Evaluation of Significant Coronary Artery Disease Based on CT Fractional Flow Reserve and Plaque Characteristics Using Random Forest Analysis in Machine Learning

Rationale and Objectives: Fractional flow reserve (FFR) is an established technique for detecting lesion-specific ischemia but is invasive. Our objective was to investigate the effects of combined assessment of coronary CT angiography (CCTA) imaging features and CT-FFR on detecting lesion-specific ischemia by comparing with invasive FFR. Materials and Methods: Forty-seven patients who had 60 coronary vessels with 30%-90% stenosis were included. Six anatomic CCTA descriptors (Agatston score, stenosis severity, mean plaque CT attenuation value, noncalcified and calcified plaque volumes, remodeling index) and a functional descriptor (CT-FFR) were measured. Random forest was used to identify which descriptors were useful to identify ischemia-related lesion. Receiver-operating characteristic (ROC) curves were calculated for 2 models: i.e. Model-1 for anatomical CT descriptors and Model-2 for anatomical CT descriptors plus CT-FFR. Results: Stenosis severity (40.8 +/- 15.7% vs 57.6 +/- 14.1%), noncalcified plaque volume (190 +/- 100 vs 254.8 +/- 133.3), and remodeling index (1.04 +/- 0.12 vs 1.11 +/- 0.13) were significantly higher in ischemia-related lesions than nonischemia-related lesions. CT-FFR was 0.84 +/- 0.14 and 0.71 +/- 0.14, respectively, for ischemia-related and nonischemia-related lesions, and the difference was significant. The area under the ROC curve was 0.738 and 0.835 in Model-1 and Model-2, respectively. Reclassification of ischemic lesion risk was significantly improved after adding CT-FFR: net reclassification improvement was 0.297 and integrated discrimination improvement was 0.254. Conclusion: Combined assessment of anatomical CCTA features and functional CT-FFR was helpful for detecting lesion-specific ischemia.