Prediction of Severe Obstructive Coronary Artery Disease Using Radiomic Features from Resting Cardiac Ultrasound Images: A Pilot Study

Background Although cardiac ultrasound is frequently performed in patients with chest pain, the probability of obstructive coronary artery disease (CAD) cannot be quantified. We investigated the potential of cardiac ultrasound radiomics (ultrasomics) to identify obstructive CAD using limited echocardiography frames, suitable for cardiac point-of-care ultrasound evaluation. Methods In total, 333 patients who were either healthy controls (n=30), undergoing invasive coronary procedures (n=113), or coronary CT angiography (n=190) were divided into two temporally distinct training (n=271) and testing (n=62) cohorts. Machine learning models were developed using ultrasomics for predicting severe CAD (stenosis >70%) and compared with regional LV wall motion abnormalities (RWMA). Results In total, 94 (28.2%) patients had severe CAD with 50 (15.0%) having high-risk CAD defined as left main stenosis >50% (n=11), multivessel CAD (n=43), or 100% occlusion (n=20). The ultrasomics model was superior to RWMA for predicting severe CAD [area under the receiver operating curve (AUC) of 0.80 (95% confidence interval [CI]: 0.74 to 0.86) vs. 0.67 (95% CI: 0.61-0.72), p=0.0014] in the training set and [0.77 (95% CI: 0.64-0.90) vs. 0.70 (95% CI: 0.56-0.81), p=0.24] in the test set, respectively. The model also predicted high-risk CAD with an AUC of 0.84 (95% CI: 0.77-0.90) in the training set and 0.70 (95% CI: 0.48-0.88) in the test set. A combination of ultrasomics with RWMA showed incremental value over RWMA alone for predicting severe CAD. Conclusions Cardiac ultrasomic features extracted from limited echocardiography views can aid the development of machine learning models to predict the presence of severe obstructive CAD.