Explainable Artificial Intelligence Computer-Aided Diagnosis for Heart Myocardial Infarction

Myocardial infarction (MI) is a significant cardiovascular condition known for its detrimental impact on the heart. The rapid and accurate diagnosis of MI is crucial for reducing mortality rates. In the realm of medical imaging interpretation, explainable artificial intelligence (XAI) has emerged as a promising research field. XAI offers a valuable tool to create interpretable saliency maps for each MRI slice, providing insights into the decision-making processes employed by AI models. Our proposed framework begins with the enhancement of MI images using Contrast Limited Adaptive Histogram Equalization. We integrate and adopt five deep learning models: InceptionV3, InceptionResNetV2, ResNet50V2, VGG16, and Xception, along with the Grad-CAM technique, to demonstrate the robustness of these models in MI detection. Among these models, InceptionResNetV2 achieved the highest accuracy with a score of 83.33%, and an impressive F-score of 88.10%. However, based on the Grad-CAM results, the most trustable and robust model is ResNet50V2, which consistently demonstrated its focus on regions of interest in both normal and abnormal predictions. These compelling findings should motivate researchers and stakeholders in the medical industry to consider implementing this framework in practice, ultimately contributing to the early and precise diagnosis of MI and reducing MI-related mortality rates.