Correctly diagnosing the disease from which the patient suffers is the first step in developing an appropriate treatment plan. Therefore, it is essential to assess the accuracy of diagnostic tests, that is, the likelihood that a reader is correct with respect to an established standard. There are difficulties involved in determining the accuracy of a diagnostic imaging test. It may not be possible to definitively diagnose all cases because of variation in the presentation of the disease, the inaccessibility of the disease within the body and the concomitant invasiveness of the definitive diagnostic procedure, and the difficulty with making a differential diagnosis from other conditions with similar presentations. The diagnostic test under study may itself be the gold standard. There are also potential biases when a panel of experts is used to obtain a definitive diagnosis. Typically, the panel reviews the clinical data as well as the results of the diagnostic procedures to arrive at a consensus diagnosis. The effectiveness of the panel depends on the quality, completeness, and unambiguity of the information which is available to them (which may be quite limited for reasons discussed above) for making their decision. When there is a lack of consensus, an influential member of the panel can sway other members of the group to hisâher view so that the panel becomes equivalent to a single reader with strong convictions (Revesz et al. , Greenes and Begg , Kundel and Polansky , and Hillman et al. ). The difficult cases require a large investment of time of the panel and in the end the diagnosis of the patient is not always resolved. One approach for insuring that the diagnoses of the study cases are accurate is to exclude cases for which the diagnosis is uncertain. However this limits the generalizability of the results of the accuracy study to general practice. It is also considered desirable to include cases which are difficult to assess radiologically but which have definitive diagnoses (based on clinical follow-up or pathological tests). These difficult-to-assess cases ("stress" cases) are believed to more likely detect differences among diagnostic tests because presumably easy cases could be correctly diagnosed by even a fairly insensitive test. As a result, initial studies of the accuracy of diagnostic tests are generally done using carefully selected subsets of all cases.
Online access to SPIE eBooks is limited to subscribing institutions.