A review of prognostic prediction of coronary artery disease patients with myocardial perfusion scintigraphy and artificial intelligence

Abstract

Background

Coronary artery disease (CAD) is a major global cause of death .. For patients with suspected CAD, single photon emission computed tomography (SPECT) which is frequently used as a diagnostic tool is primarily used for assessing myocardial perfusion.In simple terms, to find blockages in heart arteries, a method called total perfusion deficit (TPD) is used during SPECT. This method gives us a clear, numerical measure of how much and how severely the heart muscle is not getting enough blood.

Recently, there has been a growing interest in using artificial intelligence (AI) to predict health outcomes in almost all sub-field of medicine. Although AI analysis of SPECT for diagnosis of CAD has a lot of attractions and several studies already exist, however, it is not yet clear exactly how AI, when combined with SPECT imaging, can predict what will happen to patients (prognosis). This review looks closely at all the research done so far on using AI with SPECT, specifically to predict the outcomes for patients with CAD.

Material Methods

A systematic search was conducted in Pubmed, Embase and Cochrane databases from the beginning of the database until November 30, 2023, including the prognostic predictions of SPECT-based AI models. This included all clinical prognostic outcome studies focusing on deep learning applications and MPI. 6 studies reported in the literature were identified. After finding these studies, their summaries (abstracts) and full texts were carefully reviewed for detailed analysis from both clinical and AI aspects.

Results

Among these studies, two of them were planned as a multi-center while others were single-center. Among single center studies, the prognostic outcome was determined as major cardiac adverse event (MACE) while other studies showed different prognostic outcome as illustrated in Table . We have compared potential differences between follow-up periods, outcome rates, patient population / size, gender, and ages, all showing considerable differences . We have compared the prognostic outcomes of these studies too: AI-models solely derived from SPECT imaging was already found to be statistically significantly superior to models produced from stress TPD. In three of the studies, SPECT imaging based AI models were also shown to be statistically significantly superior to the Logistic Regression (LR) models produced from combined feature sets of the demographic characteristics, laboratory values . AI-models derived from SPECT imaging and demographic information was statistically significantly superior to LR and stress TPD. Last, but not least, only one study showed that inclusion of demographic information to SPECT imaging based AI-model was statistically significant.

Conclusion

This review reveals that AI models created using SPECT images alone, or in combination with demographic information, are more effective than other methods in predicting the future health outcomes of patients with CAD.