Incremental prognostic value of AI-based coronary CT angiography combined with PET/CT dynamic myocardial perfusion imaging in patients with suspected or confirmed coronary artery disease.

BACKGROUND: Coronary computed tomography angiography (CCTA) and positron emission tomography/computed tomography (PET/CT) myocardial perfusion imaging (MPI) provide complementary anatomical and functional information for coronary artery disease (CAD). However, the prognostic value of integrating CCTA-derived coronary imaging characteristics with PET-MPI parameters remains to be established. Therefore, this retrospective single-center study aimed to evaluate the prognostic and incremental value of integrating artificial intelligence (AI)-based CCTA with PET/CT MPI for predicting major adverse cardiovascular events (MACEs) in patients with CAD. METHODS: Patients with suspected or confirmed CAD underwent PET/CT dynamic MPI and contemporaneous CCTA were retrospectively enrolled and followed for at least 12 months for MACEs. MACEs were defined as the occurrence of at least one of the following endpoints: all-cause death, nonfatal myocardial infarction, new-onset heart failure, rehospitalization for angina, or late coronary revascularization. Clinical characteristics, AI-derived CCTA features, and PET-MPI parameters were collected. To reduce the number of candidate variables and mitigate potential multicollinearity, LASSO regression was applied as an exploratory variable selection method. Variables with non-zero coefficients were subsequently included in multivariable Cox proportional hazards models to assess their associations with the risk of MACEs. Dose-response relationships were further explored using restricted cubic spline analyses. Incremental prognostic value was evaluated using hierarchical Cox regression models. RESULTS: A total of 71 patients (mean age 60.17 ± 12.36 years; 59.2% male) were included, with 18 MACEs occurring over a median follow-up of 25.69 months. Patients with MACEs more frequently exhibited obstructive coronary stenosis, abnormal PET-MPI semiquantitative parameters, reduced CT-derived fractional flow reserve (CT-FFR), and a higher burden of high-risk plaque features (all P < 0.05). The total number of high-risk plaque features was significantly associated with the risk of MACEs (HR 1.898, 95% CI 1.015-3.546; P = 0.045) and showed a trend toward a linear relationship with event risk. PET-MPI semiquantitative parameters improved prognostic performance beyond the clinical baseline model (P = 0.011), and further incorporation of high-risk plaque feature count significantly improved model performance across functional models (all P < 0.05). CONCLUSIONS: AI-driven CCTA integrated with PET/CT dynamic MPI may provide prognostic information in patients with CAD. PET-MPI semiquantitative parameters and the total number of high-risk plaque features were associated with risk and may offer complementary and incremental value for predicting MACEs. These findings warrant further validation in larger cohorts.