Interpretable machine learning-based detection of coeliac disease.

Objective: Coeliac disease, an autoimmune disorder affecting approximately 1% of the global population, is typically diagnosed on duodenal biopsy. However, interpathologist agreement on coeliac disease diagnosis is only 80%. Existing machine learning solutions designed to improve coeliac disease diagnosis often lack interpretability, which is essential for building trust and enabling widespread clinical adoption. We aim to develop an interpretable artificial intelligence (AI) model segmenting key histological structures in H&E-stained duodenal biopsies, generating explainable segmentation masks, estimating intraepithelial lymphocyte (IEL)-to-enterocyte and villus-to-crypt ratios and diagnosing coeliac disease. Methods and Analysis: Semantic segmentation models were trained to identify villi, crypts, IELs and enterocytes using 49 annotated 2048×2048 patches at 40× magnification. Subsequently, IEL-to-enterocyte and villus-to-crypt ratios were calculated from segmentation masks generated by the segmentation model from 172 whole slide images (WSIs), and a logistic regression model was trained to diagnose coeliac disease based on these ratios. Evaluation was performed on an independent test set of 613 WSIs from an independent medical institution. Results: The villus-crypt segmentation model achieved mean precision-recall area under the curve (AUC) of 80.5%, while the IEL-enterocyte model reached precision-recall AUC of 82%. The diagnostic model classified WSIs with 96% accuracy, 86% positive predictive value and 98% negative predictive value on the independent test set. Conclusions: Our interpretable AI models accurately segmented key histological structures and diagnosed coeliac disease in unseen WSIs, demonstrating strong generalisation performance. These models provide pathologists with reliable IEL-to-enterocyte and villus-to-crypt ratio estimates, enhancing diagnostic accuracy. Interpretable AI solutions like ours are essential for fostering trust among healthcare professionals and patients, complementing existing black-box methodologies.