Accepted: IWBBIO2022
This research work addresses the problem of coronary artery disease (CAD) diagnosis using Single photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) data, by applying deep learning techniques. We propose a Convolutional Neural Network implementation to classify SPECT-MPI images fully automatically, also utilizing transfer learning employing VGG-16 as a pre-trained network, for comparison reasons. We developed a three-class classification algorithm capable of achieving high accuracy with minimized loss. The dataset consists of SPECT images in stress and rest representation with the following number of cases: 100 cases correspond to infarction, 180 to ischemia and 190 to normal. Data augmentation was applied due to the small size of the dataset. The RGB-CNN extracted 94,21% accuracy whereas VGG-16 84,37%. The results demonstrated that the RGB model is capable to classify unseen data having great potential as an effective and integral tool for CAD diagnosis in nu-clear medicine.
