AI Reads Pediatric IBD Biopsies: Cellular and Molecular Gastro and Hepato | June 2026

• Histopathology remains a cornerstone for diagnosing pediatric inflammatory bowel disease, but differentiating Crohn’s disease from ulcerative colitis can be challenging, especially in young children.
• This study evaluated whether artificial intelligence could automatically identify key histological features from routine endoscopic biopsy whole-slide images.
• Researchers developed three computer vision models using convolutional neural networks and multiple-instance learning, a technique particularly useful when precise lesion-level annotation is unavailable.
• The first model differentiated normal versus abnormal tissue with excellent performance, achieving an AUC of 0.91.
• The second model identified active inflammation, achieving an AUC of 0.92.
• The third model detected chronic architectural distortion, achieving an AUC of 0.93.
• Overall model performance was strong despite using a weakly supervised learning approach, suggesting that meaningful pathological signals are present within routine whole-slide images.
• The study demonstrates that AI can recognize inflammatory and structural patterns typically assessed by gastrointestinal pathologists.
• Importantly, the models were developed from real-world pediatric endoscopic biopsy specimens rather than highly curated research datasets.
• The findings support the concept that AI could function as a pathology decision-support tool rather than replacing pathologists.
• Potential future applications include:

Automated screening of biopsy slides

Standardization of histological interpretation

Reduction of interobserver variability

Faster reporting workflows

Objective disease activity assessment

• For pediatric IBD, where early diagnosis can significantly influence treatment selection and long-term outcomes, such tools may become increasingly valuable.
• The study does not yet provide direct Crohn’s disease versus ulcerative colitis classification, but establishes the foundation for future disease-specific diagnostic algorithms.
• Larger multicenter validation studies and integration with clinical, endoscopic, and molecular data will be required before routine clinical implementation.

Bottom line: This study demonstrates that artificial intelligence can accurately identify abnormal tissue, active inflammation, and chronic architectural changes in pediatric IBD biopsy slides, highlighting the growing role of computer vision as a diagnostic support tool in gastrointestinal pathology.