Endoscopic Image Classification: Advancing AI-Based Gastrointestinal Diagnostics

Gastrointestinal (GI) endoscopy is the gold standard for detecting and treating GI cancers. However, traditional endoscopic techniques require significant expertise, visual-spatial skills, and extensive practical experience, making them technically demanding. To address these challenges, artificial intelligence has emerged as a promising tool for enhancing endoscopic image analysis. Deep learning, particularly convolutional neural networks, has demonstrated significant potential in medical imaging diagnostics. This study investigates the application of two deep learning models—DenseNet-121 and ResNet-152—on the GastroVision and Kvasir GI endoscopy datasets, which contain 16,220 images. The performance of these models in endoscopic image classification is evaluated across various categories, demonstrating high accuracy and reliability, particularly in specific disease classifications. The results indicate that in small-scale classification tasks, the models achieve an accuracy of up to 97%, while for multi-class classification involving twelve categories with sufficient data, the accuracy reaches 84%. The models have been further extended to nineteen categories, with promising performance, suggesting their potential application in clinical settings. Our results highlight the potential of AI in medical image analysis and paves the way for the AI-based gastrointestinal diagnostics in the future.