Answers are grounded in this paper and cite their sources.
Retrospective study demonstrates high diagnostic accuracy of deep learning for celiac disease in adults, highlighting potential to reduce manual pathology workloads.
BACKGROUND & AIMS: Duodenal biopsy combined with tissue transglutaminase IgA (tTG-IgA) antibodies is the gold standard for celiac disease (CeD) diagnosis. Issues with biopsy include artifacts, poorly oriented specimens, interobserver reliability, and pathologist shortages. Artificial intelligence may address these challenges. We aimed to determine if deep learning-based image analysis can be used on duodenal biopsies to accurately diagnose CeD. METHODS: The study population included individuals with a duodenal biopsy and tTG-IgA antibody from 2008 to 2021. The inclusion criterion was adults ≥18 years. The exclusion criterion was those following a gluten-free diet. Patients were grouped into 5 categories: normal, CeD, increased intraepithelial lymphocytes (IELs) only, IELs with a positive tTG-IgA antibody, and seronegative villous atrophy. Whole-slide images were obtained. Clustering-constrained attention multiple-instance learning was the deep learning algorithm used. The dataset was split into training, validation, and testing splits with 10-fold cross-validation. RESULTS: A total of 2406 whole-slide images from >2100 patients were included: 1290 normal, 414 CeD, 450 IELs only, 89 IELs with a positive TTG-IgA, and 163 with seronegative villous atrophy. For all categories, the average area under the receiver operating characteristic curve (AUC) was 0.872 with an accuracy of 74%. The model reliably differentiated CeD from normal with an average AUC of 0.992 and an accuracy of 97%. To distinguish CeD from seronegative villous atrophy, the average AUC was 0.810 with an accuracy of 76%. CONCLUSIONS: This deep learning model can discriminate between CeD and normal histology. It can also separate patients into 5 classifications, representing a promising tool that would benefit from further optimization to improve diagnostic accuracy, reduce the manual pathology review burden, and enhance resource management.
No takes yet. Share an insight, caveat, or question.
Jansson‐Knodell et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: