 |
 |
 |
|||
|
|||||
 |
Visual Perception And Radiographic InterpretationMarc Papageorges, DVM,MS,PhD,DACVR
FOCAL POINTThe high error rate in radiology appears to result from the complex mechanisms of visual perception. KEY FACTS
ABSTRACTInterpretation errors are common in radiology and the causes are still debated. Perceptual mechanisms appear to be responsible for a large proportion of mistakes made by both neophytes and trained radiologists. Erroneous perception of familiar contours can be triggered by unrelated opacities. At the opposite, visual information can not induce a specific perception if the observer is not familiar with the concept represented or its radiographic appearance. In addition, the area of acute vision is smaller than we recognize. Familiarity with perceptual mechanisms and the limitations of our visual system as well as multiple readings may be necessary to reduce interpretation errors. INTRODUCTIONSeeing is believing. We consider that what we "see" is an exact representation of the physical world; but is this confidence justified? Visual errors are common. Most errors, however, remain undetected because our dominant sense, as any exercising power that does not answer to a higher authority, is blind to its own weaknesses.[1,2] The purpose of this article is to demonstrate that when we look at radiographs and other medical images, it may be a good idea to keep an eye on our visual system. VISUAL DISABILITIESDarkness adaptation and peripheral glare are familiar visual disabilities that can be minimized using optimal viewing conditions.[3,4,5] Our poor performance in the estimation of length, angle, and size (Fig 1a-1c) is less recognized,[1,6] but this handicap can easily be overcome using measuring devices. Other distortions of visual information such as the Mach phenomenon (Fig 2) and contrast background effect (Fig 3) can not be eliminated, but awareness of their existence reduces interpretation errors.[7,8] The most common and flagrant visual inaccuracies, however, are difficult to identify and correct because they seem to involve perceptual mechanisms developed by our brain to make sense of sensory signals.[1,2,9] Many believe that the high error rate (as high as 20 to 30% for subtle lesions) reported in (human) radiology is rooted in our perceptual mechanisms.[10-12] This hypothesis is supported by the fact that improving the quality of radiographs or digital manipulation of digital images do not seem to reduce the error rate significantly. Therefore, acknowledgment of the limitations of our visual system and familiarity with perceptual errors may be necessary to reduce the inconsistency of interpretation in radiology. To better understand perceptual mechanisms and their implication in radiology, we will review the vision and perception dichotomy, subjective contours, multiplicity of perceptions, and visual search.
Figure 1 — (A) Which line is longer? Actual measurements may be surprising. (B) It is very difficult to align partially covered lines. Using a ruler or a straight paper edge may be surprising. (C) Which nodule is bigger? The background influences size perception. (D) Apparent cardiomegaly on radiographs obtained durning full expiration is a very common perceptual error. THE DIFFERENCE BETWEEN VISION AND PERCEPTIONIt is widely accepted that visual signals do not produce integral images in our brain: transmission and reconstruction of integral visual information would require too much “computing power”. Visual signals are rather compared to mental images--called percepts--stored in memory. Those mental images are constantly built and modified based on experience.[1,9,14]. Apparently, a meaningful image is perceived only when visual signals induce the emergence of a percept in our brain (Fig. 4).[1,9,14] Therefore what we “see” are mental images (percepts) brought to consciousness and touched up by visual signals for shape, size, color, distance, motion, etc... 
Figure 2 — Mach lines (Mach phenomenon), which iare persistent optical illusions produced by a lateral inhibition neural network within the retina, appear as false white or black lines at sharp boundaries. This illusion enhances boundary detection but can mimic lesions (fractures) or anatomic structures (bowel wall, bladder wall).
Figure 3 — The eye is a poor judge of absolute brightness. The brightness of the small squares is identical. The contrast background effect is an extension of the Mach phenomenon.
Take a few minutes to look at various objects around you. Note how our vision provides information about physical properties that our eyes cannot detect (weight, flexibility, temperature--even how it would taste or smell given the opportunity). This non-visual information arises from the same percepts stored in memory. The theory of percepts explains how we can “see” without seeing: the vivid images "seen" during dreams (or hallucinations) are percepts that emerge spontaneously without the control of visual signals. Therefore, one could say that the purpose of our senses during waking hours is to make the "dream" as compatible as possible with physical reality. As we will see, this is not always easily done with radiographs. SUBJECTIVE CONTOURSRadiographs are two-dimensional images showing the contour of objects; they are images of shadows. Unlike the familiar shadows surrounding us, however, radiographs display a summation of several types of shades. This added complexity gives the false impression that radiographs are more than a collection of shadows. Even experienced radiologists occasionally forget that radiographs show only the contour of tissues and objects (Fig. 5). This natural oversight leads to numerous perceptual errors. Because the perception of contours may be the dominant visual mechanism at work during radiographic interpretation, it deserves our consideration. Contours are perceived when there is an abrupt change in luminance (or color) between adjacent areas.[1] Under certain circumstances, however, the perception of contours occurs where there is no change in brightness. Such "subjective contours" are the result of mental completion of partial lines to form familiar or expected shapes.[13-15] Here is one example: when viewing three dots, most of us perceive a triangle. If there are four dots, a square or a rectangle is "seen" (Fig. 6a). The objective visual information, however, is limited to three or four dots independent from one another. The dots could as well be part of a circle or a complex figure (Fig. 6b). Yet, we perceive a triangle and a square. This is a first hint about the fundamental difference between vision (what the eye sees) and perception (what the brain sees), and how previous knowledge (of triangles and squares) influences perception. 
Figure 4 — An image is perceived only when visual signals induce the emergence of a percept in the brain.
Figure 5 — Examination of familiar objects serves as a reminder that radiographs are images of shadows that show only contours.
Figure 6A
Figure 6B
Figure 6 — (A) Seven black dots are perceived as (left) a triangle and (right) a square. (B) The same dots could as well be parts of (left) a circle and (right) a complex figure.
Figure 7 is another example. The objective visual information is limited to three irregular black patches with a triangular defect. When viewing this figure, however, most of us perceive a large white triangle partially covering three black patches. Where is the white triangle coming from? The triangle has no physical reality other than in our brain. When the visual signals are compared to memorized mental images (percepts), our perception mechanisms conclude--based on our previous experience in the tri-dimensional world--that the most likely explanation for the visual stimulation is "a white triangle partially covering three black patches", and this most compatible percept emerges in consciousness.[1,14] The mental image can be so strong that the triangle appears lighter than the "background", creating definite borders between the patches, even though the retinal stimulation is the same in both regions. Would you perceive a triangle if you had never seen one before, assimilated the concept, and stored it in memory? Based on the experience of people who were blind at birth and cured later, the answer is no.[1] Would you perceive a megaesophagus on radiographs if you had never seen one before with your mind's eye and assimilated the concept? When the brain has been sensitized and expects a percept, limited or even unrelated information can induce the mental image (Fig. 8a). This phenomenon explains many radiographic errors. If the amount of visual signals supporting the mental image increases, the compatibility with the percept and the perception becomes stronger and more stable (Fig. 8b). The opposite is equally true: when the brain does not expect or desire a certain discovery, sufficient information might not induce the perception. On radiographs, subjective contours often appear as abdominal masses or false kidneys created by superimposed bowel loops, or as pulmonary lesions produced by adjacent areas of increased or decreased opacity. Superimposition of pulmonary blood vessels and ribs regularly mimics pulmonary nodules, particularly when blurred by motion or diffuse pulmonary opacifications.[13]
Although it may be very persistent, a subjective contour is not an irreversible illusion: close examination of the radiograph usually reveals that the contour is produced by unrelated opacities. But an effort has to be made; our desire--or lack of desire--to find a lesion should never prevent us from performing a close examination of the contour every time an abnormality is "seen". MULTISTABILITY OF PERCEPTIONMultistability of perception is another phenomenon that illustrates the fundamental difference between vision and perception. A classic example of this is provided in Figure 9. Such an image is called ambiguous or reversible because it may be perceived in different ways.[1,2,16] If you look at figure 9, you will most likely perceive a translucent cube seen from above. If you stare at it long enough, however, the image suddenly changes and a similar but different cube appears, seen from below this time. Once both perceptions have been triggered, they alternate as if two different images were interchanged.
Figure 9 — This image can trigger the perception of two different cubes.
Figure 10 — Different percepts triggered by the same image. If rabbiits and gulls are familiar to the obserer, the perceptions alternate. Note that both are never "seen" at the same time.
Most of us perceive the "cube from above" first because based on experience, one is more likely to find a cube resting on a surface than one floating weightlessly in mid air. Note that only one perception is "seen" at any moment: the dominant one drives the other out of consciousness. Can you see the second image in figure 9? It may help to turn the page upside down. Figure 10 is another example of different mental images triggered by the same visual signals. If one has never seen a gull, the only image perceived would be that of a rabbit, and vice versa. If both animals are familiar to the observer, the perceptions alternate. Again, note that both are never "seen" at the same time. Therefore, although a radiographic shadow may represent two possible alternatives, only one can be perceived at any given time. Recent brain mapping experiments suggest that perceptions arise from large scale quantum changes in the collective activity of millions of neurons.[17] The plurality of perception is significant when we have to make a diagnosis under conditions of uncertainty,[12] or when we interpret radiographs for evidence of progression, regression, or stability of disease.[10] Perceptual errors caused by progressive situations are illustrated by Figure 11. Such errors are common in radiology because most diseases progress gradually from a normal to abnormal state. Whether cardiomegaly, splenomegaly or hepatomegaly is present or not are good examples. In the mid portion of the spectrum, preconceptions have a definite effect on the percept triggered by the visual information. In such circumstances, the observer's attitude has an influence on interpretation. For example, a pessimistic attitude created by having missed a pulmonary metastasis in a previous patient would make a clinician more likely to call an ill-defined shadow a nodule. Such an attitude would increase the number of true lesions detected, but at the same time would increase the number of false positive diagnoses.[12] On the other hand, a more optimistic attitude would decrease the number of false positive diagnoses, but would also decrease the number of true positive findings. 
Figure 11 — Progressive figures show how preconceptions can influence perception, particularly in conditions of uncertainty. What do the "in between" figures represent?.
What is the right attitude to adopt? The ideal attitude should be determined by the number of false positive and false negative diagnoses we are willing to accept.[12] The decision should be affected by the clinical significance of a positive finding and the reliability and invasiveness of other tests likely to be used to confirm or disprove the radiographic diagnosis (first, do no harm). Even with the best intentions it is impossible for any of us to maintain a constant attitude or be aware of all possible implications over a long period of time. Knowledge and emotions change continuously and both influence perception: we always see more easily what is set or expected to be seen, and we are less likely to see what has no apparent connection with, or contradicts, the idea we have in mind when we examine radiographs. This influence is unavoidable because an act of perception is not simply the copying of an incoming stimulus: it is the fusion of sensory messages with past experience and expectations.[17] If as it has been suggested the plurality of perception is indeed a major explanation for the interobserver and intraobserver error rate of 20-30% in radiology,[10,12,18] multiple readings may be the only way to reduce the inconsistency of interpretation[10]. VISUAL SEARCHWe all know that radiographs should be searched thoroughly, but how complete is a "complete" search? Studies have shown that large areas of radiographs are unexplored, even when interpreted by experienced radiologists.[19-21] For the most part, this is not the result of carelessness but the consequence of complex visual mechanisms, many of which are not under our conscious control.[19] The area of acute vision is a lot smaller than we recognize. The fovea, the small central area of the retina densely packed with visual receptors, is the only structure of the eye capable of detailed vision. During interpretation of an image, eye movements temporarily fix the fovea on different locations to obtain detailed information. The indistinct information gathered by the peripheral vision guides the eye movements towards areas of motion or high contrast via oculomotor reflexes and occasionally activates additional percepts.[19] As we all know from experience, a radiographic abnormality is easier to detect if it stands out from the background (high contrast). The explanation is simple: the abnormality is more likely to be detected by the peripheral vision and attract visual attention (Fig. 12a). On the other hand, if the lesion is too subtle to attract visual attention, it may be very difficult to find (Fig. 12b). We rarely realize it, but our eyes constantly move from fixation to fixation in a succession of jumps. Each fixation lasts about one third of a second. Because vision is blurred during eye movements, we effectively see only when the eye is still, which can be as little as 30% of the time.[19,22] We are unaware of this intermittent input of information because the process is efficiently smoothed in our visual perception centers. This smoothing of the visual input along with the filling of the peripheral field by mental images is dangerous in radiology: it gives us the impression of seeing in detail a large area at one glance, when the actual area of acute vision is only about the size of a dime at a distance of 60 cm.[1,19] The apparently simple act of reading (what you are doing effortlessly right now) displays the effectiveness and complexity of the filling mechanisms. As an experienced reader, you probably make three or four jumps per line. As you read, you have the impression of seeing all characters clearly, but you actually see in detail only six to twelve per line. To convince yourself, try the following experiment: fix your eyes on any word of this page for a few seconds, and notice how most of the page turns into a blur. Before long you see clearly only two or three letters. This very small surface of the page is the only one that is actually seen in detail at any given time. You might be able to identify one or two short words on either side of the fixation, but this identification is based on inference from word length and grammatical construction rather than from seeing the actual letters. As soon as you start reading again, you have the impression of seeing every letter of every word again. The extrapolation and smoothing procedures are performed by our brain that fills-in the missing information using our knowledge of the English language (as the triangle is filled-in using our knowledge of triangles in figure 7). The faster we read, the longer the jumps and the more information is filled-in by the mind.[19] This is why we have such a hard time finding typos in a familiar text. Struggling with a foreign language (spoken or written) really makes one appreciate the influence of previous knowledge on perception. The same phenomenon occurs with radiographs: the more familiar we are with the images, the more comfortable we become and the quicker we make sense out of them; but at the same time, the more information is filled-in by the mind.[19] Another interesting aspect of the eye jumps is that they are not completely under conscious control: we cannot change the length or direction of a jump once it is initiated. Therefore, we cannot be certain about what we actually see and what our brain fills-in.[19,22] The best we can do to control a search pattern is to concentrate on a certain area; the details of the search are always left to an unconscious part of our visual system [19]. 
Figure 13 —Counting how many times the letter "F" is used in one sentence proves more difficult than anticipated. Finding all peritnent lesions on one radiograph is likely to be a more arduous task.
This is difficult to believe because it contradicts our perceived experience. The following experiment should challenge the confidence of skeptics: try counting how many times the letter "F" is used in figure 13. Do not read any further without giving it a serious attempt. Most of us will find four or five. Try again. The answer is seven. If you cannot find them all after multiple attempts, try reading backwards (which suppresses the "fill-in" mechanisms and reduces the length of the eye jumps). CONCLUSIONKnowing that we see in detail only a surface the size of a dime, that it is impossible to accurately control where the "dime" will rest on the image, that we see only one third of the time we are actually "looking", and that it is impossible to know how much we actually see and how much information our brain fills-in makes it easier to understand how lesions can be missed, and how psychological factors like interest, motivation, training, and emotions can modify the way we search, perceive, and interpret radiographs. This review may give the impression that radiology is less reliable than other fields of medicine, but it seems that other diagnostic procedures are affected as well: a similar error rate was found in the ability to detect and describe lesions such as heart murmurs.[10] Familiarity with the limitations of our senses and the complexity of perceptual mechanisms appears to be a positive asset for all clinicians. Hopefully, this review will stimulate you to reevaluate the way you read radiographs, and make you ponder about the way we perceive in general. BIBLIOGRAPHY
CONTINUING EDUCATION QUIZ
ANSWERS CHAPTER VISUAL PERCEPTIONQuestion 1: c)
|
|||||||||||||
