Diagnosis and Severity Assessment of Ulcerative Colitis Using Self Supervised Learning

Ulcerative Colitis (UC) is a chronic inflammatory bowel disease that causes ulcers in the large intestine and rectum. The growing prevalence of UC, combined with a shortage of gastrointestinal physicians, strains the healthcare system and delays patient care. UC is diagnosed through colonoscopy, with severity assessed using the Mayo Endoscopic Score (MES), which ranges from 0 (no inflammation) to 3 (severe inflammation). AI-based convolutional neural networks (CNNs) have shown promise in diagnosing UC and assessing severity by framing the analysis as a multi-class classification problem. However, these CNNs rely on large annotated datasets, which are time-consuming and costly to create. To address this, this research explores self-supervised learning (SSL) frameworks that can train on unannotated datasets. Results show that SSL methods like SwAV and SparK outperform supervised learning models on the LIMUC dataset, the largest publicly available annotated dataset of colonoscopy images for UC.