|Home | About | Journals | Submit | Contact Us | Français|
When interpreting radiographs, practitioners often express apprehension about missing a lesion. Most significant errors are indeed false negative interpretations, but false positive and misclassifications also occur (1). Factors contributing to this decreased confidence are the inherent subjectivity of radiographic interpretation, but also the high number of sources of error that must be considered. Gaining an understanding of these errors is essential to minimizing their occurrence and improving diagnostic accuracy.
Radiographic images are black-and-white 2-dimensional representations of disease, and interpreting them has inherent subjectivity, which is often perceived as a source of error. In fact, a more appropriate term is variation (2). Variation occurs in 2 forms: interobserver and intraobserver. Interobserver variation occurs when different radiographic interpreters do not arrive at a consensus on the radiographic findings and/or diagnosis, but where neither opinion is wrong. Intraobserver variation occurs when the same interpreter does not arrive at the same findings/diagnosis for different cases presenting the same radiographic appearance of disease. Variation can occur regarding the detectability of a lesion (for example, cardiomegaly — present or not), as well as in grading the severity of a lesion (cardiomegaly — mild or moderate) (2). Diseases where inter- and intraobserver variation is high are generally considered more difficult to confidently pose a radiographic diagnosis (2). For example, high interobserver variability is associated with a lower degree of certainty as to the presence of a bronchial pattern in cats when this pattern is less severe (3). Variability is important to consider as it can modify diagnostic sensitivity and specificity depending on the observers performing the interpretation (4).
Conversely, a radiographic error implies that a mistake has been made (2). Errors can be grouped into several categories: technical, perception (search), and analysis (judgement/cognition/decision) (2,5,6).
Technical errors are those related to the creation of the radiographic image and include: number and type of radiographic views, patient positioning, parameter selection, inappropriate use of collimation and grids, processing errors, and other image artifacts (such as, fog, extraneous objects). Although some types of technical errors are reduced with digital radiography, they are not eliminated, and a series of errors inherent to digital radiography exists (7,8). Among others, underexposure (insufficient X-ray exposure; quantum mottle or “graininess”) and overexposure (excessive X-ray exposure; saturation and planking) can still occur. In addition, consideration must be given to the medium on which the image is being viewed. For digital images, a properly calibrated, high-luminance, high-resolution gray-scale screen is essential (9) (see Radiology Corner, Can Vet J 2008;49:1122–1123). It goes without saying that any image, digital or analog, on which the technique is suboptimal should be repeated to ensure that technical errors not lead to interpretation errors (Figure 1).
Perception is defined as the awareness of the content of a radiographic image (10). Perception errors occur when searching for a lesion. Among these, visual perception errors occur in the translation of a structure to 2-dimensions and include summation, silhouetting, Mach lines, magnification and distortion, and often result in false positive diagnoses (5,11,12). Common examples include the impression of an abdominal mass where intestinal loops are summated, or mistaking the superimposition of the radius and ulna for a fracture (Mach line). The transfer from analog to digital radiographs requires an adaptation period, for example, overinterpretation of pulmonary interstitial patterns can occur due to edge enhancement of fine pulmonary structures. Other perception errors create false negative diagnoses (5). Search errors are common and most often result from incomplete evaluation of a radiograph. It is natural to focus on the center of the image and forget to systematically evaluate the structures on the edge (Figure 2). Practice helps: expert radiologists develop visual search patterns that are more efficient and complete than novice viewers (5). Visual overload and structured noise also contribute to search errors. A large number of images (such as, an equine prepurchase examination, an upper GI examination) causes visual overload. Structured noise is caused by normal anatomy obscuring a lesion. For example, correctly evaluating all of the structures on spinal radiographs is difficult, owing to the large number of overlapping structures. Detecting a lung nodule amongst the vessels of the pulmonary hilus is also difficult.
The following steps should be taken to reduce perception error. 1) Perform a complete radiographic examination appropriate for the anatomic region of interest. It is also important to understand the factors that contribute to the radiographic appearance of different structures and their anatomic variations. 2) Use a systematic (step-by-step) approach to reading radiographs to help train visual search behavior such that the image is consistently searched. 3) Change the viewing distance and use a hot lamp. With digital images, vary the zoom, pan, black/white inversion and window/level parameters. 4) The viewing environment greatly contributes to perception errors. Maintain a quiet environment by using a sufficient number of viewboxes/monitors (at least 2), reducing ambient lighting, reducing veiling glare on computer monitors, and minimizing distractions to improve visual acuity and concentration (2).
The second component of radiographic interpretation is analysis, or establishing the meaning of perception. Visual tracking technology shows that a radiographic sign may be seen, but is not recognized as being abnormal (underinterpreted) (5). Conversely, overinterpretation often stems from a lack of knowledge regarding the radiographic appearance of disease, or normal anatomic variations (1). For example, overinterpretation by veterinary students and recent graduates may occur frequently because relatively few normal radiographs are seen during classes and clinical rotations (13). Lack of information can also lead to analysis errors, in particular, by not making comparisons with previous radiographic examinations. Accurate clinical history increases diagnostic sensitivity, but misleading clinical information and patient history can bias the interpreter, leading to an increase in false positive diagnoses (14). For example, in a dog with a heart murmur and a suggestive history, one is probably more likely to think that at least mild cardiomegaly is present (Figure 3). Satisfaction of search is another result of clinical bias towards a diagnosis, defined as when an observer stops image interpretation once a meaningful lesion has been detected (2,10). Other types of bias occur when the practitioner locks too quickly into a hypothesized diagnosis (anchoring bias, tunnel vision) or is influenced by the undesirability of the diagnosis (regret bias) (15).
The following steps should be taken to reduce analysis errors. 1) Keep up-to-date and use current references on anatomic variations and radiographic signs of disease. 2) Review patient history carefully; in particular, compare current examinations with any previous radiographs. Try interpreting the radiographs first by ignoring the history; then use the history judiciously in order to rule-in or out the diagnostic hypotheses formulated during the initial search (16). 3) A “dual read” by 2 different interpreters increases the chances of detecting small lesions (2). It also helps reduce clinical bias when the second reader remains unaware of the clinical history. 4) Psychological (fatigue, frustration, boredom) and external factors (unfavorable working environment, noise) contribute to both perception and analysis errors and should be minimized (6).
Use of this article is limited to a single copy for personal study. Anyone interested in obtaining reprints should contact the CVMA office ( gro.vmca-amvc@nothguorbh) for additional copies or permission to use this material elsewhere.