A Convolutional Neural Network-based explainable classification method of SPECT myocardial perfusion images in nuclear cardiology

This study targets on the development of an explainable Convolutional Neural Network (CNN) pipeline in the form of a handcrafted CNN to identify patients’ coronary artery disease status (normal, ischemia or infarction). The proposed RGB-CNN model utilizes various pre- and post-processing tools and deploys a state-of-the-art explainability tool to produce more interpretable predictions in the task of decision making. The provided dataset includes 630 patients’ cases in stress and rest representations and comprises 257 normal, 241 ischemic and 127 infarction cases, previously classified by a doctor. The imaging dataset was split into 20% for testing and 80% for training, whose 15% was further used for validation purposes. Data augmentation was employed to increase generalization. Grad-CAM based color visualization approach was also utilized to provide predictions with interpretability in the detection of ischemia and infarction in SPECT-MPI images, counterbalancing any lack of rationale in the results extracted by CNNs. The proposed model achieved 94,06% accuracy and 0.9541% AUC, demonstrating efficient performance and stability.