Classification models for SPECT myocardial perfusion imaging

Objective: The main goal of this work is to develop computer-aided classification models for single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) to identify perfusion abnormalities (myocardial ischemia and/or infarction). Methods: Two different classification models, namely, deep learning (DL)-based and knowledge-based, are proposed. The first type of model utilizes transfer learning with pre-trained deep neural networks and a support vector machine classifier with deep and shallow features extracted from those networks. The latter type of model, on the other hand, aims to transform the knowledge of expert readers to appropriate image processing techniques including particular color thresholding, segmentation, feature extraction, and some heuristics. In addition, the summed stress and rest images from 192 patients (age 26-96, average age 61.5, 38% men, and 78% coronary artery disease) were collected to constitute a new dataset. The visual assessment of two expert readers on this dataset is used as a reference standard. The performances of the proposed models were then evaluated according to this standard. Results: The maximum accuracy, sensitivity, and specificity values are computed as 94%, 88%, and 100% for the DL-based model and 93%, 100%, and 86% for the knowledge-based model, respectively. Conclusion: The proposed models provided diagnostic performance close to the level of expert analysis. Therefore, they can aid in clinical decision making for the interpretation of SPECT MPI regarding myocardial ischemia and infarction.