Capsule Networks for COVID-19 Detection with Explainable AI via Grad-CAM

In recent years, the world has faced a severe epidemic, COVID-19, which has significantly impacted global health. Traditional laboratory methods for COVID-19 detection are time-consuming and likely to make errors. Early and Accurate detection can save lives. The urgent need for early detection of this life-threatening disease has driven the designing of advanced AI techniques to detect COVID-19. Even though advanced AI models have been developed for COVID-19 detection, they lack explanation regarding the model’s decisions. The use of explainable models is essential to support clinical decision-making systems. This research develops a methodology to overcome these challenges of time-consuming disease prediction and lack of explainable AI models. We propose a Capsule Network with a customized dynamic routing algorithm, combined with an explainable AI technique using Grad-CAM, to enhance the efficiency of detection of COVID-19 in chest X-rays. Grad-CAM visualizations highlight relevant regions in chest X-rays, providing critical insights to support the decision system of the proposed model. The proposed model achieved a test accuracy of 99.64%, training accuracy of 99.47%, recall of 99%, 100% precision, 99% F1-score, and AUC scores of 99.54% and 99.89% on training and testing data, respectively. The integration of explainability methods offers a valuable tool for medical professionals, improving the reliability and transparency of AI-driven diagnoses.