Motion-based camera localization system in colonoscopy videos.

Area of Science:

• Medical imaging
• Gastroenterology
• Computer vision

Background:

• Optical colonoscopy is vital for diagnosing gastrointestinal diseases but suffers from subjective interpretation.
• Inter-observer variability in evaluating colonoscopy findings necessitates objective assessment methods.

Purpose of the Study:

• To develop an automated camera localization system for colonoscopy.
• To enhance the classification of colon anatomical segments using estimated camera motion.

Main Methods:

• A self-training convolutional neural network estimates camera motion from endoscopic videos.
• Camera trajectory and relative location index are derived from motion estimation.
• Anatomical colon segment classification is performed using a colon template and the location index.

Main Results:

• The proposed motion estimation algorithm outperforms existing methods on an external dataset.
• Anatomical region classification achieved an average accuracy of 0.754 in clinical practice.
• The system demonstrated substantially higher accuracy than other location index methods.

Conclusions:

• The developed camera localization system accurately classifies colon segments.
• This automated approach offers a promising solution to reduce subjectivity in colonoscopy interpretation.
• The system has the potential to improve diagnostic and prognostic accuracy in gastrointestinal disease assessment.