Precision enhancement in wireless capsule endoscopy: a novel transformer-based approach for real-time video object...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Wireless Capsule Endoscopy (WCE) generates large video datasets, posing challenges for real-time abnormality detection.
• Traditional methods struggle with WCE's variable illumination, complex textures, and high processing demands.

Purpose of the Study:

• To introduce and evaluate a novel approach for WCE video analysis using Real-Time Detection Transformer (RT-DETR).
• To assess the performance of different RT-DETR variants in detecting gastrointestinal abnormalities.

Main Methods:

• Utilized transformer-based object detection (RT-DETR) optimized for WCE video analysis.
• Evaluated three RT-DETR variants (Small, Medium, X-Large) on the Kvasir-Capsule dataset.
• Assessed models for contextual information capture and handling of variable image conditions.

Main Results:

• RT-DETR-X demonstrated the highest detection precision.
• RT-DETR-M provided a balance between accuracy and speed.
• RT-DETR-S achieved 270 FPS, enabling real-time processing.

Conclusions:

• The RT-DETR framework significantly improves precision and real-time performance for WCE abnormality detection.
• This technology holds potential for faster and more accurate gastrointestinal diagnoses.
• Future research will focus on optimization and integration into clinical systems.