Stomach Deformities Recognition Using Rank-Based Deep Features Selection.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Gastroenterology

Background:

• Wireless capsule endoscopy (WCE) generates extensive data (over 57,000 frames), making manual analysis for gastrointestinal malformations time-consuming and challenging for clinicians.
• Accurate and efficient classification of abdominal infections from WCE images is crucial for timely patient diagnosis and treatment.

Purpose of the Study:

• To develop and evaluate a novel automated computerized method for classifying gastrointestinal infections using WCE images.
• To enhance the diagnostic efficiency and accuracy of WCE image analysis through advanced feature extraction and selection techniques.

Main Methods:

• A system combining segmentation, deep feature extraction (DenseNet CNN with transfer learning), and feature optimization (Kapur's entropy, Tsallis entropy) was developed.
• Ulcer abnormalities were initially extracted using a color features based low level and high-level saliency (CFbLHS) estimation method.
• A multilayered feedforward neural network classifier was used for final classification after selecting the top 50% of optimized features.

Main Results:

• The proposed automated method achieved a maximum classification accuracy of 99.5% on a collected WCE dataset.
• The entire classification process was completed in a significantly reduced time of 21.15 seconds.
• The system demonstrated high efficacy in identifying and classifying abdominal infections from WCE images.

Conclusions:

• The developed automated WCE image classification system offers a highly accurate and efficient solution for diagnosing gastrointestinal infections.
• This novel approach has the potential to alleviate the burden on gastrologists by automating the tedious frame-by-frame analysis of WCE videos.
• The method provides a promising tool for improving the diagnostic workflow in gastroenterology through artificial intelligence.