Interpretable AI-Driven Wireless Capsule Endoscopy Image Classification: Advancing Explainability in Medical Diagnostics

Authors- Dr.A.Radha Krishna, Ch.Raja Rajeswari, Y.V.S.Kartheek, L.Bhuvan Sai Ram, R.Arya Vinayaka Venkata Siva Sai, K.Teja Ramyasr

Abstract-Deep learning has significantly advanced medical imaging and computer-aided diagnosis (CAD), providing powerful tools for disease detection. While various deep learning (DL) models exist for medical image classification, further analysis is needed to better understand their decision-making processes. To address this, several explainable AI (XAI) techniques have been proposed to improve the interpretability of DL models. In the field of endoscopic imaging, medical professionals primarily rely on visual inspections for preliminary disease diagnosis. However, integrating automated deep learning systems can enhance both efficiency and accuracy in medical assessments. This study aims to increase the reliability of model predictions in endoscopic imaging by implementing multiple transfer learning models on a balanced subset of the Kvasir-Capsule dataset, a widely used wireless capsule endoscopy (WCE) imaging dataset. The dataset subset includes the top 9 classes for training and testing. Our results demonstrate that the Vision Transformer model achieved an F1-score of 97% ± 1%, outperforming previous studies on the same dataset. Other models, such as MobileNetV3Large and ResNet152V2, also performed exceptionally well, achieving F1-scores above 90%. These results mark a significant improvement over prior benchmarks. To enhance model interpretability, we employ multiple XAI techniques, including Grad-CAM, Grad-CAM++, Layer-CAM, LIME, and SHAP, to generate heat maps that highlight key decision-making regions.

DOI: /10.61463/ijset.vol.13.issue2.260