Computer-based diagnostic support systems in histopathology: what should they do?

In many fields in histopathology diagnosis making is notoriously difficult. We explored the diagnostic process in the area of CL and related disorders. The field of cutaneous lymphomas (CL) and borderline lesions is complex and representative for the type of diagnostic problems encountered. A review of diagnostic support systems in diagnostic pathology revealed that the usefulness of these systems has been disappointing. We contribute this to the fact that these systems target only part of the diagnostic process. A tentative model of diagnosis making in pathology is presented. This model assumes a two-step process, from observation to feature recognition and from features to diagnosis. In a retrospective study of existing skin biopsy pathology reports we assessed detail and scope of the histological descriptions. In a second, prospective study, a pathologist panel described a set of 16 skin biopsies using a standard set of descriptors. The retrospective study showed a large variability in the nature and details of described features, whereas the prospective study showed lack of consensus regarding both feature descriptions and diagnostic category. Both studies provide an indication that lack of consensus in feature recognition may be an important contributor to lack of consensus at the diagnostic level. Diagnostic expert systems target the step from feature to diagnosis. Evidently different input into such systems produces different output. We conclude that support of the feature recognition step can contribute to better diagnostic consensus because of more uniform interpretation of observations.