Introduction: Over the last decade, implementation of machine learning (ML) algorithms in cardiology, particularly in the field of nuclear cardiology, has revolutionized individualized diagnosis and prognostic estimations of myocardial ischemia-related major adverse cardiovascular events (MACE). The use of non-invasive, high-quality imaging via single-photon emission computed tomography (SPECT), together with clinical data, improves accuracy of diagnosis and prognosis of MACE compared to traditional methods. The aim of this review is to analyze the evidence for diagnostic and prognostic estimations of the combined use of SPECT myocardial perfusion imaging (MPI) and clinical data using ML.     

Methods: A literature search of databases (e.g., PubMed and SpringerLink) was refined to identify reports of combined SPECT MPI and clinical data using ML focusing on diagnostic and prognostic values in MACE outcomes (e.g., non-fatal myocardial infarction, unstable angina, and coronary revascularization). 

Results: Three cohort studies were included for review. A boosting ML ensemble of MPI total perfusion deficits (TPD) and clinical variables (e.g., age, sex, and post-electrocardiogram CAD probability) for detection of CAD were compared to standard imaging quantification and to visual analysis by two experienced readers. The diagnostic accuracy of ML (87.3%) was similar to expert 1 (86.0%), but superior to combined supine and prone TPD (82.8%) and expert 2 (82.1%, p < 0.01). Also, the AUC for ML algorithm (0.94) was higher than TPD and both visual readers (p < 0.001). In prediction of revascularization, the ML ensemble sensitivity (73.6%) was similar when compared to one expert reader (73.9%) and to imaging quantification (75.5%), though ML specificity (74.7%) was better than both expert readers (67.2% and 66.0%, p < 0.05). The specificity of ML was similar to ischemic TPD (68.3%, p < 0.05). The AUC for ML was similar to expert 1 (0.81) but superior to expert 2 (0.72, p < 0.01) and to imaging quantification (0.77, p < 0.01). In a 3-year risk prediction of MACE, ML ensemble had higher predictive accuracy compared to MD diagnosis, automated stress TPD, and automated ischemic TPD (AUC: 0.81 vs 0.65 vs 0.73 vs 0.71, respectively; p < 0.01).

Conclusion: ML demonstrated to improve SPECT MPI diagnostic accuracy, as well as a superior predictive accuracy for 3-year risk of MACE and for early revascularization after MPI. The integration of MPI with clinical data via ML is shown to be equally as effective, and in some instances a superior method to generate individualized diagnosis and prognoses of MACE in patients undergoing SPECT MPI.

Machine Learning Integration of SPECT Myocardial Perfusion Imaging and Clinical Data in MACE