Computer assisted gastric abnormalities detection using hybrid texture descriptors for chromoendoscopy images

Area of Science:

• Medical imaging analysis
• Computer vision in gastroenterology

Background:

• Early stomach cancer detection relies on screening procedures like chromoendoscopy (CH).
• Manual inspection of numerous endoscopic frames is time-consuming and exhaustive.
• Automated analysis of gastroenterology images can assist endoscopists in identifying abnormalities.

Purpose of the Study:

• To develop a computer-aided detection system for abnormal chromoendoscopy frames.
• To introduce a novel hybrid texture extraction method for enhanced image analysis.

Main Methods:

• A new feature extraction method, Gabor-based gray-level co-occurrence matrix (G2LCM), was developed.
• G2LCM combines local and global texture descriptors by analyzing Gabor filter responses.
• Second-order statistics of co-occurrence matrices represent image texture for classification.

Main Results:

• The G2LCM method achieved high classification performance for abnormal frames.
• Sensitivity, specificity, and accuracy were 91%, 82%, and 87% respectively.
• Area under the curve (AUC) was 0.91, indicating strong discriminatory power.

Conclusions:

• The proposed system effectively aids gastroenterologists in gastric tract screening.
• Automated analysis using G2LCM can significantly reduce endoscopic procedure times.
• This technology supports earlier and more efficient detection of potential abnormalities.