Precision Enhancement in Wireless Capsule Endoscopy: A Novel Transformer-Based Approach for Real-Time Video Object Detection

Tsedeke Temesgen Habe ^* Keijo Haataja Pekka Toivanen

• University of Eastern Finland, Kuopio, Finland

Wireless Capsule Endoscopy (WCE) enables non-invasive imaging of the digestive tract but generates vast video data, posing challenges for real-time and accurate abnormality detection. This paper presents a novel approach that uses a transformer-based model, specifically optimized for WCE video analysis, to improve both the precision and speed of object detection. The proposed method integrates the Real-Time Detection Transformer (RT-DETR) to address a unique challenge posed by WCE data, specifically the uncontrolled illumination variations, highly textured backgrounds, and the requirement for high-speed processing to handle vast amounts of video frames efficiently. The RT-DETR model is designed to capture contextual information between video frames effectively, leading to a more accurate detection of gastrointestinal abnormalities.The RT-DETR models achieved better detection performance on Kvasir-Capsule while operating at real-time speeds. RT-DETR-X demonstrated highest precision while RT-DETR-M succeeded in maintaining a practical balance of accuracy and processing speed. RT-DETR-S ran at 270 FPS for real-time analysis making it the most suitable model for WCE video applications. The experimental outcomes validate that RT-DETR can operate effectively for clinical applications through results which show better accuracy as well as computational efficiency than earlier frameworks.