AI-Based Peptic Ulcer Detection Using Deep Convolutional Neural Networks: A Diagnostic Accuracy Study
Keywords:
Peptic Ulcer Detection, Deep Convolutional Neural Network (DCNN), Endoscopic Image Analysis, Artificial Intelligence, Medical Image Classification, Computer-Aided Diagnosis, Gastrointestinal Disorder, Peptic ulcer disease (PUD), wireless capsule endoscopy (WCE), nonsteroidal anti-inflammatory drugs (NSAIDs).Abstract
Peptic ulcer disease remains a major gastrointestinal disorder that often goes undiagnosed due to subtle visual cues and variability in endoscopic interpretation. To address these diagnostic challenges, this study proposes an AI-based diagnostic framework utilizing Deep Convolutional Neural Networks (DCNNs) for the automated detection and classification of peptic ulcers from endoscopic images. This study presents an AI-based diagnostic framework employing pretrained Deep Convolutional Neural Networks (DCNNs) — VGG16, ResNet50, InceptionV3, and the Hugging Face Vision Transformer (ViT) — for automated classification of ulcerous and non-ulcerous endoscopic images. The proposed model integrates transfer learning and Grad-CAM visualization, enhancing both diagnostic accuracy and interpretability. Comparative analysis reveals that the Hugging Face Vision Transformer (ViT) architecture achieved the highest classification accuracy of 96.8%, outperforming traditional CNN models. The developed system offers a clinically interpretable, scalable, and user-friendly diagnostic tool that assists in the early identification of gastric and duodenal ulcers, thereby reducing the dependence on manual endoscopic evaluation. Experimental results demonstrate that the DCNN models can effectively identify early-stage ulcers with reduced false negatives, highlighting its potential as a reliable clinical decision-support tool.
