Explainability of Disease Classification from Medical Images Using XGrad-Cam

Abstract

In this study, a novel approach to disease classification from medical images using a customized convolutional neural network (CNN) and an advanced Explainable AI algorithm, XGrad-Cam, is presented. The proposed CNN model is streamlined with 11 layers, ensuring efficient training and accurate disease classification. XGrad-Cam, an extension of the Grad-Cam algorithm, offers improved visualization by highlighting critical areas contributing to the classification decisions. This model was evaluated on medical images, including chest CT scans for SARS-COVID and brain MRIs for Alzheimer’s disease. Through extensive experimentation with both binary and multi-class disease datasets, the model demonstrated high accuracy, precision, recall, sensitivity, specificity, and F-1 scores, notably achieving 90.8% accuracy for SARS-COVID and 91.0% for Alzheimer’s disease classification. The utilization of XGrad-Cam enhances the interpretability of the CNN’s decisions, providing valuable insights into the disease classification process. This study contributes to the fields of medical image analysis and Explainable AI by offering a highly efficient and interpretable model for disease classification, paving the way for future advancements in automated diagnosis and treatment planning.