Inflammation Detection Using Ensemble Endoscopic Multimodal Assessment in Inflammatory Bowel Disease

Inflammatory bowel diseases (IBD), comprising Crohn's disease (CD) and ulcerative colitis (UC), present chronic inflammatory gastrointestinal disorders with substantial implications for patients' quality of life. Traditional endoscopic evaluation remain pivotal for monitoring and managing IBD. Recent advancements in Virtual Chromoendoscopy (VCE) technologies, such as Flexible Spectral Imaging Color Enhancement (FICE) and iScan with digital enhancement, offer noninvasive alternatives for evaluating gastrointestinal diseases. While overcoming some limitations of White Light Endoscopy (WLE), these technologies introduce challenges related to scoring systems and deep learning algorithm training due to the qualitative nature of existing endoscopic scores. To address these challenges, we propose a combination of a generative (cycleGAN) and an ensemble model that integrates assessments from white light endoscopy (WLE), and generated Virtual Chromoendoscopy (VCE) to enhance inflammation detection and prediction. The ensemble model aims to combine the strengths of diverse modalities, providing a holistic understanding of a patient's inflammation status. Experiments demonstrated in this paper show that by integrating endoscopic findings with other modalities using an ensemble learning method can greatly improve the accuracy of prediction of IBD.