Bleeding Detection in Wireless Capsule Endoscopy Image Video Using Superpixel-Color Histogram and a Subspace KNN...

Area of Science:

• Medical Imaging
• Gastroenterology
• Computer-Aided Diagnosis

Background:

• Wireless Capsule Endoscopy (WCE) is a valuable tool for gastrointestinal (GI) disease diagnosis.
• Manual review of WCE images is time-consuming, creating a need for automated systems.
• Computer-aided diagnosis (CAD) systems can improve efficiency in WCE image analysis.

Purpose of the Study:

• To develop an automatic bleeding detection algorithm for WCE images.
• To improve the speed and accuracy of GI bleeding detection.
• To reduce the workload for clinicians analyzing WCE data.

Main Methods:

• A three-stage algorithm was developed: preprocessing (key frame extraction, edge removal), bleeding frame discrimination using superpixelcolor histogram (SPCH) features and a subspace KNN classifier, and bleeding region segmentation with a 9-D color feature vector.
• Utilized novel SPCH features based on the principle color spectrum.
• Employed a subspace KNN classifier for frame discrimination and superpixel-level feature extraction for segmentation.

Main Results:

• The proposed algorithm achieved a high accuracy of 0.9922 in GI bleeding detection.
• The method demonstrated superior performance compared to existing state-of-the-art techniques.
• The algorithm offers low computational costs, making it efficient for clinical application.

Conclusions:

• The developed automatic bleeding detection algorithm is highly accurate and efficient for WCE.
• This CAD system can significantly assist clinicians in diagnosing GI bleeding.
• The proposed method represents an advancement in automated analysis of WCE images.