Advancing Gastrointestinal Disease Diagnosis: A Fine-Grained Approach Using Swin Transformer and Explainable AI Techniques

Diagnosing gastrointestinal (GI) disease detection is often challenging because the variations present in pathology demand careful and detailed examination. This study integrates Swin Transformer (Swint-t) architecture within a fine-grained visual classification framework to enhance the accuracy of GI disease detection using a recent endoscopic dataset, the GastroVision dataset. Integrating Explainable AI (XAI) techniques, such as GradCAM and LIME, provides deeper insights into the model's decision-making process, promoting trust and transparency. Moreover, the proposed approach in this paper effectively addresses class imbalance and image size variability issues in the dataset by introducing preprocessing and data augmentation techniques. This study also presents an XAI-based comparison against state-of-the-art models to demonstrate our approach's superior performance, offering promising improvements over traditional endoscopic diagnostic procedures.