Antral Gastritis Segmentation based on Deep learning models in Endoscopic Images

The study focuses on utilizing machine learning techniques for segmentation, feature extraction, and classification of antral gastritis in endoscopic images to accurately identify affected stomach regions. To identify the impacted stomach areas, deep learning-based segmentation methods such as UNet and SegNet is used for automated segmentation process. Grey Level Co-occurrence Matrix (GLCM) features were used for statistical feature extraction, and three machine learning classifiers—K-Star, Random Forest, and Stochastic Gradient Descent (SGD)—were used to classify the features that were extracted. The segmented regions showed clear patterns in the GLCM feature extraction. SGD and Random Forest had the best accuracy of 97% among the classifiers. K-Star, Random Forest, and SGD classifiers were shown to have specificity and sensitivity of 93% and 90%, 97% and 96%, and 96% and 97%, respectively. These results highlight the potential of the proposed methodology for reliable and automated detection of antral gastritis, which could aid in improving diagnostic accuracy and reducing subjectivity in medical assessments.