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

Histopathology diagnosis presents significant challenges due to the complexity of interpreting tissue samples. Prior research has shown that diagnostic systems in this field have yielded limited success. Established knowledge includes the recognition that variability in diagnostic outcomes is common, especially in areas like cutaneous lymphomas (CL) and borderline lesions. This paper's contribution lies in identifying a gap in understanding how diagnostic systems can better support pathologists. The authors propose that current systems often overlook a critical phase of the diagnostic process. No prior work had resolved how to standardize feature recognition in histopathology. This gap motivated an investigation into the diagnostic process itself. The study aims to address how systems might improve consensus by supporting feature recognition. The findings suggest that diagnostic variability may stem from inconsistent feature descriptions.

The study aimed to explore the diagnostic process in histopathology, focusing on cutaneous lymphomas and related disorders. The authors sought to understand why current diagnostic systems have failed to improve consensus among pathologists. They proposed that the issue lies in the diagnostic process's two-step nature: from observation to feature recognition and from features to diagnosis. The study's motivation stemmed from the need to identify where diagnostic systems could better support pathologists. A key problem is the lack of standardized feature descriptions in pathology reports. The authors hypothesized that supporting the feature recognition step could reduce diagnostic variability. This approach aligns with the broader goal of improving diagnostic reliability in complex cases. The study aimed to test this hypothesis through retrospective and prospective analyses of histopathology reports.

The authors conducted two studies to assess the diagnostic process in histopathology. The first was a retrospective analysis of existing skin biopsy pathology reports. The second was a prospective study involving a panel of pathologists describing 16 skin biopsies using a standard set of descriptors. The retrospective study evaluated the detail and scope of histological descriptions in current reports. The prospective study aimed to assess consensus among pathologists using standardized criteria. Both studies focused on cutaneous lymphomas and borderline lesions. The authors used a two-step model to analyze the diagnostic process: observation to feature recognition and features to diagnosis. The studies compared variability in feature descriptions and diagnostic outcomes. The results indicated a lack of consensus in both feature recognition and diagnostic categorization.

The retrospective study revealed significant variability in the nature and detail of histological descriptions across pathology reports. The prospective study showed a lack of consensus among pathologists regarding both feature descriptions and diagnostic categories. These findings suggest that variability in feature recognition contributes to diagnostic inconsistency. The authors observed that diagnostic systems typically target the second step of the diagnostic process, from features to diagnosis. However, different input into these systems leads to different diagnostic outputs. The studies indicate that diagnostic variability is closely linked to variability in feature recognition. The authors propose that supporting the first step—feature recognition—could improve diagnostic consensus. This conclusion is based on the observed relationship between feature descriptions and diagnostic outcomes.

The authors conclude that diagnostic systems should support the feature recognition step in histopathology diagnosis. They propose that more uniform interpretation of observations could lead to better diagnostic consensus. The findings suggest that current systems often overlook this critical phase of the diagnostic process. The authors emphasize that diagnostic variability may stem from inconsistent feature descriptions. Their model of the diagnostic process highlights the importance of standardizing feature recognition. The studies indicate that different input into diagnostic systems produces different outputs. The authors suggest that improving feature recognition could enhance system performance. These conclusions are based on the observed variability in both retrospective and prospective studies.