FICE: Fovea-Inspired Closed-Loop Edge-AI System for Enhanced Capsule Endoscopy Imaging

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Wireless capsule endoscopy offers non-invasive gut imaging but faces limitations in power, image quality, and frame rate.
• Existing systems struggle to balance comprehensive imaging with resource constraints.

Purpose of the Study:

• To develop an advanced capsule endoscopy system addressing power, image quality, and frame rate limitations.
• To improve diagnostic accuracy and efficiency in gastrointestinal imaging.

Main Methods:

• Implementation of a Fovea-Inspired Closed-Loop Edge-AI (FICE) system.
• Integration of a lightweight transformer-based model for onboard lesion detection.
• Utilizing a Gated Corner Proposal Network for precise region-of-interest (ROI) localization.

Main Results:

• Achieved 99.88% recall and 99.40% precision for lesion detection.
• Obtained a mean Intersection over Union (mIOU) of 0.804 for bounding box localization.
• Demonstrated a processing speed of 9.58 fps with improved frame rate and pixel density compared to state-of-the-art.

Conclusions:

• The FICE system effectively overcomes capsule endoscopy limitations without compromising diagnostic quality.
• This closed-loop approach mimics human visual attention for efficient resource allocation.
• The system offers significant improvements in imaging performance and data transmission efficiency.