Deep Learning Based Antral Gastritis Detection Using Endoscopic Images

The proposed study aimed at detection and classification of antral gastritis using pre-trained deep learning models and compared its performance with custom F Net model. This study systematically evaluates and compares the performance of multiple deep learning models in classifying endoscopic images for the detection of antral gastritis. Pre-trained networks such as MobileNet, DenseNet121, and ResNet152V2 were benchmarked against a custom-designed F-Net model to assess their effectiveness. The models were trained on endoscopic images and evaluated based on accuracy, sensitivity, specificity, precision, F1-score, and AUC. MobileNet achieved an accuracy of 85%, sensitivity of 76%, specificity of 100%, and an AUC of 0.99. DenseNet121 and ResNet152V2 outperformed Mobile Net, each attaining 90% accuracy, 83% sensitivity, 100% specificity, and an AUC of 0.99. The proposed F-Net model achieved the highest classification accuracy of 95%, demonstrating superior performance over pre-trained models. These findings highlight the potential of deep learning, particularly F-Net, in enhancing automated detection of antral gastritis, contributing to improved diagnostic accuracy and clinical decision-making.