Search

PURPOSE: Artificial intelligence (AI) has high diagnostic accuracy for coronary artery disease (CAD) from myocardial perfusion imaging (MPI). However, when trained using high-risk populations (such as patients with correlating invasive testing), the disease probability can be overestimated due to selection bias. We evaluated different strategies for training AI models to improve the calibration (accurate estimate of disease probability), using external testing. METHODS: Deep learning was trained using 828 patients from 3 sites, with MPI and invasive angiography within 6-months. Perfusion was assessed using upright (U-TPD) and supine total perfusion deficit (S-TPD). AI training without data augmentation (Model 1) was compared to training with augmentation (increased sampling) of patients without obstructive CAD (Model 2), and patients without CAD and TPD

BACKGROUND: Single photon emission computed tomography (SPECT) myocardial perfusion is frequently used to predict risk of major adverse cardiovascular events (MACE). We performed an external validation of the CRAX2MACE score, developed to estimate 2-year risk of MACE in patients with suspected coronary artery disease (CAD). METHODS: Patients who underwent clinically indicated SPECT with available follow-up for MACE were included (N=2,985). The prediction performance for MACE (revascularization, myocardial infarction, or death) within 2-years for CRAX2MACE was compared with stress and ischemic total perfusion deficit (TPD) using area under the receiver operating characteristic curve (AUC). Calibration was assessed with calibration plots, Brier score and the Hosmer-Lemeshow test. RESULTS: MACE occurred within 2 years in 243 (8.1%) patients. The AUC for CRAX2MACE (0.710, 95% CI 0.677 – 0.743) was significantly higher compared to stress TPD (AUC 0.669, 95% CI 0.632 – 0.706, p=0.010) and ischemic TPD (AUC 0.664, 95% CI 0.627 – 0.700, p