ColonNet: A Hybrid Of DenseNet121 And U-NET Model For Detection And Segmentation Of GI Bleeding

TL;DR

This work tackles automatic detection and segmentation of GI bleeding in Wireless Capsule Endoscopy frames by proposing ColonNet, a hybrid model that combines a DenseNet121‑based ColonSeg branch for classification/detection with a UNetModel for pixel‑wise segmentation. Trained and evaluated on MISAHUB Auto-WCEBleedGen V2 data, ColonNet achieved the top performance among $75$ competing teams, notably attaining about $0.80$ classification accuracy on Test set 2 and strong segmentation metrics. The study demonstrates that DenseNet features are particularly effective for detection in this domain, while the UNet segmentation branch provides meaningful pixel‑level bleeding masks, enabling more precise localizations. The approach has practical potential to reduce clinician workload by rapidly flagging bleeding frames and delineating affected regions, though challenges remain for subtle patches, bubbles, and noisy frames that limit generalization.

Abstract

This study presents an integrated deep learning model for automatic detection and classification of Gastrointestinal bleeding in the frames extracted from Wireless Capsule Endoscopy (WCE) videos. The dataset has been released as part of Auto-WCBleedGen Challenge Version V2 hosted by the MISAHUB team. Our model attained the highest performance among 75 teams that took part in this competition. It aims to efficiently utilizes CNN based model i.e. DenseNet and UNet to detect and segment bleeding and non-bleeding areas in the real-world complex dataset. The model achieves an impressive overall accuracy of 80% which would surely help a skilled doctor to carry out further diagnostics.

Figures (4)

• Figure 1: Architecture of Proposed model
• Figure 2: Architecture of ColonSeg
• Figure 3: Result on test dataset 1
• Figure 4: Result on test dataset 2