Combination of quantitative parameters from 13N-NH3 PET/CT myocardial perfusion imaging and attenuation corrected CT radiomics for predicting myocardial viability in patients with coronary artery disease

Background Accurately assessing myocardial viability ‌in the left ventricular (LV) infarcted area of coronary artery disease (CAD) patients‌ is crucial for predicting functional recovery after revascularization. While ¹⁸ F-FDG PET/CT myocardial metabolism imaging (MMI) combined with myocardial perfusion imaging (MPI) is the gold standard, its protocol is ‌complex and may be unreliable in diabetic patients. ‌Quantitative parameters from MPI show inconsistent diagnostic performance.‌ Low-dose attenuation corrected CT (ACCT) may contain imperceptible tissue heterogeneity, which ‌has the potential to be used for assessing viability‌. Therefore, we developed and evaluated a combined model for predicting myocardial viability in patients with CAD by combining quantitative parameters from MPI and ACCT radiomics. Results A total of 140 CAD patients performing myocardial ¹³ N-NH ₃ / ¹⁸ F-FDG PET/CT were enrolled. The myocardium was segmented using 17-segment model. The quantitative parameters including myocardial blood flow (MBF), perfusion score (PS) and metabolism score (MS) were calculated. The segments with PS ≥ 2 were identified as infarcted segments and then divided into training, testing and validation groups. These infarcted segments were further classified into scarred segments (MS = 2–4) and viable segments (MS = 0–1). MPI model was built by MBF and PS, and radiomics model (Rad model) was built by ACCT radiomics. A combined model (MPI-Rad model), incorporating MBF, PS and ACCT radiomics, was constructed using logistic regression and evaluated using AUC, decision curve analysis (DCA) and calibration curve. In the training, testing and validation groups, the AUCs of MPI model were 0.852, 0.784 and 0.860, respectively, while the AUCs of Rad model were 0.766, 0.770 and 0.789. The AUC values (ranging from 0.852 to 0.897) of MPI-Rad model in the three groups were all significantly higher than those of MPI model (all p  < 0.01), and DCA and calibration curve demonstrated improved net benefits and good consistency. Conclusions ACCT radiomics features exhibited moderate diagnostic performance in predicting myocardial viability, nevertheless, radiomics could provide incremental value to MPI quantitative parameter model to enhance the predictive value of the combined model.