learning curve for archery (sk15p51), or like Pirsig's struggle with a headbolt, slowed so that skill grew from more laborious conscious competence to relatively effortless and fast unconscious competence - identical to the learning process described by Dreyfus (sk6c, sk6), Schneider and Shiffrin (sk51, sk52), and Lesgold (sk223).

The state of active receptivity, of mindful attention, is hard to get used to. Just sitting, just looking, and nothing happening feels like a waste of time, life, and attention. Just sitting, confronting a film for the time it takes to look at everything, while saying nothing, can feel like doing nothing, and looks like it too. People start bringing you coffee. There is always pressure to DO something, or at least look and sound like you are. Referring docs standing behind you won't wait more than about 8 seconds for you to look at all 60 or more panels of a CT scan.

Connelly (sk31pp173-174), in Empathic Practitioner, similarly observes that just listening, and being actively receptive, is less valued than speaking - a general cultural bias which favors perceptive failure.

Connelly describes a special kind of attention that promotes perception, similar to the recommendation of others to first obtain a fundamental and complete representation of what you are looking at (sk44,sk15b,sk63,sk79,sk31pp173-176,sk223p51, sk195) - in other words, first doing the basic perception work (sk79,sk47,sk39), while having the discipline and self control to leave conclusions wait until all the data is in (sk63) - separating the work of seeing and judging. Such an approach avoids the pitfalls of anchoring bias, availability bias, regret bias, and framing bias (sk89p941).

Staff radiologists (sk14), even in the context biased (sk48) setting of research studies, were reluctant to use more than half of the allotted 4 minutes, and content to detect only 68% of clinically significant lesions, with the length of reading sessions limited by the energy of the observers. Perhaps the readers' usual search habits were exhausted before they had fully evaluated the film, leaving them feeling bored (sk6c-p38). Goodall observed a similar phenomenon in the jungle among (sk195-p58) a team of high-school students with well-trained habits of inattention. When animals and information were not jumping out and announcing themselves, they saw nothing, became bored, and declined to consider using the powerful observational tools at their sides - binoculars- and instead concentrated on playing cards. They lacked perceptual discipline, and were similarly unaware that they lacked it.

Patience to just sit with a film is especially hard if the reader brings less in the way of search habits, perceptual resources, differential possibilities, or interest, to the work (see also sk31pp170-182).

Goodall (sk195p55) describes the heightened perception that comes when the searcher breaks off the habit of the premature use of words - perhaps a variation on the the need for image interpreters to just sit with the film in receptive silence, and look at the film, and refrain from an immediate outpouring of words in the report. Similar to satisfaction of search effects, she notes that premature labeling leads to less careful looking. It would seem that it is better to be perceptive first, and productive second.

The feeling of being stuck with nothing happening might be overcome by greater deliberateness brought to the task. The deliberate archery of Eugen Herrigel's master archer, focusing on each element of technique, with no thought as to when and where the arrow would land prior to seeing it land, is an example of the discipline which might make longer film viewing more tolerable.


It may be that, as Bradbury, Kundera, Pirsig, Dreyfus, Dr. Heilman, and Drs. Ania/Asenjo (sk3) point out, technology blunts perceptual capacity by diminishing patience and attention span. Attention deficit disorder probably affects all citizens of technology - and applicable to more than "Why Johnny Can't Read". The virtual reality of television and the computer may velocitize adults as well (sk65), radiologists included, so that the patience required to just sit with a film becomes a scarce perceptual resource. Garland (sk63) reports, from long ago in a slower era with perhaps longer attention spans, that a dubiously negative case might require 5 or 10 minutes - not a 250 millisecond flash diagnosis.

Patience is needed for the slow, controlled search processes needed to find subtleties, and for the inspection of the many image elements that vary between patients (sk52p160). Those inclined to speed read (sk79p407,sk14) may lack this level of patience. Controlled search for example is generally required in comparing new to old imaging findings, if one hopes to catch disease in its early stages. Linear structures such as vessels, or ribs (sk210,sk75,sk64,sk49p73) demand a slow serial controlled search. But controlled serial search is time and labor intensive (sk51p3;sk52p127,p161). As in text reading (sk79,sk223p51), because automatic processing does not demand conscious attention, and is very rapid (sk52,p160) economic pressures favor shifting as much perceptual work as quickly as possible from the slow controlled processing into automatic processing.

The development of faster automatic processing begins with an initial slow serial/controlled processing learning phase. If detailed parts of the serial search task are omitted in the learning phase, the automatic processing that finally develops will omit the habits of detailed search and understanding. The reader develops, and is left with, an automatic processing equivalent to a blunt axe, rather than the more effective automatic processing of a sharp axe.

Recognition error may simply represent faulty controlled processes repeatedly performed under time pressure, such that they are prematurely transformed into automatic processes that are perceptually deficient. These "veiled control processes" (sk52p159) are harder to improve by instruction because they occur unconsciously, and rapidly, without giving time to see slowly - to see grass in detail (sk1) This creates a situation where subtleties are missed, since their detection was never learned in the first place - the false zebra phenomenon (sk103p367).

Time pressure may truncate the learning process, in a fashion similar to the person who reads, and while not knowing phonics, or not knowing the words, also fails to look them up in the dictionary, satisfied that a general idea of what the word means is adequate.

Perhaps radiology could be compared to text reading in some ways (sk14p365,sk79, sk223pp37,50,51). Looking at each letter in reading may be analogous to looking at each body part for a given pathology. A word may be analogous to looking at a body part - one rib, for example - for all signs of abnormality.

Image interpretation lacking detailed understanding and care - not knowing grass in detail - may be analagous to a partially literate person who speaks or composes text, misusing various 'sounds close enough' homonyms, homophones, or other 80/20 malapropisms, resulting in a text that seems just fine to the composer, but whose error results in a different message than that intended - misleading and/or amusing. Unlike malapropisms, medical errors and illiteracy tend not to be amusing.

In radiologic interpretation, this type of illiteracy is less apparent or detectable, since unlike text - with words composed of explicit and generally agreed upon letter symbols - radiologic words and sentences are chosen based on images that are not generally available to the reader of the report, and with the radiologic appearance and vocabulary far less standardized than for the alphabet and dictionary.

Short attention span is likely to be even more problematic in film reading than in children's alphanumeric reading - the latter itself a complex task (sk79,sk223,p50). Reading text is a highly structured task, done from top to bottom, generally left to right, with letters and words linked rigidly in sequence. The feedback from the task itself is immediately, and tightly, linked to further task performance. This structure makes it clear when the task is being done, or isn't being done - counteracting lapses in attention - a corrective that image interpretation is not blessed with. Image interpretation is far less structured than reading text is, with little in the way of task structure to counteract interpretive attention deficits, and little in the way of task feedback to prompt attention. Lacking the structural cues for quality performance present in text reading, image interpretation may have an especially great potential for being disrupted by culturally cultivated short attention spans.

Tunnel vision (sk138), immediate gratification, the Pareto Law, short attention span, time truncation, attention deficit disorder, and satisfaction of search may all be synonyms for the same phenomenon.


There are many impediments to slowness. Slowing down is tough to sell. Everyone hates speed bumps. Speed appeals to the demand for immediate gratification, and to the shorter attention spans of the electronic age. For radiologists, speed offers obvious financial and lifestyle benefits. Demand for immediate service and rapid report turnaround justify speed on the basis of medical necessity, cost savings, and public relations. Lack of cost effective measures of quality leaves the bias for speed and quantity unbalanced, since quantity of work is so much easier, cheaper, less painful, and less controversial to measure, than the quality of that work.

Besides the threat of lawsuit, or more frequently, professional embarrassment, there is often little penalty for wrong answers, so speed based errors don't count as heavily as speed based productivity. Tests that don't penalize wrong answers predictably reward guesses. Like stock brokers, getting paid as much for wrong answers as for correct ones, there is no braking mechanism to assure quality, but unlike lawyers, we are not paid by the hour - both being situations that offer incentive for the use of speed.

For payors and government overseers, speed produces an income or service generating stream that can be diverted into more revenue and profits, or tapped for higher service levels in a failing health care system. Similar arguments have been applied to the link between diminished NASA funding and the shuttle disasters (sk29,sk244). With shrinking financial commitment to health care, speed helps any health care entity serve larger populations with fewer resources, and still remain viable in the face of cost cutting, while anomalously violating the law of supply and demand that would usually predict less output for less reward - but workable if there is less effort, and less time, employed per unit billed.

Speedy, economically productive medicine allows payors, health care managers, and legislators to claim to subscribers, shareholders, or voters that health care can be continually had at a lower cost, and that they are able to perform this miracle of bounty, sometimes for large bonuses. In the Enron/Andersen era, with lavish entertainer, sports, and CEO salaries, and with pressure for health care as a right (which means cheap), and with the increasing legitimacy of lotteries and gambling, perhaps doctors can be excused for thinking that the general public concurs with an 80/20 approach (sk87). A belief that higher work speed can be had without loss of quality may be necessary for those producing, or participating, in reimbursement systems that offer less and less.

Relative anonymity also makes radiology susceptible to a bias for speed, since cutting time per case is not likely to be observed by peers or patients. Invisible to patients, few watch our work. The technologists who do watch have themselves a bias for speed. Paid per procedure, and with minimal if any patient contact, there are fewer speed bumps in radiology than in the other fields of medicine where the crude measure of time given by the doctor is visible to the patient, who then has at least some measure of effort expended, if not of competence.

Biases For Normal, Bias for Speed

Biases for normal reports are closely linked to the bias for speed, each producing, and resulting, from the other. Speedy reports are most often normal, and normal reports are speedy, and at least inspire confidence (sk231) - tempting the hopeful to confuse confidence with correctness, and tempting the productivity-minded to believe that less time spent on a case might somehow confer accuracy, ignoring the fact that most errors in radiology are false, confident, negatives.

Normal allows a faster procedure, faster film review and interpretation, faster dictation, faster transcription, and faster correction time. Normal is faster and simpler for the referring doctor, or office staff, to understand. Normal offers quick phone reports, with good feelings all around. Normal doesn't require the added effort of specifying exactly how, why, and what - sometimes veiling poor observational disciplines under the descriptive anonymity of 'otherwise normal'. Normal, even when wrong, has the virtue of appearing decisive (sk231), and avoids the agony of open uncertainty (sk58). Normal, at the initial and most critical fork of diagnosis, is probably one of the most powerful feeling diagnoses (sk66). Normal offers Dr. Pangloss' "best of all possible worlds" (sk202), and avoids the pitfalls of differential diagnosis, which by its use casts a pall on the ability of physicians to come up with answers, rather than more questions (sk229p601,sk231).

Searching for abnormal can be unrewarding, since predominantly normal case material infrequently rewards the thorough searcher (sk16). The low index of suspicion that goes with the bias for normal is understandable (sk48), but can result in underdiagnosis (sk37, sk48,sk79p407,sk14p365,sk29), operating like emotional factors in the stock market, with bubbles based on hopes, - gambling for the best of all possible outcomes, rather than sticking to the fundamentals - hoping for bull's eyes, rather than just doing the archery right.

Meaningful normal only follows the real accomplishment of knowing abnormality, and having looked for it, failing to find it. While HMO market planners may seek populations with a high percentage of normal - providing health care to well patients who don't really need medical skills and services - such a view is alien to ethical medicine. Normal ought to be the patient's good fortune, not ours.

The bias for brevity in perception and reporting might be conflated with confidence, and from there conflated with accuracy (sk231), but this sequence works best with easy cases, and holds most reliably if one starts it in reverse, with accuracy, and then works on to brevity. Confident dead wrong normal reports - the false negatives - are the most frequent radiologic error. While difficult incorrect readings may take longer than easy correct ones, with long reports reflecting more uncertainty, this may reflect the complexity of case material, rather than being a comment on the intrinsic value of brevity (sk231p262). Believe the brief, believe the negative model in double reading is actually worse than individual readings (sk232). Uncertainty about an assessment is not less information, but more information, and is what the proper understanding of error and variance estimates are about.

It is in the last 20% of cases - the hard ones - that departing from the fast and facile requires a departure from the brevity and the 80/20 rule that works best only for the easy ones. The difficult last 20%, the difficult cases, necessarily involve more effort, and longer reports, than the do the easy 80%.

Dreams of Flash Diagnoses Promote Speed, and Promote 'Normal' Diagnoses

Literature reporting flash diagnoses, with abnormalities detectable in a fraction of a second, probably encourage speed, and a bias for normal, and discourage the long look (sk18,sk56,sk64).

Christensen (sk103,sk14) found that 60 percent of all radiographic findings important to management are discovered in the first 15 seconds. A 250 millisecond look at a chest can identify major pathologic features, even peripherally (sk103), detecting 70 percent of obvious tumors, and 30 percent of subtle ones (sk127). Conspicuous abnormalities can even be detected by non foveal vision (sk112). Flash viewing for only 200msec (sk106p1246) can yield 70 percent true positives although viewing for an unlimited, and unspecified, time yielded 95% true positives, or better.

Obviously more time improves matters. Unlimited viewing of an unspecified length increases detection of lung cancers (sk127). Flash diagnosis is not the same as a detailed search for tough targets (sk79p404). The easy abnormalities can be diagnosed in seconds (sk14), but too many misses go unexplained if flash radiology were really that good, and the retrospective miss rate of 30%, obtained sometimes even with the added benefit of context bias (sk48), wouldn't be as high as it is.

Clearly, spending too much time is not a problem from the standpoint of getting too accurate. Several authors (sk127,sk14) find that radiographs should not be " speed " read, a reading rate which requires excessive use of the imagination and self-deception (sk79). Though obvious findings are evident with rapid viewing (sk66), radiologists cannot be content with being masters of the obvious.

False Positive Fears Feed a Bias for Normal, and for Speed

An understandable fear of false positives also can cause a bias for normal reports, and thereby also encourage false negatives.

False positives trigger anxiety, expense (sk132,sk58,sk200), and the medical risks of unnecessary procedures, and avoiding false positives observes the dictum of primum non nocere - first do no harm. Workups for false positives trigger that uniquely foolish feeling that comes when exertions cause more trouble than doing nothing would have. False positives, by distracting from the real cause of disease, can delay the diagnosis of the true cause of patient's complaints.

Past hard lessons with incorrect borderline positive calls can fuel skillful rationalization (sk223p42), probably transforming some subtly abnormal cases into normals, and again skewing mental set, interpretive sensitivity, and personal ROC curves to avoid wasting time looking for abnormalities that often are not there. Working hard at finding subtleties in a population of mostly normals is a behavior that can be readily extinguished. The false zebra phenomenon - extinguishing of the ability to find the subtly abnormal in mostly normal populations - is probably the reverse of context bias (sk48).

False Positive Errors are Harder to Defend than are False Negatives.

With a high followup rate, false positives are generally a much more public error than are false negatives, since they often prompt immediate, embarrassing followup - building notoriety, and low credibility. False negative reports have no finding that demands followup, so that one can make false negative errors and still appear mostly right.

The fear of false positives can produce the confusing accusation that one is "seeing things", paradoxical when used in a field which is supposed to be about about seeing things. Samuel Shem, in his book The House of God (sk228), invokes the shame of false positives (sk216p498) to chastise in law XII those house staff and radiology residents prone to conjure up abnormalities.

While radiologists might claim that false negative errors are more important than false positives (sk125), radiologists' are uncomfortable with false positives (sk125,sk200). The use of language is telling. Interpretive styles that result in more false negatives while producing fewer false positives have been termed "cautious" (sk112), while styles that produce more false positives, with fewer false negatives have been termed "liberal". One might think that the term 'cautious' would be applied to interpretive methods that attempt to insure that no significant lesion is missed - addressing the patient's interest, rather than applied to approaches that have the effect of shielding the radiologist from the embarrasment of prompt followup on false alarms.

Because perception is incompletely understood, false negatives are probably more likely to be excused than are false positives, since once a finding has reached the perceptual level, a mistaken analysis of it seems more obviously wrongheaded (sk39). A false negative on a tough case is rarely faulted, partly from "there but for the grace of God go I", and partly because criticism seems a hard thing to add to a patient's or radiologist's misfortune.

False negatives that do progress to become obvious positives still allow one the "apples and oranges" defense that negates criticism - of course it's easier to see lesions (retrospectively) after they've progressed, which is not comparable to the situation when an earlier phase (prospective) false normal occurred (sk2).

Difficult false negatives, for which veracity might demand equivocation and uncertainty, when left to progress to a later stage of disease, eventually become clear and definitive true positives, and then offer the radiologist the good feeling of being definitive, correct, helpful, and competent. Pointing out the late stage, definite apple core lesion in the sigmoid colon, feels more helpful and more competent than risking being wrong, and agonizing over an earlier, tougher call a few years prior, that, though tougher on the radiologist, might have saved patient morbidity, and offered efficacy - or, that might have prompted misguided and complicated workups might have been a normal patient.

When the criteria for distinguishing subtle abnormal from normal are sufficiently imprecise, and with many normals in the population, the radiologist may be persuaded in favor of false negative rather than false positive errors because of a powerful multiplier effect - since two or more false positives due to a very sensitive diagnostic threshold may seem a poor trade compared to making one false negative (sk232). For breast imaging, some studies have suggested that a ratio of false positives per additional cancer detected using double reading might range from 45:1 to 101:1 (sk255).

But False Positives Aren't Always that Bad

Counterbalancing such high ratios of false positives, relatively benign followup imaging procedures might minimize the adverse impact of risking numerically greater false positives, while confirmation and early therapeutic measures on what would have been false negatives might enhance overall outcome results. Accepting false negatives, and fearing false positives, made more sense in the past when the next level of investigation was not cross sectional imaging, or skinny needle biopsy, but exploratory surgery, with its attendant complications. With better treatments, with early detection important to prognosis (sk134), and with corroborating non invasive studies of lower morbidity, the cost of missing subtle lesions becomes greater than it used to be, and the cost of false positives is lessened (sk111). When false negative errors are extremely injurious, false positives may be more acceptable (sk232), and fear of false positives becomes hard to justify when false negatives, not false positives, are the more frequent, and the more significant, error (sk79p414,sk233p66).

Though some (sk60,sk23,sk39) find that false positives become more likely as more time is spent on a case, rather than being a sign of incompetence, false positives may indicate a higher level of engagement with the task. Siegle et al (sk16) report that radiologists whose errors included a relatively high proportion of false positive findings were those who made fewer total errors. Garland (sk63) also found that readers who accepted more false positives, and read with the most pessimistic, and sensitive, perceptual set performed almost as well as a double reading , with the cost in false positives relatively minimal, missing fewer true positives than those who employed a less sensitive threshold for abnormality.

With the more intensive followup and with the learning that potentially occurs when interpretive methods are willing to risk false positives, patient injury from false negatives might be decreased (sk232), and the cost in false positives might decrease over time as feedback and improvement occurs. False positives might be learned from rather than feared.

With a low threshold for possible abnormality, the radiologist has the option to consider a finding, and then choose between abnormal or normal variation or technical difference. Not identifying a subtle finding precludes considering it at all.

Having Mostly Normal Patients Encourages a Bias for Normal, and for Speed

The high percentage of normals in a routine practice and in prospective studies plays to the error of labelling as normal the subtly abnormal, and promotes a normal bias, since on a statistical basis calling it normal is usually correct (sk16). This is somewhat similar to the manner in which there is no immediate penalty for sloppiness in military operations in areas where one is unopposed (sk9p14-19). Many normals dilutes the higher error rate of 30% found in retrospective studies that are usually enriched in abnormal cases (sk2,sk16). With the diluting, and deluding, effect of many normals (sk140), the greater proportion of errors made on tough cases gets watered down from 30% to a more reassuring and respectable 5% - an error rate applicable mostly to patients who are mostly normal, and who aren't most in need of radiologic skill.

Such a 'prospective' measure of accuracy - 5% error - derived from mostly normal populations, overstates the accuracy applicable to dealing with the cases that become problematic (sk16) - the ones with subtle abnormalities that experts are expected to be good at (sk14,sk48,sk140).

Finding abnormality is a valid measure of skill. Calling it normal properly comes only after that. Adjusting one's perception level, ROC curve, or reading sensitivity, when knowing beforehand which films are really normal, would be scientific and economic efficiency, but an uninteresting accomplishment. Without the luxury of knowing beforehand which films are the normal ones, prospectively lowering standards of film review and perception, even in a mostly normal population, is less in the tradition of science, and more in the tradition of gambling.

Prospective Population, Retrospective Performance, and Accuracy

Probably neither the usual prospective nor retrospective studies give the best picture of radiologic accuracy in practice.

Retrospective studies make perception look easier than it usually is, with more false positives, with underestimation of false negatives and underestimating the difficulty of finding subtle lesions in routine practice (sk138,sk135,sk132) - always having the benefit of knowing that there is a missed, or likely, lesion (sk2) - and having the benefit of context bias (sk48,sk135).

The retrospective look's value is that it shows the highest potential level of perceptive functioning that a radiologist can attain (sk138,sk48), even if unable to maintain it in the 80/20 routine reality of the workplace. Like those standardized tests that don't penalize wrong answers, retrospective studies allow radiologists to stretch their talents on tougher cases, in ways that are usually inhibited by the penalties that come with false positives. That abnormalities falsely called normal are shown to garner unconscious radiologist attention (sk125) indicates that there is some barrier to radiologists bringing to bear the perceptual resources they already have. Retrospective studies show that radiologists already possess, and have access to, perceptual powers that go unutilized, but that might potentially be tapped to enhance routine work.

Perhaps the best measure would be to eliminate spectrum and context bias (sk48) by giving the reader the routine practice mix of a high percentage of normals, and then measure what fraction of abnormals were correctly identified. This would count what counts most - the ability to identify abnormal even when it does not jump up and announce itself among the normals (sk94,sk31p217).

Perceptual error might be minimized if the observational disciplines employed by the retrospective interpreter of known misses could become so habitual that they would be rapid enough to allow their application to routine work. Maybe this would require more specialization, or better histories. Maybe with computer generated prompts specific to present and past patient history brought up on the interpretive screen - the heads up (sk263) display of jet fighter aircraft transposed to a medical setting - technology might repay us for any speed related losses in attention span. Then again, maybe that would be just another distraction (sk39p333).

Speed vs. Time On Task

What approaches are there to put more effective radiologic time on task - but in less time? Of course, attorneys are a counterbalance to the indiscriminate use of speed, but this seems dubious, like balancing devils to achieve virtue.

Different work organization and improved work conditions (sk90) might increase efficiency, making more time available for the depth of perceptive tasks. Radiologists might obtain more view boxes (sk64), making comparison readings easier, and minimizing the impact of interruptions.

Physical isolation of the radiologist might eliminate distractions (sk64)that interfere with perception (sk84,sk26), enabling better attention (sk49p95,sk49pp97,99). This would avoid working in the radiologic equivalent of deer hunting with buddies who every 3 minutes want to show you pictures of their kids, while discussing the Bears game on their walkie talkies, as your own cell phone goes off every 4 minutes, and the sun is going down right behind the scent rag where you thought you saw a 12 point buck.

Subspecialization Uses Time/Experience Accumulated in the Past to Minimize Time Spent in the Present and Future.

More subspecialization might give a reader enough career time dealing with a limited set of problems to develop exceptional perceptive competencies and search disciplines, and yet retain the efficiencies of useful speed, so that the time taken per case is only 30 seconds - and 30 years. Multiple studies find that subspecialist readings are superior.

Loughrey et al (sk240) found that oncology specialist radiology review of outside cross sectional imaging changed staging in 19 percent of cases, with major interpretational differences in 34% of exams. The differences had little impact on patient management (sk240p154) perhaps because the films were reviewed after treatment decisions had already been made on the basis of the outside report.

In the military setting, skills have been described as diminishing rapidly in a matter of months (sk130p218), and with rotation of duties (sk9p16) , perhaps explaining the paradox that resident performance has at times been found superior to that of staff radiologists (sk233) due to a beginner's mind effect, or due to a (sk233p66) more recent intense period of engagement in the imaging area that was tested. Subspecialist immersion in a limited breadth of imaging studies allows good habits and lessons to build more quickly on one another.

In another study, only half or radiologists evaluated were 100% reliable in excluding intracranial hemorrhage at CT, with a subset of more capable radiologists consistently scoring better than others (sk224,sk225).

Some studies suggest that neither physicians nor the public (sk87) believe that shifting procedures and patients to high-volume centers would effectively diminish errors, making less likely a shift to superspecialization.

Centralized high volume imaging centers have already reported favorable quality assurance experience (sk211,sk221), and such models, especially with superspecialization, might offer low cost, lower malpractice exposure, high quality readings, and offer radiologists sustainable CME demands, sustainable income, and sustainable and more flexible career development and lifestyle.

Optimally, one might imagine a large group of radiologists reading both accurate and fast from their megacenter in Bora Bora, with digital images coming in from distant locations sent to highly accurate and fast superspecialists reading only highly specific categories of images - Knee MRI, head and neck MRI, head CTs, etc.

Early subspecialization, with a whole career - not 5 years - to learn all modalities, might yield a shorter time to be economically productive for a limited range of studies, with limited range of malpractice probability and exposure. This might allow young radiologists, after anatomy/modality specific qualifying exams, to gradually increase the number and types of exams they are responsible for and qualified to read, giving them early debt relief, and giving radiology a series of small steps along a more gradual career path - similar to that of business executives - rather than the abruptly increased status/ income steps / professional demands that physicians encounter with the present (dis?)organization of medicine. A subspecialized and long incremental learning curve might even sidestep wasting time learning methods that become extinct at a faster rate.

Cheaper labor (sk255), such as nurse subspecialists, technologists (sk89:p942), or Masters in anatomy (sk102), using effective CAD programs that disconcertingly seem to work best when the radiologist is not involved (sk141), might do the perceptive work for a prescribed range of cases, if radiologist time is too precious(sk79,sk85).

Perhaps anticipating resistance to giving up speed, Oestmann suggests the solution of "enforced" viewing times might be necessary to improve the detection rate of lung cancers (sk127).

Double Readings, Time On Task, and Speed

Double readings do increase accuracy, slowness, and time per case (sk119,sk78,sk112,sk88p286), but the generally modest increase in sensitivity and accuracy (sk119) is no surprise, and may not be cost effective (sk8, sk35). Sometimes double reading may add nothing but expense (sk255). Two readers, especially if using time truncation technique, will each ensure that the most obvious abnormalities are not overlooked, and thus duplicate effort, adding only small improvements attributable to idiosyncratic variations in search habits and discipline, perhaps mostly only repeating search habits which were in some way lacking to begin with. Some studies reporting accuracy enhancement by double reading may be making the case more for subspecialist reading, than for pure and simple double reading (sk2,sk41).

Double reading offers one of the few instances where quantity and quality can be compared on the same scale. The additional FTEs required, as added over-readers, to attain a given level of performance over and above the performance of a solitary reader, can be expressed as the number of single read cases that could otherwise be handled by the same number of FTEs. For example, when both quality and quantity are factored in, it may be that two fast, but average accuracy interpreters may be required to match the same quality and quantity output as one highly accurate, but slower interpreter.

Various studies have attempted to find ways to maximize the yield from double reading, and to calculate increase in true and false positives, the costs involved in eliminating such errors, the cost of error free reports (sk232), and the effectiveness (sk234) and substantial time costs of various ways of combining double readings (sk230p49). The variability in the ROC curves for individual radiologists make it difficult to predict the yield of double reading in a given practice (sk255).

One low cost approach to cost effective double reading proposes the same reader search first without, and then with, clinical history (sk79,sk112). Lack of efficiency enhancers like automated viewers probably makes double reading untenable (sk234). There are conflicting opinions as to whether double reading is economically feasible (sk2,sk88,sk102,sk119, sk89p942) or efficient. I suppose it depends on how much you're getting paid (sk234).

Computer Assisted Diagnsos (CAD), and Speed

Computer assisted diagnosis (CAD) could help, offering a double reading that complements (sk35,sk55), rather than duplicates, radiologists' search strategy (sk35), increasing accuracy, partly through technology, but maybe also by simply requiring more time interacting with the image. By encouraging the radiologist to give more time per case, preclude errors, and teach the reader about his/her own perceptual errors, CAD might enhance perceptive skills.

The use of a sophisticated technological device could make the act of spending more time with an image into an act of power, rather than an act of nervous uncertainty, or embarrassed confession of potential fallibility. The use of a sacred gadget, with its own billing code, might confer special powers on the radiologist, and exert a mind altering influence on payors, who then would allow the radiologist to get paid more for looking and doing more. Coming up with cash for CAD might be a sacrificial act that communicates to payors that we actually do intend to spend more time on these cases, since the extra work becomes tangible, transsubstantiated into a device. This is quantifiable - maybe even billable - since the ritual CAD object makes tangible the difference between cursory and complete.

But maybe CAD won't make things much better. The cost/ benefit of more observers may outweigh the cost/benefit of buying, adapting to, and depending on, new technology (sk119,sk112). Technology doesn't solve, but only displaces, the problem of perceptual error to a new and different technology (sk2), offering the opportunity to make a whole new, and maybe longer, list of mistakes. You can't buy excellence in a box, though you can keep buying newer, and more expensive boxes. Tellingly, a productivity enhancing voice transcription program first printed my dictated 'perceptual error' as 'perpetual error' - and of course my spell checker missed it.

To the extent that radiologists might attempt, with the availability of CAD, to spend less time looking at films, CAD has the potential for diminishing non CAD time-on-task, with atrophy of innate skills, and diminishing perceptive performance. Such a tendency to lean on the machine has been described (sk35,sk39), increasing the chance that gains in perceptive power are more likely to be used to increase productivity by decreasing time spent per unit billed, rather than increasing accuracy and quality - essentially, 80/20 time truncation. In this way, radiologists might use CAD to spend less time with the image, not unlike the clinician who uses the CT first, and then sees the patient, transmogrifying his/her own need for rapid clinical case throughput, with less time on task, into more (sk198p57,sk271), and some would say too much (sk54,sk3), technology and imaging being done (sk22p93). Such things have been anticipated. Dreyfus (sk6c,p27) quotes Phil Agre's book Telematics and Informatics, cautioning against large expenditures on high tech equipment that is then limited by, and cannot expect to overcome, fundamental flaws in the purchasing institution.

CT - A Speed Trap?

Reviews of CT interpretations might serve as a marker, or speed trap, for a more generalized use of speed in image interpretation involving other modalities. Some authors have found that satisfaction of search errors were particularly common at CT (sk138).

Since optimal CT interpretation involves a close evaluation of an information-rich, extensive number of images and structures (sk138), time truncation technique might be especially easy to spot at CT, where excellent spatial and contrast resolution eliminate the confusing shadows (sk27,sk107) that generally hamper plain film/projection image diagnosis, and blur the difference between lesions not looked at and lesions not seen. With CT, perception is more straightforward, misses more obvious, and defects in search discipline more blatant. Failure to mention a structure becomes a clue to failure to look at all. Worse, findings missed at CT may occur in regions different from those raised by plain film findings that prompted the CT (sk138), all this tending to mislead the cursory radiologist.

With CT there is a more constant appearance of target and background, unlike the perceptual challenge of magnetic resonance where normal target or background appearances vary depending on which of the 67 flavors of sequence parameter are in use.

Since the potential for such false negative errors increases with disease prevalence (sk233p64, sk16,sk20,sk231p261), and with enhanced imaging detectability of disease, CT's ability to show otherwise undetectable incidental disease increases the probability for speed related perceptual misses (sk16). At CT, attempts at flash diagnosis become more blatant, being long on describing the obvious, missing subtleties, and short on enumerating normal structures, pertinent negatives, slice numbers, or visually inconspicuous anatomy like the adrenals and the spine (sk240,sk237), prompting calls from referrers - "Hey, did you look at the adrenals?"

The seductions of speed inherent in fast machines is suggested by a study (sk133) evaluating automatic cine viewing of CT images with frame rates up to 21 frames per second, equipment that invites the use of speed. With equipment such as this, the increased frame rate on the dial may exert a subtle pressure, not entirely dissimilar from the higher voltage levels on the equipment in the Milgram (sk143) experiment - always asking for more, faster, or both. Predictably, results favored a human pace that varied with interpretive flow, rather than a machine determined rate of viewing - this with the perceptive task limited to evaluation of abdominal masses that were mostly located in those solid organs that especially garner radiologist attention. Had abnormalities included been of a vascular nature, or in body wall, muscular, skeletal, or intestinal structures - generally more difficult to evaluate - interpretive times would likely have been longer, or performance worse than the minimal AZ of .86 obtained.

Despite the perceptual power that CT technology offers, the weak link of human perceptual error persists, even with CT (sk132). Satisfaction of search, or the two bomb fallacy, also occurs in CT (sk138), with 40 percent of the significant, missed lesions accompanied by other distracting findings. Even with the benefits of retrospective bias, review of CTs with known missed tumors identified only 47 percent of missed cancers (sk138). The tendency to miss paravascular and bronchial lesions in the chest (sk132,sk138) might be explained by omitting or deferring the painstaking task of linearly following the margins of multiple vessels, looking for aberrations - a task more laborious than merely looking for nodules obviously out of place in the low noise lung fields. This is similar to the task of evaluating the ribs, needing to looking along the entire length of each surface of each rib, one of the more exacting tasks in radiology (sk64). The slow phase perception that Christensen (sk14) et al found was needed for the peripheral chest and upper abdomen reflects the need for looking along many structures, evaluating margins and interfaces, not merely looking for substantial density changes that are the clue to easier, more readily detected lung masses.

The tendency of overlooked lesions to occur in the lower chest (sk138) may have resulted from fatigue from already looking at the upper chest (Sk138p112), but then one would expect that the rest of any radiologist's day's work would be all downhill after the first 40 CT slices - which would make for a very short span of accurate reading ability. Conceivably, the risk of many false positives due to lesser air contrast offered by atalectatic lung can explain this, or perhaps lower chest CT errors are merely due to search time truncation. The lesser parenchymal volume of the costophrenic angles may be thought statistically less likely to harbor lesions, and this may encourage an 80/20 " don't sweat the small stuff " interpretive strategy - especially so if the signal of tumor cannot be distinguished from the noise of atalectasis.

The Missing Body Part - Speed Trap II?

Another test of the rigor with which interpretation is done is the test of the missing body part. While even cursory examinations can detect many abnormalities, normal structures that aren't present are more likely missed by cursory observational discipline (sk94). Failure to identify missing legs on bone scans, or surgical abscence of the female breast on chest radiographs (sk214) are clues to a deficit in the perceptual inventory maintained, and in the standards that the radiologist employs.


Verbalizing a search strategy, as in structured reports(sk236p1517), seems to improve vigilance, and lesion detection (sk17,sk47) and protects against satisfaction of search (sk79p416,sk237), perhaps by making explicit any ommissions or lapses in diagnostic rigor, and thus forcing the slowdown that creates time for attending to the film (sk237) - and time for perception. A checklist of specific queries to be addressed by each exam type helps sharpen perception (sk88p288).

Using report structure to break the work into smaller valued sub-tasks promotes giving proper time to each sub-task - which - set off from the others - is respected as an essential part of the whole act. This focusing on each element of the task becomes much like Herrigel's archery (sk15,sk15b), or like prayer beads employed by various religions to encourage and promote slowness, consideration, and deliberation.

Lesgold (sk223p57) makes a similar observation, suggesting that a verbal plan for a search - a form of controlled conscious knowledge - acts as a conscious structuring agent for the sub-tasks whose execution is often automatic and under less conscious control.

While some observers failed to find any significant advantage in the use of a structuring checklist (sk230), there are several reasons why the accuracy enhancing effect of a checklist might not be observed. A checklist might increase the sense of being hurried in a reader, prompting less rather than more attention to each subtask. Readers who already have an effective and comprehensive method of film interpretation might find an external checklist to be distracting rather than helpful. The use of a checklist might instill false confidence, rather than humility in the reader, who becomes too easily satisfied that all the boxes, if checked off, insures infallibility - similar to White's description of the observational deficits of the tenderfoot deer hunter.

In studies evaluating a limited number of known diseases in a limited number of organs (sk230), the value of a checklist is minimized, since a checklist is most valuable when a large number of unknown and diverse disease processes are sought. Subspecialist radiologists, especially if they already have excellent ROC curves (sk230), are less likely to benefit as much by interpretive enhancements than are those radiologists farther down on the performance scale. Finally, learning effects occurring due to multiple interpretive enhancements being tested might improve performance of all enhancements tested, such that distinguishing which of these was best would be difficult, since learning effects can push the performance enhancement of each method into the same small corner of high performance where they become near-equivalent, particularly if pre-learning performance, as measured by ROC curves, was already excellent.

While the use of checklists in structured reports has the potential to privilege accuracy of finding one type of abnormality compared to others - as a function of order in the checklist (sk210) - the breaking up of the total task into subunits might minimize some of the error potential that occurs when multiple, rather than one abnormality must be looked for (sk210).

Structured reports, listing findings as they are found, keeps track of what has been evaluated (sk79p415,sk47), and can be helpful in promoting a complete look (sk240,sk237), since perception of abnormalities is impaired by interrupting search (sk47). Structured reports function like random access computer files, making it easier, when interrupted, to find your place, and resume in sequence. Structured reports can minimize the effect of distractions and the too speedy use of automatic processing (sk52) that creates errors.

A structured report, using explicit comment on each anatomic area evaluated (sk204), encourages a corrective cognitive dissonance when speed prompts haste and omissions. Structured reports, a bare outline of the visual search, lend deliberateness, by removing the fig leaf of vagueness that might otherwise lump areas not specifically, or much looked at, into the anonymity of "otherwise normal".

Verbalizing a report structure, as in a structured report, does not detract from being able to attend to the film, since speech does not interfere with the visual system (sk17). Whether this applies to voice transcribed reports, which increase potential for visual fatigue by dividing the radiologist's attention between visual screens,is uncertain (sk39). Structured reporting make radiology reports easier to dictate, transcribe, read, and compare from one exam to the other, even with different interpreters.

Structured reporting, with an explicit checklist of sub-tasks (sk47), working from a well thought out plan, might build the deliberate observational habits and completeness (sk79) that could minimize the tendency to search time truncation (sk237), and make excellence in radiology a routine accomplishment.

The Implicit Audience Effect may be Underemphasized in Medical Imaging

When a verbalized structured protocol removes vagueness, the reader is more likely to engage the image as if providing service to a person, as if implicitly visible to an audience of observers, and therefore within the kind of disciplined social environment that promotes accountability - doing radiology as if someone were watching you do it.

The power of the implicit audience to enhance accuracy has not been emphasized in imaging literature, but the accuracy enhancing effects of explicit structured reports, verbalized reporting, of double reading, and of clinical history may rely to some degree on their creating an implicit audience. An implicit audience effect (sk143), by removing anonymity, may create a sense of accountability (sk120), that other known aids to sharper perception , such as the retrospectoscope, also rely on.

While studies have shown that interpretive accuracy is enhanced by pooling interpretive skills in double reading (sk203), this does not seem to have been distinguished from the potential enhancing effect of merely knowing that someone else was also looking at the same images. For example, the second reader in double reading is an implicit audience, so that each radiologist is in a less isolated, or less anonymous social environment, perhaps reducing perceptual errors (sk58). Dreyfus' (sk6c-p38) observes that the presence of others, and an emotional connection to the work, becomes important to learning and optimal performance. Perhaps part of the improvement from double reading is not just the added skills of a second interpreter (sk203), but the inducement to better performance on the part of each interpreter. It might be interesting to see if double reading, performed with each interpreter aware of the other's role, is more accurate than a composite double reading performed by two independent readers unaware of each other's contribution.

Clinical history, creating an implied expectation and question, likewise represents an implicit audience, improving perception of anatomy and lesions pertinent to the clinical problem (sk61,sk47,sk17,sk62,sk111,sk110), and inhibiting satisfaction of search errors (sk61) for that type of finding. By explicitly stating what is to be searched for, and demanding attention to specific items, a clinical question removes vagueness (sk47) about what is expected, promoting accountability and the sense of involvement (sk6c) - needed for looking as if looking counted. Accordingly, fewer external cues to interpretive task decreased lung nodule detection rates to less than half (sk47).


Even if good methods to improve perceptual accuracy are found, there might be obstacles to implementation. The reality of medical and radiological error seems to persist unchanged over decades (sk2,sk102,sk16,sk27,sk119,sk40,sk137), impervious to improvement, similar to findings with the NASA shuttle program (sk197). When education is used to remedy some errors, readers may then obstinately go on to make new and different errors, or take up making errors that they had in the past gotten beyond (sk84).

Forgive and Remember: Managing Medical Failure (sk5) by Charles Bosk, may offer some explanation of persistent medical error, identifying an error class impervious to the benefits of education and knowledge, probably similar to what Hilfiker refers to as the "failure of will" (sk271p79). Bosk divides medical errors into technical and normative categories. with technical errors ascribed to lack of knowledge, and therefore remediable. Normative error, from (sk246,sk5-p51,57, 60,64) failure to apply basic knowledge and skill, is generally impervious to improvement by education, since normative error is rooted in the value system and character structure of the physician, recalling Smith's past (sk49p96) observations on how basic life perspective impacts radiology errors. Normative error can negate any amount of training and education.

Some radiology studies confirm similar effects. Studies where less experienced or less trained (sk233,sk245,sk225) readers have superior performance suggest that maybe those with more experience were somehow not making best use of their experience, with performance more dependent on personality variables or other traits or characteristics (sk233).

Some perceptive errors may be normative errors, persisting unchanged, since the level of rigor and care applied to medical problem solving is obstinately operator dependent, as noted by Drs. Ania, Asenjo, and Heilman (sk54,sk3). Goodall's high school students had the technical equipment and know-how to use their binoculars, but committed the normative error of not bothering to use them (sk195). Lack of comparison with old films, held to be the standard of care (sk49pp86-88,sk53, sk24,sk118,sk201), is one example of failure to observe established quality standards. Failure to perform comparison reports, a common failing, and an important cause of missed cancers (sk37), would - according to Bosk - fall into the normative error category. Other articles describe how breaches of standards of care (sk240) - essentially normative errors - risk and result in more litigation (sk238). This is a general societal trend, since similar flaws in insuring the safety of shuttle crews have been identified in the space program, where an implicit responsibility to take due care for the safety of others would be expected (sk244).

The 80/20 approach inhibits getting too proficient (sk233), and is probably a major contributor to performance failure, evident even to ancient craftsmen with a vocational level of training (sk209). Many problems, and their solutions, in radiology call for levels of performance that are adequate to handle even the difficult cases (sk231,sk233), at the 99/1 level or better (sk232p592,sk225,sk240,sk16,sk209,sk29pp152,154,168), as noted in the problem of screening for breast cancer (sk255p892) or detecting intracranial hemorrhage (sk225).

When the goal is to find the single abnormal among 100 cases, calling them all normal produces correct calls in 99% of cases, but produces 100% failure in the task for which the entire 100 were examined/radiated - finding the one in 100. With such performance, one might as well not image them at all, and still miss one in 100, without the benefit of radiography, which - costing nothing - would be more cost effective, though perhaps not revenue effective.

It is in the setting of reading many normals - where calling it normal produces few errors (sk232) - that normative error has the greatest potential to produce negative effects. If the pretest probability of normal is high, will the doctor maintain the best standards of practice and care needed to detect the one in 100 subtle findings needing to be found, or relax perceptive standards, and congratulate him/herself on the 99 (sk16,sk225,sk209) of 100 easy and confident calls that the population and providence provide?


Radiologists, like other physicians, are surpised at the magnitude of their own error (sk58,sk63,sk90,sk49,sk19). Coexisting with technologies having a stunning potential for diagnostic accuracy is a disappointingly high error rate in routine work. If personal experience weren't enough, studies report that most radiologists eventually miss significant abnormalities, sometimes below an acceptable standard of care (sk100,sk135,sk58,sk64,sk16,sk19,sk20).

Discussing Errors is Depressing

In older literature, critics were occasionally candid about their reservations concerning error review (sk58), fearing negative effects. Reluctance to believe that error is common, may reflect (sk58) physician reluctance to accept the uncertainty inherent in the probabilistic nature of radiology and medicine. Imaging ordered in hopes of attaining black and white clarity and certainty becomes depressing when the results are in shades of gray (sk100,sk58p36).

Error may persist unchanged because it is so depressing to review one's own errors (sk58) - probably because it prompts the feeling of futility that comes when the past cannot be rescued by getting smarter too late in the future. Working on error probably elicits in physicians the same dread and sense of waste that going to a physician elicits in patients - "I'm going to spend all this time and money, just to be OK - for all this effort I would rather have something new and good in my life." Looking at your own errors is depressing because it means more work to just stay in place. By robbing time from efforts at actually learning new tasks, time given to error review feels like being stuck in the past, rather than moving ahead. In this way, continuous quality improvement (CQI) brings to mind other safety/maintenance practices like exercycles, physical therapy, weeding, edging or fertilizing the lawn, vaccinations, antabuse, new roofs, walking the dog, neckties, condoms, and taxes.

Discussing Errors Feels Hazardous

Similarly, physicians do not welcome appraisal of their work (sk78,sk72). 86% of physicians believe that hospital reports of error should be kept confidential (sk86p1966), opposed to 62% of lay people who think such results should be public. Whistleblowers and critics, even in the medical field, have found that institutions generally have a low tolerance for bad news, conflict, complaint, or candor (sk89,sk83,sk121,sk122,sk123,sk124,sk129,sk257).

Radiologists are often held to a high standard (sk131), in part because retrospective film review is easily done (sk49, sk119p1086), making errors more salient. The view that individual perceptual errors might not constitute malpractice (sk19, sk100) seems hard to reconcile with the reality that perception applied to individual cases is at the heart of the service that medical imaging provides. If not evident in a given case, but only in a population (sk19,sk90), malpractice might conceivably be defined away into practical non-existence, leaving the post modern conundrum that without the possibility of mal-practice, there probably cannot be a meaningful good practice.

Discussing Errors Meets Impediments

There are also natural impediments to improvement. In radiology, unlike jig saw puzzles, or text reading, it is not always immediately evident that our diagnostic verdicts are correct, or incorrect. In radiology, immediate feedback is rare, and even delayed feedback is difficult to obtain (sk71), often limited by inability to get pertinent pathological or surgical confirmation. (sk119,sk21).

Some studies of radiologic accuracy prefer to ascribe differing performance results to methods of technique employed, preferring to ignore the contribution of observer error (sk119pp1088,1090).

Deliberate avoidance of significant terms like mistake or error (sk41,sk219) nurtures a climate of denial, even though legitimately protecting some practitioners from overzealous judgements, and shielding the flagrant errors of others. This type of strategy seems to be successful, since (sk90) a search of the medical literature for articles including " error " or " mistakes in radiology " is reported to produce few papers. Similarly, widespread news media hype of medical advances can produce the unduly high patient expectations of medical care (sk239) that, when disappointed, lead to litigation, although investigators seeking grants and funding, and those providing the funding, probably encourage similarly high hopes (sk29).

Resistance to error review has prompted forces outside radiology to demand reviews (sk16), or to appeal Medical Board dismissals of charges against radiologists (sk80). Radiologists are said to show reservations, and lack of enthusiasm about continuous quality improvemnt (CQI) (sk72, sk25). Similar exasperation with the counterproductive energy draining character of multilayered oversight institutions is observed in other fields as well (sk268pp203-204). Outcome studies and quality reviews are sometimes valued little (sk83), most often being a lot of work to tell us things that we knew all along, and don't have the required ability or resources to change (sk89p944). In a field squeezed by funding cuts, we may not have the time to look carefully at our misses, needing to work hard and fast simply to generate them. Also, errors are outliers, not the dominant trend, and so seem more excusable, as exceptions.

Some studies report that even the public seems not much more concerned about medical errors than physicians are (sk120,sk87), although perhaps this reflects the public's lack of awareness of the extent of error, and an abscence of effective mechanisms to lodge grievances. Maybe the public also prefers to deny the reality of medical error, except when that reality is thrust upon them, and then they become plaintiffs.

Accepting Error, Not Accepting Errors

Some authors have addressed the issues of legitimate human fallibility, the inherent uncertainty in medicine, and negligence, and described how more observer variation, and error are allowable for difficult cases (sk119), where most radiologists will make errors. Isolated errors might be excusable, even if not the best potential performance, but a pattern of misses might point to incompetence (sk16). Perhaps negligence is the reading of a given type of case material after inadequate performance on a standard series of known subtle cases of that type, this defining negligence before, rather than after, patients suffer injury, and before, rather than after, physicians are sued. Or perhaps negligence is not changing some aspect of practice methods after making serious class 3 or class 4 errors (sk16,sk19,sk20). Consistent with such an approach, some authors observe that (sk225p1297) board certification is an inadequate marker for competence, and suggest that particularly qualified physicians with enhanced training in the interpretation of cranial CT should have cases triaged to them for optimal interpretive performance.

Flagrant radiological errors reveal a defect in some element in the chain of perception and reporting, and because they occur regularly, does not mean they are normal. Errors may mean that there are too many interruptions and distractions in the reading area. They may mean that hospital information system software, written with subroutines inadequate for error trapping of keypunch errors, is scrambling and misidentifying patient reports - harder to detect when using uniform, canned report formats, and perhaps explaining, in some cases, the apparent outrageous errors on the radiologist's part, where the programmer's/vendor's sloppy code or poor system design is actually the source of error.

Or errors may mean that the speed of reading is excessive - time truncation.


Error persists unchanged in part due to pressures outside the profession producing latent sources of error, thus preventing the changes in work habits required to reduce errors (sk79p407). Both Richard Feynman (sk29p155) and Christiansen (sk14p365), from different fields, deplore speed driven habits that risk others' lives, using the terms 'Russian roulette' and 'gambling' to describe this practice.

Richard Feynman's, in the The Pleasure of Finding Things Out (sk29), discusses the Challenger disaster, a model of conflict between productivity oriented managers, with quantitative performance goals, producing top-down latent errors, and engineers who actually do the work (at the sharp end), with a quality improvement orientation. The substantial discrepancy between working engineers and management in terms of the acceptance of imperfection, and subversion of the mission, in favor of the schedule and appearances, is reminiscent of recent financial scandals, and of changes in health care (sk217sk274p219).

The business/customer mode of medical practice invites similar pressures (sk31p195) as the doctor/patient relationships evolves to a vendor/customer relationship, with all the negatives that that implies - like caveat emptor, and the tendency for business mores to favor the development of company identity (sk145p66), rather than a professional identity with customer interests at heart.

Like space missions, radiology also has a schedule, determined externally by the medical and psychological needs of patients and referring doctors, and by pressures from administrators and cost accountants who seek to preserve financial viability, similar to the shuttle project's need to retain funding from sources who know, and care comparatively less, about the actual details of space flight. Such a dichotomy between managers and technical workers is supported as a norm by the accepted wisdom that managers need not know much about the field they manage (sk145p57,58), and by management systems that favor "non-ability" factors (sk145p61).

Latent Errors, and the Slow Rot of Speed

Feynman describes the corrosive effect of shuttle launch schedule priorities, and public relations driven management culture, on the rigor of the process used to give the OK to launch. Overlooking blow-by around gaskets ultimately had disastrous effects. Radiological blow-by errors, those particularly egregious errors that signal an underlying defective diagnostic process, should induce a change, rather than validate that such things happen routinely, and are therefore normal. (sk100). Foam insulation flying apart during shuttle launches (sk244) was regarded as an acceptable abnormality rather than a safety risk - in other words, because it happened frequently, it was normal.

Given the high rate of diagnostic error in retrospective studies, it is not difficult to see how Feynman's observations in chapter 7 might extend to medical care. Persistent known flaws were repeatedly allowed to go unchecked (sk29p152). Managers concerned about their schedule, funding, and organizational image failed to listen to warnings from those closer to the work (sk29pp152,168,169). Known failures and near misses (sk29p154-55) went unheeded as clues to potentially larger problems - somewhat like the class 3 or class 4 errors (sk16,sk20, sk19) that radiologists occasionally make that signal opportunities for improvement, as well as chagrin.

Feynman describes a situation where the reliability of accepted methods is exaggerated (sk29p155), and safety considerations gradually eroded (sk29p162). Quantitative estimates of outcome probabilities might be used to obscure failures to observe optimal engineering practices (sk29p154,155,156) - somewhat like the manner in which the prospective 5% error rate in radiology performed on the usually normal population sounds more respectable than error rates on the order of 20-20% found in retrospective studies (sk29p156), or the manner in which the overall outcome of high efficacy of treatment of stroke might mask disasters deriving from using thrombolytics on patients with undiagnosed intracranial hemorrhage (sk224, sk225.)

Like Bradbury's character Clarisse in Fahrenheit, Feynman also argues for a detailed, knowing grass in detail, complete understanding (sk29p159), in order to have the awareness and understanding of flaws in technology well enough to fix them (sk29p169), rather than ignoring the reality of imperfections in favor of maintaining public relations (sk29p169). The British health system is reported to have similar problems (sk120) with a management culture of cursory analysis that fails to employ lessons learnable from near accidents.

The 'best of all possible worlds' (sk202) approach is satisfied that infrequent errors are exceptions and can't be addressed by systemic changes, and Catch 22 (sk261) - that errors that do happen regularly and routinely, because they are routine and regular, are a normal part of the landscape, and nothing can be done about them either. This is somewhat like my car's "maintenance required" light that is constantly on - "Don't worry, it's been like that for years now".

Amid the controversies and uncertain grey areas of medical practice, by selective manipulation of definitions and criteria, or ignoring other considerations, there is the potential for sculpting data to arrive at favorable and/or desired conclusions (sk272). The down side of impression management done well, with statistical reports contoured to truthfully tell misleading things, is that it can suppress the humility needed to stimulate better practice (sk29p169). The 4% failure rate of solid fuel rockets, quoted by Feynman, is surprisingly close to the 5% prospective failure rate in radiology (sk16), or the 3% substandard misses found (sk16). Perhaps 3-5% error is the administratively and publicly acceptable rate of error in any field, with measures of error adopted as reasonable only when they reasonably result in such a sufficiently low, publicly acceptable rate of error.

Redemptive Redundancies, Outcomes Studies, Fig Leaves

As with space launches, redundancies in medicine save doctors and patients, even, and especially, amid the reality of medical imperfection (sk89p943). Redemptive redundancies can produce favorable overall results, despite flaws in individual steps of the diagnostic process. A good radiologist picks up something missed by a distracted, or overly focused clinician. An astute and persistent clinician picks up something missed by both the ER doc and a sleepy radiologist.

Redemptive redundancies (sk41p43) unfortunately can also create the illusion that underlying component processes are more robust than they in fact are. Individual flaws might be ignored, and not considered on their own merits, because they may not lead often, in the aggregate, to a negative result, a potentially deceptive and misleading effect of only looking at final outcomes (sk29p159). As long as cost cutting, economies, or other pressures don't remove redundancies, and don't disastrously expose the failings of component processes mistakenly thought adequate, there can be acceptable overall performance on final outcomes, and no quality concerns.

A study looking at the role of interpretation of head CTs for hemorrhage when treating stroke patients with thrombolytics makes this point (sk225). The authors note that a population based perspective provides a less stringent demand on imaging performance. That is, due to the nature of the disease,and the efficacy of the new treatment, sensitivity for hemorrhage greater than 75% (note the approximation to 80/20) will maintain a potential net death benefit in treated patients. In other words, the treatment of those with stroke is so good that efficacy of treatment overwhelms the disasters of diagnosis that occur when thrombolytics are given to patients erroneously thought not to have intracranial hemorrhage. Focusing on the final aggregate outcome masks diagnostic deficiencies that produce disaster for individual patients. In this way, much touted outcomes studies might be more a fig leaf than a tool for more cost effective medicine. The authors found that sensitivity for intracranial hemorrhage could dip to 22% before overwhelming improvements in global neurologic result. People have been reading head CT scans for hemorrhage for quite awhile now. One wonders what the problem is.

This is similar to the pattern that Feynman found with shuttle launches. In radiology, the analog of Feynman's more rigorous evaluation of launch component processes would be testing accuracy on subtle abnormal cases, not testing how we did on cases where the redemptive fact that the patients were mostly normal served to veil defects in perceptual discipline.


Internal resistance to quality change can also occur inside a field or discipline through custom, habit, or other consensus enforcing mechanisms. Groupthink: Psychological Studies of Policy Decisions and Fiascoes, by Irving Janis (sk30), explores, in the Munich decision, Bay of Pigs, and other debacles, how a bias for short term, premature, comfortable consensus subverts accomplishing the long term stated goals of an organization. The shuttle disasters, Bay of Pigs, multiple nautical SNAFUs (sk208), including Titanic, Andrea Doria, Greenville (sk54), 'Herald of Free Enterprise' (sk274p37), or the firehouse culture in Fahrenheit, are a few examples of the down side of groupthink (sk269).

Forces inside any organization may resist needed changes for improvement, as found in a British review (sk120), and after the Columbia disaster (sk244). Groupthink may function as a tool to blow smokescreens over dysfunctional teams. Physicians have been reported to switch their own blood for that of a drinking colleague, in order to circumvent police blood tests, and thus protect him from a DUI charge going to the extreme of placing him on a ventilator to stall for the time needed to clear the alcohol from his blood (sk274,viii). Likewise, the down side of teamplayer groupthink is generally glossed over, although uncommonly these negatives are identified (sk120) by sources who distinguish between teams and cliques (sk89, sk274).

Numerous reports have described the toxic effect of strong intragroup bonds in scenarios where small, insular elites have unmonitored control of, and responsibility for, relatively powerless subject populations, producing ethical failures (sk143, sk30, sk267pp74-76). Groupthink among NASA managers as described by Feynman (sk29), may have played a part in the organizational inertia which repeatedly tolerated flawed operation. Groupthink ingroup/outgroup dynamics may have been operative in well-known cases like the Tuskegee syphilis study in African-Americans (sk246,sk247) and the Willowbrook (sk235) study of mentally defective children infected with viral hepatitis. Groupthink is common, and other less flagrant examples of actual or potential ingroup/outgroup groupthink interactions have also been described in other more prosaic Anglo-American medical and non-medical settings (sk37,sk31pp125-126,190; sk252, sk253,sk256, sk258,sk262pp30,229-238,270,275,276,281-291; sk267pp17,55,56; sk228pp201, 203, 205,209,211,242,273,342,383,384). Probably more examples could be collected if there were not a bias for repressing such things (sk235).

With groupthink, loyalty tends to be reserved for those of similar social status - like teammates - and operates more powerfully between colleagues who are the constants in one's social milieu. Responsibility toward subject populations such as prisoners, experimental subjects, or patients, has far less durable psychological and social support. As with the jailers in the Zimbardo/Abu Ghraib (sk243,sk242) examples,the strongest interpersonal forces are between physician colleague team players - not between doctors and patients - since patients come and go, and less commonly impact significantly on an individual physician. This dual loyalty problem may apply to military settings such as Abu Ghraib (sk243, sk242), or to managed care settings (sk217), particularly if these institutions disempower physicians. The Abu Ghraib incident (sk242), and the experiments of Zimbardo and Milgram (sk143) have demonstrated this phenomenon in persons generally thought to be normal, well adjusted Americans. The use of various epithets applied to physicians or patients, such as SHPOS (Sub-Human Piece of Shit) (sk227), or Gomer, etc. (sk228) is symptomatic of a similar trend to dehumanization latent in the health care milieu.

The detachment that Goodall describes in scientists (sk195p59), with the investigators adhering to the experiment - as in the Milgram example (sk243) - despite the risk of harm to animals, subjects, or patients that they have distanced from themselves, might also occur in highly manipulated medical reimbursement systems. The interests of the animals, prisoners, subjects, or patients - unvoiced and invisible - may be disappeared as economic or career motives eventually outweigh ethical considerations. To the extent that they are subordinated and disempowered by the systems in which they function, physicians become more susceptible to groupthink derived ethical failures and top-down abuse along the lines of the Milgram model (sk143).

Groupthink in Radiology

Social pressure is explicitly cited as one cause of false negative errors in radiology (sk40p149). Image quality is held to be determined by community consensus (sk76).

Groupthink effects have been identified in radiologic practice, usually - though not always - negative. A refreshing exception is described by an administration initiated quality review (sk16), where one hospital F was identified as an outlier, producing unusually accurate reports, which was attributed to organizational culture at that hospital.

Looking at the diagnostic process applied to already perceived lesions, a study by Hillman (sk264) found that group consensus had a positive influence on the accuracy of a majority of radiologists, with the radiologists obtaining the most benefit from the consensus process being those whose initial diagnostic accuracy was the least. The information exchange of the consensus process produced results superior to simply voting the group's initial diagnoses, a finding dramatized by Larson's Far Side cartoon spoof (sk259) depicting surgeons deciding, in the midst of operative dissent, to vote on how many chambers the heart has.

Otherwise, groupthink effects tend to the negative. The accuracy of the highest performers in the Hillman study was actually lessened by the group consensus process, demonstrating the pressure to the mean of group processes, the significant influence of groupthink pressure on dissenters, and the inability of groups to discern their most able performers (sk264), who generally benefited little from group interaction. The tendency of physicians to group themselves according to competence reinforces this effect (sk274p17).

Another paper (sk232) on chest interpretations describes how a single individual arbitrator performed better than a committee of arbitrators. With a single arbitrator, consensus pressures are minimized, but personal performance pressure in an accountable social context is maximized. These investigators found that pseudo-arbitration, the use of a third arbiter unaware of the specifics of disagreement between two previous interpreters, and therefore not swayed by knowing their opinions, was an overall superior method.

Negative groupthink effects are consistent with the diffused, lesser felt accountability of committees, and with the possibility of consensus pressures playing a damping role on leading edge performers - consensus always exerting a pull toward the mean (sk264). The Ucurve of performance found by Lesgold (sk223) may simply be the mirror image of the Gaussian distribution of physician abilities and talent.

Well recognized manifestations of groupthink, such as alliterative error, (sk24), can result in previous reports adversely affecting independent analysis of subsequent exams. Marcus Smith explicitly describes alliterative error as the result of groupthink (sk49pp75-77), and then goes on (sk49p96) to describe how field dependent and field independent personalities might impact film interpretation, the field dependent type particularly subject to pressures from their environment, and susceptible to groupthink.

Garland (sk63) alludes to groupthink when, writing in an era that had time for such things, he quotes the four fallacies described by Sir Francis Bacon (sk128). These fallacies or idols - idols of the tribe, idols of the den, idols of the market, and idols of the theatre - essentially represent social pressures and effects that in various ways hamper understanding.

When radiologists were asked to read CT scans at .5 - 21 frame rates per second (sk133), the total reading time decreased as the frame rate increased, suggesting that there was an insidious, and velocitized, field-dependent response to externally, implicitly encouraged speed norms at the higher frame rates, a pattern reminiscent of the Milgram experiment (sk143), with the dial of the machine always wanting, and subliminally demanding more, faster, or both. Thankfully, the radiologists instinctively preferred non machine selection of frame rates.

Group error, with enough solidarity of error, may not count as error. In one series (sk20), cases were excluded when more than two faculty members made the same error. Another study (sk32) excluded erroneous consensus conclusions from error analysis, thus minimizing the apparent number of errors. (sk119) Robinson clearly distinguishes between accuracy and precision, consensus group error.

Groupthink derived pressures for consensus may have the effect of damping down the more perceptive who are necessarily out beyond the norm. In mammographic screening, 25% of cancers detected, were found by only one of two screeners, these being disproportionatley the more difficult, subtle, and generally less advanced cancers (sk234). These investigators went to particular trouble to avoid premature groupthink consensus pressures, removing wax marks, or other consensus clues, made by the initial screener, emphasizing that they were using independent double reading. Recalls for further imaging occurred when even one of the screeners found an abnormality. Such a method minimizes groupthink, but uses social pressure positively, by enhancing the implicit audience effect, since each screener knows that at least one other person will look closely at every case, perhaps increasing the viewer's felt jeopardy/vulnerability, and thereby enabling them to empathize with the patient's vulnerability, like a sudden death playoff. The calculated second screener contribution emphasizes the value of this independent opinion, tending to enhance felt responsibility, rather than consensus which - dissolving personal accountability - tends to diffuse, and diminish, felt responsibility.

A study of CT interpretation in patients with ovarian carcinoma found that independent 'replicated readings', a form of pseudoarbitration, were superior to consensus readings (sk230), perhaps offering the benefit of two readings, but unhampered by consensus interpretations that would diminish the effect of uncommonly perceptive observations (sk264).

Another study of mammograpic double reading similarly chose to avoid consensus double reading in favor of logical rule driven independent double reading (sk255), noting how, for a true double reading, subjective influence and pressures to consensus must be excluded to avoid interpretation bias - essentially, groupthink.

Maybe we all like to believe that we belong to the best of all possible worlds (sk202), one of the known risk factors for groupthink. Amiability, a risk factor for group think (sk30/intro,p13) may result in physicians - prone to avoid conflict (sk86) - underestimating the degree to which medical care is unsafe, preferring to emphasize the positive, despite (sk120) studies that describe under-reported preventable injuries due to medical error (sk91). Esprit de corps, another groupthink risk factor (sk30,intro,p13), is encouraged by the demands and stressors of medical education and practice that build group cohesion and homogeneity (sk30pp239,174,185), these latter traits also risk factors for groupthink.

Contrary to advice to avoid mentioning error (sk41,sk219), radiologists occasionally suggest being candid about acknowledging the limits of our accuracy (sk100), but institutions permeated by groupthink are probably structurally unable to embrace such recommendations for candor (sk257). Referring doctors are probably not willing to deal with such candor in reports, and candor for purposes of risk management may conflict with the needs of marketing, and patient confidence, as for example, with mammography.

Groupthink, by enforcing speed norms, is likely to be important in maintaining the use of speed (sk269), despite the negative effects of speed on perception. Being carried along by the dominant spirit of the social environment is a common human tendency. The usual and customary rate of image interpretation, and attendant perceptual error, constant over so many decades, functions as a norm, like the normal speed of 80 mph on a 70mph freeway, velocitizing drivers, so that going slower feels stressful, like being wronged - sometimes eliciting road rage. Others have noticed common factors between road accidents, and medical accidents (sk274p41,219). Once the use of speed becomes habitual, the conditions of its existence, and the tangible benefits it provides, then become the baseline, and it becomes unthinkable to slow to a safer speed (sk269). Speeding violations become normative, and slowing down becomes falling behind - declasse - maintaining the use of speed, and precluding the less measurable benefits to the self and others that slowness offers (sk1pp9-11). Validated as normal by their fellows, speedsters in a fog zoom along thoughtlessly, an accident waiting to happen, and perceiving a problem only when faced with crises and 200 car pileups (sk147). Inevitably, the fog or some other factor will be blamed for the accidents that inevitably occur when unsafe speeds are normal. The roast burned.


Denial of error, and ignoring errors that occur in the process of radiology, or any other field, can result in a situation where what we are supposed to do, say we do, and then actually do - are different things. Here again, Bradbury's Fahrenheit, together with Janis' book on Groupthink, becomes pertinent, illustrating how even foolish or brutal acts can become normalized. Radiologists might enjoy Fahrenheit 451, noticing the toxic effect of groupthink (sk30) on any profession, and describing how in the past one of the tasks of firemen was to actually put out fires, calling it normal.


For those patients who depend on our skill and care,

to my teachers, those named and unnamed, for their skill and care,

and to my First Teachers,

my parents,

of course.

Thank you for any corrections, comments offered.

Contact: speedinput04@yahoo.com

Copyright 2004, RW Jasinski, all rights reserved



sk1- Bradbury, R. (1950) Fahrenheit 451. Random House. New York. - see pp 9-11.

sk2 - Berlin L. (August 2002) Reducing Errors in Radiology. RSNA News. p 10-12.

sk3 - Ania BJ, Asenjo M. (2002) Why a CT Scan Instead of a History and Physical Examination? Radiographics 22(6): 1524.

sk4 - Kundera M. (1996) Slowness. Harper Collins. New York.

sk5 - Bosk CL (1979) Forgive and Remember: Managing Medical Failure. University of Chicago Press. Chicago.

sk6 - Dreyfus SE, Dreyfus HL. (February 1980) A Five-Stage Model of the Mental Activities Involved in Directed Skill Acquisition. Operations Research Center Berkely, California.

sk6b - LeBlond RF. (December 2002) Dreyfus Model of Skill Acquisition, Handout. Personal Communication, University of Iowa College of Medicine.

sk6c - Dreyfus, HL. (2001) On the Internet Routledge. London, New York. - see esp. chapter 2.

sk7 - Pirsig RM.(1974) Zen and the Art of Motorcycle Maintenance: An Inquiry Into Values. William Morrow. New York. - see esp. pp9-17, 20-23,24,25, 45-50,59-67,69-72,148-150 ,247-269,281

sk8 - MacMahon H, Engelmann R, Behlen FM, et al. (1999) Computer-aided Diagnosis of Pulmonary Nodules: Results of a Large-Scale Observer Test. Radiology 213: 723-726.

sk9 - Shay J. (1994) Achilles in Vietnam. Simon and Schuster, New York.

sk10 - Henderson C. (1986) Marine Sniper. Berkeley Publishing Group. New York.

sk11 - Dillard A. (1985) Pilgrim At Tinker Creek. Harper and Row. New York.

sk12 - White SE. (1904) The Mountains. Hurst and Co. New York. (see the chapter "On Seeing Deer", pp116-128)

sk13 - Keller EF. (1983) A Feeling for the Organism: The Life and Work of Barbara McClintock, WH Freeman and Co., New York.

sk14 - Christensen EE, Murry RC, Holland K, et al. (1981) The Effect of Search Time on Perception. Radiology 138: 361-5.

sk15 - Herrigel E. (1953) Zen in the Art of Archery. Random House, New York.

sk15b - Herrigel GL. (1958) Zen in the Art of Flower Arrangement. Routledge and Kegan Paul, Ltd. London.see esp. p6-12, 19-23.

sk16 - Siegle RL, Baram EM, Reuter SR, et al. (1998) Rates of Disagreement in Imaging Interpretation in a Group of Community Hospitals. Academic Radiology 5:148-154.

sk17 - Berbaum KS, Franken EA Jr., Dorfman DD, et al. (2000) Role of Faulty Decision Making in the Satisfaction of Search Effect in Chest Radiography. Academic Radiology 7(12): 1098-1106.

sk18 - Carmody DP, Nodine CF, Kundel HL. (1980) An Analysis of Perceptual and Cognitive Factors in Radiographic Interpretation. Perception. 9: 339-344.

sk19 - Franken EA Jr. (1998) Individual Error in Radiology. Academic Radiology 5:147.

sk20 - Cascade PN, Kazerooni EA, Gross BH, et al. (2001) Evaluation of Competence in the Interpretation of Chest Radiographs. Academic Radiology 8(4): 315-321.

sk21 - Berbaum, KS. (1995) Difficulty of Judging Retrospectively Whether a Diagnosis Has Been "Missed". Radiology 194:582-3.

sk22 - Sobel R, Elata G. (1998) The Problems of Seeing and Saying in Medicine and Poetry. Perspectives in Biology and Medicine 44(1): 87-98. Johns Hopkins University Press, Baltimore.

sk23 - Nodine CF, Mell-Thomas C, Kundel HL, et al. (2002) Time Course of Perception and Decision Making During Mammographic Interpretation. AJR 179: 917-923.

sk24 - Berlin L. (2000) Malpractice Issues in Radiology: AlliterativeErrors. AJR. 174: 925-931.

sk25 - Deitch CH, Chan WC, Sunshine JH, et al. (1994) Quality Assessment and Improvement: What Radiologists Do and Think. AJR. 163: 1245-1254.

sk26 - Berbaum KS. Franken EA, Dorfman DD, et al. (1991) Time Course of Satisfaction of Search. Investigative Radiology 26: 640-648.

sk27 - Revesz G, Kundel HL, Graber MA. (1974) The Influence of Structured Noise on the Detection of Radiologic Abnormalities. Investigative Radiology 9(6): 479-86.

sk28 - Robinson HM. (Feb. 1990) A Double Standard of Care in Mammography? ACR Bulletin.

sk29 - Feynman R. (1999) The Pleasure of Finding Things Out. Perseus Books, Cambridge, Massachusetts.

sk30 - Janis IL. (1982) Groupthink: Psychological Studies of Policy Decisions and Fiascoes, 2nd ed. Houghton Mifflin. Boston.

sk31 - More ES, Milligan M, eds. (1994) The Empathic Practitioner: Empathy, Gender, and Medicine. Rutgers University Press, New Brunswick, New Jersey.

sk32 - Herman PG, Gerson DE, Hessel SJ, et al. (1975) Disagreements in Chest Roentgen Interpretation. Chest 68(3): 278-282.

sk33 - Shah PK, Austin JHM, White CS, et al. (2003) Missed Non-Small Cell Lung Cancer: Radiographic Findings of Potentially Resectable Lesions Evident Only in Retrospect. Radiology 226:235-241.

sk34 - deleted.

sk35- Kobayashi T, Xin-Wei X, MacMahon H, et al. (1996) Effect of a Computer-aided Diagnosis Scheme on Radiologists' Performance in Detection of Lung Nodules on Radiographs. Radiology 199: 843-848.

sk36 - deleted.

sk37- Austin JHM, Romney BM, Goldsmith LS. (1992) Missed Bronchogenic Carcinoma: Radiographic Findings in 27 Patients with a Potentially Resectable Lesion Evident in Retrospect. Radiology 182: 115-122.

sk38 - Ferrari M. (1992). Colors For Survival: Mimicry and Camouflage in Nature. Barnes and Noble Books. New York.

sk39 - Krupinski, EA. (2000) The Importance of Perception Research in Medical Imaging. Radiation Medicine 18 (6):329-334.

sk40 - Renfrew EL, Franken EA, Berbaum KS, et al. (1992) Error in Radiology: Classification and Lessons in 182 Cases Presented at a Problem Case Conference. Radiology 183: 145-150.

sk41 - Yoon LS, Haims AH, Brink JA, et al. (2002) Radiology Evaluation of an Emergency Radiology Quality Assurance Program at a Level I Trauma Center: Abdominal and Pelvic CT Studies. Radiology 224:42-46.

sk42 - Potchen EJ, Cooper TG, Sierra AE, et al. (2000) Measuring Performance in Chest Radiography. Radiology 217: 456-9.

sk42b - Maimonides (1958) The Preservation of Youth: Essays on Health Philosophical Library, New York.

sk43 - Sunshine JH, Burkhardt JH. (2000) Radiology Groups' Workload in Relative Value Units and Factors Affecting It. Radiology 214(3):815-22.

sk44 - Wood BP. Visual Expertise. (1999) Radiology 211: 1-3.

sk45 - Bhargavan M, Sunshine JH, Schepps B. (2002) Too Few Radiologists? AJR 178(5):1075-82.

sk45b - Potchen EJ. (1998) The Medical Image Perception Society and the Future of Imaging Research. Radiology 209(3): 613-4.

sk46 - Sunshine JH, Cypel YS, Schepps B. (2002) Diagnostic Radiologists in 2000: Basic Characteristics, Practices, and Issues Related to the Radiologist Shortage. Radiology 178 (2): 291-301.

sk47a - Bhargavan M, Sunshine JH. (2002) Workload of Radiologists in the United States in 1998-1999 and Trends Since 1995-1996. Radiology 179(5): 1123-8.

sk47 - Samuel S, Kundel HL, Nodine CF, et al. (1995) "Mechanism of Satisfaction of Search: Eye Position Recordings in the Reading of Chest Radiographs". Radiology 194: 895-902.

sk48 - Egglin TKP, Feinstein AR. (1996) Context Bias: A Problem in Diagnostic Radiology. JAMA 276(21):1752-1755.

sk49 - Smith MJ. (1967) Error and Variations in Diagnostic Radiology. CC Thomas. Springfield, Illinois.

sk50 - Kundel HL, Nodine CF. (1978) Studies of Eye Movements and Visual Search in Radiology, pp 317-328. in Eye Movements and the Higher Psychological Functions., Eds. Senders JW, Fisher DF, Monty RA. Lawrence Erlbaum Associates. Hillsdale, NJ.

sk51- Schneider W, Shiffrin RM. (1977) Controlled and Automatic Human Information Processing: I. Detection, Search, and Attention. Psychological Review 84(1):1-66.

sk52 - Schneider W, Shiffrin RM. (1977) Controlled and Automatic Human Information Processing: II. Perceptual Learning, Automatic Attending, and a General Theory. Psychological Review 84(3):127-190.

sk53 - Berlin L. (2000) Must New Radiographs Be Compared with All Previous Radiographs, or Only with the Most Recently Obtained Radiographs? AJR 174: 611-615.

sk54 - Heilman RS. (2002) What Did the CT Scan Show? Radiographics 22(4):894.

sk55 - MacMahon H. (2000) Improvement in Detection of Pulmonary Nodules: Digital Image Processing and Computer-Aided Diagnosis. Radiographics 20(4): 1169-1177.

sk56 - Kundel HL, Nodine CF, Carmody D. (1978) Visual Scanning, Pattern Recognition and Decision-making in Pulmonary Nodule Detection. Investigative Radiology 13 (3): 175-181.

sk57 - Bass JC, Chiles C. (1990) Visual Skill: Correlation with Detection of Pulmonary Nodules. Investigative Radiology 25: 994-998.

sk58 - Garland L H. (1959) Studies on the Accuracy of Diagnostic Procedures. AJR 82(1): 25-38.

sk59 - Parker TW, Kelsey CA, Moseley RD, et al. (1982) Directed Versus Free Search for Nodules in chest Radiographs. Investigative Radiology 17 (2): 152-155.

sk60 - Berbaum KS, Franken EA Jr., Dorfman DD et al. (1990) Satisfaction of Search in Diagnostic Radiology. Investigative Radiology 25 (2): 133-140.

sk61 - Berbaum KS, Franken EA Jr., Anderson KL, et al. (1993) The Influence of Clinical History on Visual Search with Single and Multiple Abnormalities. Investigative Radiology. 28 (3): 191-201.

sk62 - Berbaum KS, Franken EA Jr, Dorfman DD, et al. (1986) Tentative Diagnoses Facilitate the Detection of Diverse Lesions in Chest Radiographs. Investigative Radiology 21 (7): 532-539.

sk63 - Garland L H. (1949) On the Scientific Evaluation of Diagnostic Procedures. Radiology 52(3): 309-327.

sk64 - Riebel FA. (1958) Use of the Eyes in X-ray diagnosis. Radiology 70(2): 252-8.

sk65 - Tuddenham WJ. (1962) Visual Search, Image Organization, and Reader Error in Roentgen Diagnosis. Radiology 78: 694-704.

sk66 - Kundel HL, Nodine CF. (1975) Interpreting Chest Radiographs Without Visual Search. Radiology 116: 527-532.

sk67 - Kundel HL, La Follette PS Jr. (1972) Visual Search Patterns and Experience with Radiological Images. Radiology 103: 523-528.

sk68 - Smoker WRK, Berbaum KS, Luebke NH, et al. (1984) Spatial Perception Tests in Diagnostic Radiology. AJR. 143:1105-1109.

sk69 -Muhm JR, Miller WE, Fontana RS et al. (1983) Lung Cancer Detected During a Screening Program Using Four-Month Chest Radiographs. Radiology 148: 609-615.

sk70 - Carmody DP, Kundel HL, Toto LC. (1984) Comparison Scans While Reading Chest Images: Taught, But Not Practiced. Investigative Radiology 19(5): 462-466.

sk71 - Nodine CF, Kundel HL, Mello-Thomas C, et al. (1999) How Experience and Training Influence Mammography Expertise. Academic Radiology 6:575-585.

sk72 - Cascade PN. (1990) Quality Improvements in Diagnostic Radiology. AJR 154: 1117-1120.

sk73 - Friedenberg RM. (2000) An Endangered Art: Teaching. Radiology 214: 317-319.

sk74 - deleted.

sk75 - Jackson FI. (1990) Perception of Radiologic Images. AJR 154:1121. letter.

sk76 - Krupinski EA, Kundel HL, Judy PF, et al. (1998) The Medical Image Perception Society: Key Issues for Image Perception Research. Radiology 209: 611-612.

sk77 - Bechtold RE, Chen MYM, Ott DJ, et al. (1997) Interpretation of abdominal CT: Analysis of Errors and Their Causes. Journal of Computer Assisted Tomography 21(5): 681-685.

sk78 - Pentecost MJ. (1998) Measuring Professional Quality in Radiology. AJR 170: 843-846.

sk79 - Thomas E. L. (1969) Search Behavior. Radiologic Clinics of North America 7(3): 403-417.

sk80 - Berlin L, Hendrix RW. (1978) Malpractice Issues in Radiology: Perceptual Errors and Negligence. AJR 170:863-867.

sk81 - deleted.

sk82 - Taylor CR. (1998) The Measure of Professional Quality in Radiology. AJR 170: 847-850.

sk83 - Dockray KT, Sawyer JW, Elks ML, Vines DL. (1997) Using a Computer to Track the Image Quality in Diagnostic Errors. AJR 169:635-636.

sk 84 - Tudor GR, Finlay DB. (2001) Error Review: Can This Improve Reporting Performance? Clinical Radiology 56: 751-754.

sk85 - Gale AG, de Silva ES, Walker GE, et al. (1989) "Vigilance Decrement and Radiological Reporting" In: Megaw ED, ed. Contemporary Ergonomics. London, England: Taylor and Francis. pp461-467.

sk86 - Lee TL. (2002) A Broader Concept of Medical Error. New England Journal of Medicine 347(24): 1965-1967.

sk87 -Blendon RJ, DesRoches CM, Brodie MB, et al. (2002) Views of Practicing Physicians and the Public on Medical Errors. New England Journal of Medicine 347(24): 1933-1940.

sk88 - Tuddenham W J. (1963) Problems of Perception in Chest Roentgenology: Facts and Fallacies. Radiologic Clinics of North America 1: 277-289.

sk89 - Fitzgerald R. (2001) Error in Radiology. Clinical Radiology 56: 938-946.

sk90 - Goddard, A Leslie, A Jones, et.al. (2001) A Short Communication, Error in Radiology. British Journal of Radiology 74(886): 949-51.

sk91 - Robblee JA, Nicklin WL. (2003) Views of Practicing Physicians and the Public on Medical Errors. New England Journal of Medicine 348(12):1184-1185.

sk92 - Armstrong P, Waters DB. (2000) Invited Commentary on Job Stress and Satisfaction Among Clinical Radiologists. Clinical Radiology 55:186.

sk93 - Graham J, Ramirez AJ, Field S, Richards MA. (2000) Job Stress and Satisfaction Among Clinical Radiologists. Clinical Radiology 55(3): 182-5.

sk94 - Buschi AJ, Brenbridge ANAG. The Dog in the Nighttime: A Quiet Clue. AJR 133(1):157.

sk95 - Mitchell JM, Sunshine JH. (2002) Determinants of Differences Among Radiologists in Starting Salaries. American Journal of Roentgenology 178: 1067-1073.

sk96 - Anonymous (September 1995) ACR Study Helps Radiologists Evaluate Their Hospital Practices. ACR Bulletin 51(9); p1,12.

sk97 - deleted.

sk98 - Practice Builders (1979) from Decisions in Imaging Economics: the Journal of Imaging Technology Management, 1800-679-1200, extension 235. see p3.

sk99 - Kundel HL. (1979) " Images, Image Quality, and Observer Performance. Radiology 132:265-271.

sk100 - Bates BF. (May 1989) To err, per se, is not malpractice. ACR Bulletin p12.

sk101 - deleted.

sk102 -Tuddenham WJ. (1969) Roentgen Image Perception - A Personal Survey of the Problem. Radiologic Clinics of North America 7(3): 499-501.

sk103 -Kundel HL, Nodine CF. (1983) A Visual Concept Shapes Image Perception. Radiology 146: 363-368.

sk104 - deleted.

sk105 -Kelvin FM, Gardiner R, Wenzel V, Stevenson GW. (1981) Colorectal Carcinoma Missed on Double Contrast Barium Enema Study: A Problem in Perception. AJR 137: 307-313.

sk106 -Nodine CF, Kundel. (1987) Using Eye Movements to Study Visual Search and to Improve Tumor Detection. Radiographics 7(6): 1241-1250.

sk107- Hendee WR. (1987) The Perception of Visual Information. Radiographics 7(6): 1213-1219.

sk108 - Gray JE, Taylor KW, Hobbs BB. (1978) Detection Accuracy in Chest Radiography. American Journal of Roentgenology 131: 247-253.

sk109 - deleted.

sk110 - McNeil BJ, Hanley JA, Funkenstein HH, Wallman J. (1983) Paired Receiver Operating Characteristic Curves and the Effect of History on Radiographic Interpretation. Radiology 149: 75-77.

sk111 -Doubilet P, Herman PG. (1981) Interpretation of Radiographs: Effect of Clinical History. American Journal of Roentgenology 137: 1055-1058.

sk112 - Gale AG, Worthington BS. (1984) Image Viewing Habits and the False Diagnosis. Diagnostic Imaging, (November 1984): 130-133.

sk113 - Anonymous (1994) Observations Concerning the Uneducated Retina. JAMA 271(3): 172B. (from Journal of the American Medical Association 100 years ago ). (Still, this seems strange to render things upside down - perhaps the native artists were merely being polite to their new visitors, and - if facing them - drew objects in an inverted perspective for the benefit of the visitors, rather than from their own viewer's perspective. - RWJ).

sk114 -Ridley EL. (1994) PACS Centers Achieve Widespread Acceptance. Diagnostic Imaging September 1994, p3-6.

sk115 - Sunshine JH, Bushee GR, Mallick R. (1998) US Radiologists' Workload in 1995-1996 and Trends since 1991-1992. Radiology 208(1): 19-24.

sk116 - Sunshine JH, Bansal S. (1994) Operational Characteristics of Radiology Groups in the United States in 1992. Radiology 193(3): 613-618.

sk117 - Sunshine JH, Bansal S. (1992) Operational, Professional, and Business Characteristics of Radiology Groups in the United States. Radiology 183(2):535-540.

sk118 - Berlin L. (1999) Comparing New Radiographs with Those Obtained Previously. AJR 172: 3-6.

sk119 - Robinson PJA. (1997) Radiology's Achilles Heel: Error and Variation in the Interpretation of the Roentgen Image. British Journal of Radiology 70: 1085-1098.

sk120 - Barach P, Small SD. (2000) How the NHS can Improve Safety and Learning (By Learning Free Lessons from Near Misses). BMJ 320: 1683-1684.

sk121 - Dobson R. (2001) Whistleblower Thanked but Still Suspended. BMJ 322: 634 (17 March).

sk122 - Dyer C. (2000) Whistleblower Inquiry Clears British Hospital. BMJ 321: 68. (8 July).

sk123 - Dyer C. (1999) London Whistleblower Left with Career in Tatters. BMJ 319: 142. (17 July).

sk124 - Kiger PJ. (2001) Truth and Consequences. Working Woman May, pp. 56-61.

sk125 - Kundel HL. (1981) Predictive Value and Threshold Detectability of Lung Tumors. Radiology 139: 25-29.

sk126 - Kundel HL, Revesz G. (1976) Lesion Conspicuity, Structured Noise, and Film Reader Error. American Journal of Roentgenology 126: 1233-1238.

sk127 - Oestmann JW, Greene R, Kushner DC, Bourgouin PM, Linetsky L, Llewellyn HJ. (1988) Lung Lesions: Correlation between Viewing Time and Detection. Radiology 166(2): 451-453.

sk128 - Bacon F. (1620) Novum Organum First Book, paragraphs 40-62. in Encyclopedia Britannica, Great Books of the Western World, Volume 30 - Francis Bacon. 1952, Chicago, London, Ontario.

sk129 - Tom Nugent (Spring 2003) Witness for the Whistleblowers. Valpo, the Magazine of Valparaiso University. pp20-22.

sk130 - Shay J. (2002) Odysseus in America: Combat Trauma and the Trials of Homecoming. Scribner. New York.

sk131 - Gurney JW. (1996) Missed Lung Cancer at CT: Imaging Findings in Nine Patients. Radiology 199(1): 117-122.

sk132 - Davis SD. (1996) Through the Retrospectoscope: A Glimpse of Missed Lung Cancer at CT. Radiology 199(1): 23-24.

sk133 - Gur D, Good WF, Oliver JH, et al. (1994) Sequential Viewing of Abdominal CT Images at Varying Rates. Radiology 191(1): 119-122.

sk134 - Samei E, Flynn MJ, Peterson E, et al. (2003) Subtle Lung Nodules: Influence of Local Anatomic Variations on Detection. Radiology 228(1): 76-84.

sk135 - Nodine CF, Mello-Thomas C, Weinstein SP, et al. (2001) Blinded Review of Retrospectively Visible Unreported Breast Cancers: An Eye-Position Analysis. Radiology 221(1): 122-129.

sk136 - deleted.

sk137 - Samei E, Flynn MJ, Eyler WR. (1999) Detection of Subtle Lung Nodules: Relative Influence of Quantum and Anatomic Noise on Chest Radiographs. Radiology 213(3): 727-734.

sk138 - White C S, Romney B M, Mason A C, et al. (1996) Primary Carcinoma of the Lung Overlooked at CT: Analysis of Findings in 14 Patients. Radiology 199:109-115.

sk139 - Conoley PM, Vernon SW, Burtram SG. (1992) Productivity of Radiologists in the United States by Imaging Technique: a 16-year analysis based upon relative value units. European Journal of Radiology 15(3): 258-263.

sk140 - Rockette HE, King JL, Medina JL, et al. (1995) Imaging Systems Evaluation: Effect of Subtle Cases on the Design and Analysis of Receiver Operating Characteristic Studies. AJR 165(3): 679-683.

sk141- Shiraishi J, Abe H, Engelmann R, et al. (2003) Computer-aided Diagnosis to Distinguish Benigh from Malignant Solitary Pulmonary Nodules on Radiographs: ROC Analysis of Radiologists' performance -- initial Experience. Radiology 227(2): 469-474.

sk142 - Fuhrman CR, Britton CA, Bender T, et al. (2002) Observer Performance Studies: Detection of Single Versus Multiple Abnormalities of the Chest. AJR 179(6): 1551-3.

sk143 - Schwartz J. (May 6, 2004) Simulated Prison in '71 Showed a Fine Line Between 'Normal' and 'Monster". New York Times pA20(L) col 01.

sk144 - Oestmann JW, Greene R, Bourguoin PM, et al. (1993) Chest "Gestalt" and Detectability of Lung Lesions. European Journal of Radiology 16(2): 154-7.

sk145 - Kanter RM. (1977) Men and Women of the Corporation. Basic Books, New York.

sk146 - Livingstone M. (2002) Vision and Art: The Biology of Seeing. Harry Abrams, New York. (This is a fascinating book - see esp.110, 104, 91,77, 66, 64, 46-52, 158

sk147 - Anonymous (1/13/05) Fatal Highway Pileups in 2 States Kill at Least 3. http://www.cnn.com/2005/US/01/12/highway.crash.ap, [last accessed 1/21/05]

sk148 - Ueland B. (1987) If You Want to Write. Graywolf Press. Saint Paul.

sk149 through sk 194 - deleted.

sk195 - Goodall J, Berman P. (1999) Reason for Hope: A Spiritual Journey. Warner Books, New York. excerpted as "When Words Are Abandoned: Feeling the Forest". by permission in Parabola 29(2, Summer): 54-59, with interview by Ryden H.

sk196 - Kincade K. (1998) Computer-aided Readings Improve Diagnostic Accuracy. Diagnostic Imaging May: p11-12.

sk197 - Anonymous. (8/26/03) Space Shuttle Columbia - 16 Minutes from Home. US News and World News Digest. http://usnewslink.com/shuttlecolumbia.htm.

sk198 - Brice J. (1998) What's Right (and Wrong) with Radiology. Diagnostic Imaging December 1998, pp54-67. see esp. Brant-Zawadzki p57.

sk199 - Robert Jasinski Sr., Personal Communication.

sk200- Kundel JL, Nodine CF, Krupinski EA. (1989) Visual dwell indicates locations of false-positive and false-negative decisions. Investigative Radiology 24: 472-8.

sk201 - Aideyan UO, Berbaum K, Smith WL. (1995) Influence of Prior Radiologic Information on the Interpretation of Radiographic Examinations. Academic Radiology 2(3): 205-208.

sk202 - Voltaire, JFMA. (1759) Candide. Random House (©1928), Barron's Educational Series. Great Neck, New York.

sk203 - Park SH, Song HH, Park JM, et al. (1992) Impact of a Primary Reader's Opinion on the Detection of Rib Fractures. Investigative Radiology 27(10): 785-9.

sk204 - Gary Glazer, personal communication.

sk205 - John J. Fennessy, personal communication.

sk206 - deleted

sk207 - Wu YC, Doi K, Giger ML, et al. (1994) Reduction of False Positives in Computerized Detection of Lung Nodules in chest Radiographs using Artificial Neural Networks, Discriminant Analysis, and a Rule-Based Scheme. Journal of Digital Imaging 7(4): 196-207.

sk208 - footnote - "Situation Normal" - "All Fouled Up".

sk209 - Bible. Matt. 18:10-14.

sk210 - Fuhrman Carl P., Britton Cynthia A., Bender Thomas, et al. (2002) Observer Performance Studies: Detection of Single Versus Multiple Abnormalities of the Chest. American Journal of Roentgenology 179:1551-1553.

sk211 - Berlin L. (2000) Malpractice Issues in Radiology: Liability of Interpreting Too Many Radiographs. American Journal of Roentgenology 175:17-22.

sk212 - Seckel A. (2000) The Art of Optical Illusion IllusionWorks/Carlton Books.

sk213 - Gale A. G., Murray D., Millar K, et al. (1984) Circadian Variation in Radiology, in Theoretical and Applied Aspects of Eye Movement Research, eds., Gail A. G., Johnson F. Elsevier Science Publishers BV, North-Holland.

sk214 - Gail A. G., Johnson F, Worthington BS. (1979) Psychology and Radiology, pages 453-460. In Research in Psychology and Medicine. , eds. , Oborne D J, Gruneberg MM, Eiser J R. Academic Press, London/New York.

sk215 - Eastwood, Clint. (1971, 1979) Dirty Harry. VHS, 103 min. Malpaso/Warner Brothers. New York.

sk216 - Wear D. (2002) The House of God: Another Look. Academic Medicine 77(6): 496-501.

sk217 - Slomka J. (1995) A Patient's Guide to Managed Care in the House of God: The Best Care is Less Care. Journal of Family Practice 41(5): 441-442. see also Journal of Family Practice 42(4): 417-418.

sk218 - RW Jasinski, Jr., personal observation.

sk219 - Berlin L. (1994) Reporting the "Missed" Radiologic Diagnosis: Medicolegal and Ethical Considerations. Radiology 192: 183-187.

sk220 - (An old medical adage is - "when you hear hoofbeats, think horses, not zebras" - see House of God.)

sk221 - Krotz D. (1998) Centralized Image Reading Foster High-Volume Output. Diagnostic Imaging December 1998: 13-14.

sk222 - Reit S. (1978) Masquerade: The Amazing Camouflage Deceptions of World War II. New American Library. New York.

sk223 - Lesgold AM. (1984) Acquiring Expertise. In: Anderson JR, Kosslyn SM, eds. Tutorials in Learning and Memory: Essaygs in Honor of Gordon Bower. San Francisco, California: Freeman, 1984. pp31-60.

sk224 - Krotz D. (1998) First CT Reads Often Miss Stroke-Related Pathologies. Diagnostic Imaging December: 21,25.

sk225 - Schriger DL, Kalafut M, Starkman S, et al. (1998) Cranial Computed Tomography Interpretation in Acute Stroke: Physician Accuracy in Determining Eligibility for Thrombolytic Therapy. JAMA 279:1293-1297.

sk226 - Nodine CF, Carmody DP, and Kundel HL. (1978) Searching for Nina, pp 241-258, in Eye Movements and the Higher Psychological Functions., Eds. Senders JW, Fisher DF, Monty RA. Lawrence Erlbaum Associates. Hillsdale, NJ.

sk227 - Robert Jasinski, Jr. - Personal Observation.

sk228 - Shem S. (1978) The House of God. Dell. New York.

sk229 - Wood BP. (1999) Decision Making in Radiology. Radiology 211: 601-603.

sk230 - Fultz P. J., Jacobs CV, Hall W J, et al. (1999) Ovarian Cancer: Comparison of Observer Performance for Four Methods of Interpreting CT Scans. Radiology 212:401-410.

sk231 - Lehr JL, Lodwick GS, Farrell, et al. (1976) Direct Measurement of the Effect of Film Miniaturization on Diagnostic Accuracy. Radiology 118: 257-263.

sk232 - Hessel SJ, Herman PG, Swensson RG. (1978) Improving Performance by Multiple Interpretations of Chest Radiographs: Effectiveness and Cost. Radiology 127: 589-594).

sk233 - Herman PG, Hessel SJ. (1975) Accuracy and Its Relationship to Experience in the Interpretation of Chest Radiographs. Investigative Radiology 10(1): 62-67.

sk234 - Thurfjell EL, Lernevall KA, Taube AAS. (1994) Benefit of Independent Double Reading in a Population-based Mammography Screening Program. Radiology 191: 241-244.

sk236 - Berlin L. (2000) Malpractice Issues in Radiology: Pitfalls of the Vague Radiology Report. American Journal of Roentgenology 174: 1511-1518.

sk237 - Kinard RE, Orrison WW, Brogdon BG. (1986) The Value of a Worksheet in Reporting Body-CT Examinations. American Journal of Roentgenology 147: 848-849.

sk238 - Brenner RJ, Lucey LL, Smith JJ, et al. (1998) Radiology and Medical Malpractice Claims: A Report on the Practice Standards Claims Survey of the Physician Insurers Association of America and the American College of Radiology. American Journal of Roentgenology 171: 19-22.

sk239 - Berlin L. (1999) Malpractice Issues in Radiology: Emerging Technologies. American Journal of Roentgenology 173: 267-270.

sk240 - Loughrey GJ, Carrington BM, Anderson H, et al. (1999) The Value of Specialist Oncological Radiology Review of Cross-sectional Imaging. Clinical Radiology 54: 149-154.

sk241 - deleted.

sk242 - Miles SH. (2004) Abu Ghraib: Its Legacy for Military Medicine. Lancet 364: 725-729.

sk243 - Ross, E - AP new service. (2004) Doctors Collaborated in Prison Abuse, Study Says. Times Reporter (AP). www.timesreporter.com/left.php?ID=32793&r=2, last accessed 8/28/04.

sk244 - Anonymous (2003) NASA Culture Blamed in Shuttle Report. US News & World Digest, 8/26/03. http://www.usnewslink.com/shuttlecolumbia.htm.

sk245 - Carmody DP, Nodine CF, Kundel HL. (1980) Global and Segmented Search for Lung Nodules of Different Edge Gradients. Investigative Radiology 15(3): 224-233.

sk246 - Thomas SB, Curran JW. (1999) Tuskegee: From Science to Conspiracy to Metaphor. American Journal of the Medical Sciences. 317(1): 1-4.

sk247 - Corbie-Smith G. (1999) The Continuing Legacy of the Tuskegee Syphilis Study: Considerations for Clinical Investigation. American Journal of the Medical Sciences 317(1): 5-8.

sk248 - Hricak H, Kahlbhen C L, Scheidler JE, et al (1997) Value of Expert Interpretation in Abdominal Oncologic Imaging: A Multi-Center Study. Radiological Society of North America 1997 scientific programme volume 205 (P), abstract number 361, page 225.

sk249 - Gleick J. (1999) Faster: The Acceleration of Just About Everything. Pantheon Books, New York.

sk250 - deleted.

sk251 - Honore C. (2004) In Praise of Slowness: How a Worldwide Movement is Challenging the Cult of Speed. A.A. Knopf, Toronto, Canada.

sk252 - Conway JK. (1994) True North: A Memoir. A.A. Knopf, New York.pp229-230.

sk253 - Ochberg FM. (1988) Post-traumatic Therapy and Victims of Violence. Brunner/Mazel, New York. see 128-133.

sk254 - Jaffe CC. (1984) Medical Imaging, Vision, and Visual Psychophysics. Medical Radiography and Photography, 60(1): 1-48. (p29 and other great optical illusions).

sk255 - Beam CA, Sullivan DC, Layde PM. (1996) Effect of Human Variability on Independent Double Reading in Screening Mammography. Academic Radiology 3: 891-897.

sk256 - Nolen WA. (1970) The Making of a Surgeon. Random House/Pocket Books, New York. p194-5.

sk257 - Maas P. (1973) Serpico. Viking Press, New York. (see also the movie/video, Serpico, with Al Pacino, Paramount, 1973, 1991, with Al Pacino.

sk258 - Parker N. (1972) Malingering. Medical Journal of Australia 2(23):1308-1311.

sk259 - Larson G. (1984) The Far Side Gallery. Andrews, McNeal, and Parker. Kansas City.

sk260 - Brehm J, Nelson P. (2000) That Others May Live. Crown Publishers. New York. see pp 202-203.

sk261 - Heller J. (1955) Catch 22. Simon and Schuster, New York.

sk262 - Chase A. (2003) Harvard and the Unabomber. WW Norton, New York.

sk263 - Seltzer SE, Judy PF, Adams DF, et al. (1995) Spiral CT of the Chest: Comparison of Cine and Film-based Viewing. Radiology 197(1):73-78.

sk264 - Hillman BJ, Swensson R G, Hessel SJ, et al. (1976) The Value of Consultation Among Radiologists. American Journal of Roentgenology 127:807-809.

sk265 - Louis Pasteur.

sk266 - deleted

sk267 - Herman J. (1992) Trauma and Recovery Basic Books.

sk268 - Magueijo J. (2003) Faster Than the Speed of Light Perseus Books, Cambridge Massachusetts.

sk269 - Jacobs RC, Campbell DT. (1961) The Perpetuation of an Arbitrary Tradition through Several Generations of a Laboratory Microculture. Journal of Abnormal Social Psychology 62: 649-658.

sk270 - Daly CA, Apthrop L, Field S. (1998) Second Round Cancers: How Many Were Visible on the First Round of the UK National Breast Screening Programme, Three Years Earlier? Clinical Radiology 53(1): 25-8.

sk271 - Hilfiker D. (1985) Healing the Wounds: A Physician Looks at his Work. Pantheon Books, New York. see esp. 133-145.

sk272 - Hayes JC. (1998) Barium Enema Struggles Against High-tech Rivals. Diagnostic Imaging May; pp 60-63. see p63, col 1.

sk273 - Paul H. Frank, personal communication.

sk274 - Lens P, van der Wal GA. (1997) Problem Doctors: A Conspiracy of Silence. IOS Press. Amsterdam, Washington.

sk275 - Moorefield JM. (1998) Covering the Radiology Practice. AJR 170: 271-273.