Noninvasive machine-learning models for the detection of lesion-specific ischemia in patients with stable angina with intermediate stenosis severity on coronary CT angiography.

Abstract

This study proposed noninvasive machine-learning models for the detection of lesion-specific ischemia (LSI) in patients with stable angina with intermediate stenosis severity based on coronary computed tomography (CT) angiography. This single-center retrospective study analyzed 76 patients (99 vessels) with stable angina who underwent coronary CT angiography (CCTA) and had intermediate stenosis severity (40-69%) on invasive coronary angiography. LSI, defined as a resting full-cycle ratio < 0.86 or fractional flow reserve ≤ 0.80, was determined in 40 patients (46 vessels) using a hybrid resting full-cycle ratio-fractional flow reserve strategy. The resting full-cycle ratio and/or fractional flow reserve were measured using invasive coronary angiography as references for functional severity indices of coronary stenosis in the machine-learning models. LSI detection models were constructed using noninvasive machine-learning models that predicted the resting full-cycle ratio and fractional flow reserve by feeding machine-learning models with image features extracted from CCTA. The diagnostic performance of the proposed LSI detection models was assessed using a nested 10-fold cross-validation test. The LSI detection models with the highest diagnostic performance achieved an accuracy of 0.88 (95% CI: 0.81, 0.94), sensitivity of 0.78 (95% CI: 0.70, 0.86) and specificity of 0.96 (95% CI: 0.92, 1.00) on a vessel basis and 0.88 (95% CI: 0.81, 0.95), 0.80 (95% CI: 0.70, 0.86) and 0.97 (95% CI: 0.92, 1.00), respectively, on a patient basis. These findings suggest that LSI detection models with features extracted from CCTA can noninvasively detect LSI in patients with stable angina with intermediate stenosis severity.