ColonNet: A Hybrid of DenseNet121 & U-Net Models for Detection and Segmentation of GI Bleeding

This study presents a combination of deep-learning model for automatically detecting and classifying gastrointestinal bleeding in frames extracted from Wireless Capsule Endoscopy (WCE) videos. The dataset was made available as part of a challenge organized by the MISAHUB team - the Auto-WCBleedGen Challenge Version V2. Our model achieved the best performance among the 75 teams that took part in the competition. It utilizes a CNN-based approach, specifically DenseNet and U-Net, to detect and segment bleeding and non-bleeding regions in the complex real-world dataset published in the Auto-WCBleedGen Challenge. With an overall accuracy of 80%, the model can significantly aid skilled doctors in further diagnostic procedures.