Leveraging Interpretable AI for Deciphering Signature Histopathologic Patterns

Leukocytoclastic vasculitis (LCV) and microvascular occlusion (MVO) are distinct histopathologic patterns underlying dermatologic diagnoses of purpura. This study explores the potential of attention-based artificial intelligence (AI) models to enhance diagnostic accuracy and provide interpretable insights in differentiating these conditions, serving as a proof of concept for the application of explainable AI in dermatopathology.

We compared the performance of two attention-based AI models, clustering-constrained-attention multiple-instance learning (CLAM) and attention multiple instance learning (MIL), in analyzing whole slide images of LCV and MVO cases. The models were trained and evaluated using a cohort of 69 biopsies. Performance metrics included precision, recall, accuracy, AUROC, and F1 score. Attention-based heatmaps were generated to highlight diagnostic regions and reveal histopathologic patterns.

The CLAM model outperformed the attention MIL model across all evaluation metrics. Generated heatmaps effectively highlighted key diagnostic regions, including subtle areas of occlusion in the superficial papillary dermis of MVO cases.

This study demonstrates the potential of attention-based AI models to improve diagnostic accuracy and provide interpretable insights in differentiating LCV and MVO. The use of explainable AI and heatmaps offers a valuable tool for pathologists, enhancing their ability to identify and understand subtle histopathologic patterns.