P432 Real-time automated assessment of histological disease activity in patients with ulcerative colitis using single wavelength endoscopy technology

Abstract

Abstract Background Assessment of mucosal healing as a key denominator in the treat-to-target strategy for managing ulcerative colitis remains challenging. To address this, objective evaluation of histological disease activity holds promise. Recent offline research shows encouraging results using a new deep-learning convolutional neural network based on single wavelength endoscopy technology (SWE-CAD) (Fujifilm Co, Japan). We aimed to validate the real-time performance of a new bedside prototype SWE-CAD model during standard colonoscopy. Methods A bedside module for real-time use was integrated in the endoscopy room to evaluate histological disease activity in patients with ulcerative colitis with Mayo Endoscopic Scoring (MES) ranging between 0 and 3. Imaging was performed in rectum and sigmoid following a standardized protocol based on white light and SWE (i.e. monochromatic light of 410nm). Biopsies were taken as reference at the center of the imaged region and were scored for the Geboes score (GBS). The SWE-CAD output was displayed on the separate monitor of the bedside module as a blue or red-colored indication, corresponding to histological remission or non-remission, respectively. Each region (2 in rectum and 2 in sigmoid) was simultaneously scored for MES by the endoscopist. Results In a total of 36 patients histological disease activity was automatically scored using the SWE-CAD. On a section level this CAD-system showed an accuracy of 96.4%, corresponding sensitivity was 99.3% and specificity was 85.5%. When differentiating for disease activity as mild, moderate and severe, accuracy was 97.7%, 62.8% and 95.0%, respectively. On a per-patient level, overall diagnostic accuracy remained high with 94.4%, with only 2/36 underestimations when compared to GBS. Conclusion In this pilot trial, we successfully tested a novel CAD-system, utilizing SWE technology, for real-time assessment of histological disease activity in patients with UC. The system demonstrated exceptional clinical accuracy at 94.4% per-patient level, potentially aiding physicians in interpreting subtle endoscopic abnormalities, leading to cost-effective and individualized patient management.