Machine learning algorithm diagnoses disease from duodenal biopsies with accuracy comparable to pathologists
By Elana Gotkine HealthDay Reporter
THURSDAY, April 3, 2025 (HealthDay News) — A machine learning algorithm can diagnose celiac disease from duodenal biopsies with accuracy comparable to pathologists, according to a study published online March 27 in the New England Journal of Medicine AI.
Florian Jaeckle, Ph.D., from the University of Cambridge in the United Kingdom, and colleagues developed a novel, accurate, machine learning-based diagnostic classifier to improve celiac disease diagnosis. The model was trained on 3,383 Whole-Slide-Images (WSI) of duodenal biopsies from four hospitals featuring five different WSI scanners and their clinical diagnoses. The model was trained using the multiple instance-learning paradigm in a weakly supervised manner, and it was assessed on an independent test set of 644 unseen scans. The model’s predictions were compared to independent diagnoses from four specialist pathologists.
The researchers found that the model diagnosed celiac disease in the independent dataset with accuracy, sensitivity, and specificity exceeding 95 percent and an area under the receiver operating characteristic curve exceeding 99 percent, indicating that the model can potentially outperform pathologists. Comparing the model’s predictions with those from independent pathologists, the results were statistically indistinguishable between pathologist-pathologist and pathologist-model interobserver agreement (>96 percent).
“This is the first time artificial intelligence has been shown to diagnose as accurately as an experienced pathologist whether an individual has celiac or not,” Jaeckle said in a statement. “Because we trained it on data sets generated under a number of different conditions, we know that it should be able to work in a wide range of settings, where biopsies are processed and imaged differently.”
Two authors disclosed ties to the biotechnology research industry.
Copyright © 2025 HealthDay. All rights reserved.
