Comparative assessment of feature extraction methods for visual odometry in wireless capsule endoscopy

Area of Science:

• Medical Imaging
• Gastroenterology
• Computer Vision

Background:

• Wireless capsule endoscopy (WCE) is a non-invasive tool for examining the gastrointestinal (GI) tract.
• Accurate capsule localization is crucial for precise medical interventions like biopsies and polyp resections.
• Current localization methods rely on external sensors or transit time estimations.

Purpose of the Study:

• To evaluate the feasibility of capsule localization based entirely on visual features, eliminating the need for external sensors.
• To determine the optimal visual feature extraction technique for accurate capsule motion estimation.
• To assess the trade-off between localization accuracy and computational efficiency.

Main Methods:

• Utilized a motion estimation algorithm to measure capsule distance and rotation from acquired video frames.
• Conducted an extensive comparative assessment of state-of-the-art visual feature extraction techniques.
• Employed a publicly available dataset for evaluating localization performance.

Main Results:

• Demonstrated the feasibility of sensor-less capsule localization using visual features.
• Identified a trade-off between minimizing localization error and computational efficiency.
• Achieved a localization error approximately one order of magnitude higher than the minimum for computationally efficient techniques.

Conclusions:

• Visual feature-based localization is a viable alternative to traditional methods in WCE.
• Selecting an appropriate feature extraction technique is key to optimizing localization accuracy and computational cost.
• A compromise between minimal error and computational efficiency is acceptable for current practical applications.