Automated Bleeding Detection and Classification in Wireless Capsule Endoscopy with YOLOv8-X

TL;DR

This work tackles automated GI bleeding detection in Wireless Capsule Endoscopy by deploying a unified YOLOv8-X model capable of both detection and classification. A carefully curated expansion to a 6,345-image WCE dataset, with strict reannotation and an 80/20 train/validation split, underpins training and evaluation. The model achieves strong validation performance, with classification metrics at $96.10\%$ and detection performance of $mAP@0.5=76.8\%$ (IoU $=80.75\%$), reflecting robust localization and sensitivity. Clinically, the approach offers rapid analysis (minutes rather than hours) and consistent performance across diverse bleeding presentations, with code and models publicly available for broader adoption.

Abstract

Gastrointestinal (GI) bleeding, a critical indicator of digestive system disorders, re quires efficient and accurate detection methods. This paper presents our solution to the Auto-WCEBleedGen Version V1 Challenge, where we achieved the consolation position. We developed a unified YOLOv8-X model for both detection and classification of bleeding regions in Wireless Capsule Endoscopy (WCE) images. Our approach achieved 96.10% classification accuracy and 76.8% mean Average Precision (mAP) at 0.5 IoU on the val idation dataset. Through careful dataset curation and annotation, we assembled and trained on 6,345 diverse images to ensure robust model performance. Our implementa tion code and trained models are publicly available at https://github.com/pavan98765/Auto-WCEBleedGen.

Figures (4)

• Figure 1: Architecture of YOLOv8-X showing the Backbone for feature extraction, Neck for feature aggregation, and Head for detection outputs, with multi-scale processing using different stride lengths.
• Figure 2: Examples of successful bleeding detection from our validation dataset, showing the model's ability to identify different types of GI bleeding.
• Figure 3: Detection results on Test Dataset 1 showing three different bleeding cases with confidence scores. The white bounding boxes indicate detected bleeding regions with scores of 0.62, 0.48, and 0.88 respectively, demonstrating accurate detection across varying bleeding patterns and lighting conditions.
• Figure 4: Detection results on Test Dataset 2 showing bleeding regions (marked by white bounding boxes) with confidence scores of 0.87 (left), 0.78 (middle), and 0.59 (right).