Automated bleeding detection in capsule endoscopy videos using statistical features and region growing

Area of Science:

• Medical Imaging
• Gastroenterology
• Computer Vision

Background:

• Wireless Capsule Endoscopy (WCE) enables small intestine visualization.
• Automated analysis of WCE videos is crucial for identifying critical findings.
• Manual annotation of WCE data for training is time-consuming and inconsistent.

Purpose of the Study:

• To develop a supervised method for automated detection of bleeding regions in WCE images.
• To propose a semi-automatic algorithm for efficient training data creation.
• To optimize feature selection for enhanced detection performance.

Main Methods:

• Image regions characterized using statistical features from RGB color space histograms.
• A semi-automatic region-annotation algorithm developed for training data generation.
• Exhaustive analysis of feature combinations to identify optimal feature sets.
• Segmentation methods applied to extract image regions.
• A trained neural network used for classifying bleeding and non-bleeding regions.

Main Results:

• The proposed method effectively detects bleeding regions in WCE frames.
• The semi-automatic annotation tool improves efficiency in creating training datasets.
• Optimized feature sets enhance the performance of the bleeding detection algorithm.
• Neural network accurately recognizes patterns associated with bleeding.

Conclusions:

• The developed supervised method offers an efficient and accurate approach for automated bleeding detection in WCE.
• The semi-automatic annotation tool addresses limitations of manual data preparation.
• This technology has the potential to improve diagnostic accuracy and workflow in WCE analysis.