Integrating SHAP, LIME and Grad-CAM with CNNs for Transparent Kidney Condition Classification

This research introduces a method for diagnosing kidney conditions by combining Convolutional Neural Networks (CNNs) with cutting-edge explainable AI techniques. The study aims to classify CT kidney images into four categories: Normal, Cyst, Tumor, and Stone. The proposed lightweight CNN model demonstrates high performance with validation and test accuracy of 92.71% and 92.12%, respectively, and a training accuracy of 97.92%. Precision, recall, and F1-score are 0.92%, 0.86%, and 0.88%, respectively, with an outstanding AUC of 0.99%. To improve the clarity of the model's decision-making process, SHAP (SHapley Additive exPlanations), LIME (Local Interpretable Model-agnostic Explanations), and Grad-CAM (Gradient-weighted Class Activation Mapping) were used. LIME provided insights into the most influential image regions, SHAP values explained feature contributions to predictions, and Grad-CAM visualizations highlighted critical areas of images impacting classification. The integration of these explainable AI methods not only improves diagnostic Kidney condition accuracy but also fosters trust in the model prediction and makes medical image analysis.