Performance of Deep Learning Models in Endoscopic Severity Assessment of Inflammatory Bowel Disease: A Systematic...

Endoscopic severity assessment is fundamental to the management of inflammatory bowel disease (IBD). However, traditional scoring systems are limited by significant interobserver variability. Deep learning has emerged as a promising tool for objective and reproducible analysis of endoscopic images and videos. This systematic review evaluates the performance of deep learning models for endoscopic severity assessment in IBD. We conducted a comprehensive literature search across PubMed, Scopus, Embase, and Web of Science for studies published between 2021 and 2025. We included studies that evaluated deep learning models for endoscopic severity assessment in patients with ulcerative colitis or Crohn's disease. Two reviewers independently performed data extraction and risk of bias assessment using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool. Due to heterogeneity across studies, we conducted a narrative synthesis. Twelve studies met the inclusion criteria. Deep learning models demonstrated high diagnostic accuracy for both ulcerative colitis and Crohn's disease. Architectures ranged from convolutional neural networks to Vision Transformers. For ulcerative colitis, models achieved the area under the receiver operating characteristic curve (AUROC) values up to 0.966 for distinguishing Mayo endoscopic subscore grades. They also achieved weighted F1-scores up to 0.731 for the Ulcerative Colitis Endoscopic Index of Severity scoring. For Crohn's disease, models achieved AUROC values up to 0.971 for stricture detection and accuracy up to 98.6% for lesion classification. One study reported a 97% reduction in images requiring expert review and 87-94% faster reading times. Risk of bias assessment revealed that nine studies had a low overall risk of bias, while three had an unclear risk due to insufficient reporting. Deep learning models consistently achieve high performance in endoscopic severity assessment of IBD. Emerging evidence supports their potential to standardize scoring, improve correlation with patient-centered outcomes, and enhance clinical workflow efficiency. Future research should prioritize external validation across diverse populations, standardized reporting, and prospective evaluation of clinical implementation and long-term outcomes.