A Machine Learning-Driven Approach to Quantifying Coronary Artery Tortuosity

Background

The associations between coronary artery tortuosity and age, sex, and cardiovascular risk factors are not fully established. Prior studies enrolled fewer than 1,000 subjects and relied on heuristic metrics.

Objectives

The purpose of this study was to develop and validate an automated right coronary artery (RCA) tortuosity measure using machine learning (ML)-based vessel segmentation and examine its association with sex, age, coronary artery disease (CAD), and cardiovascular risk factors.

Methods

We developed an ML-enabled pipeline to quantify RCA tortuosity as a continuous measure in 38,691 RCA angiograms from 22,334 patients and evaluated concordance with blinded interventional cardiologist review in a subset of 300 angiograms. Regression models were used to study the association of tortuosity with age, sex, hypertension, diabetes, hypercholesterolemia, smoking, and presence of CAD.

Results

Tortuosity ranged from 0.007 to 0.289. Blinded interventional cardiologist classification of high tortuosity vs not had 85.6% precision. Tortuosity was higher in women (β per SD = 0.17; P < 0.001) and higher with hypertension (β per SD = 0.06; P = 0.002), but lower with diabetes (β per SD = -0.11, P < 0.001. After adjustment for risk factors, age was not independently associated with tortuosity. Hypercholesterolemia and smoking were not associated. Higher tortuosity was associated with CAD (OR per SD = 1.05; P < 0.001), severe CAD (OR per SD = 1.09; P < 0.001), and higher Gensini score (β per SD = 0.05; P < 0.001), even after adjustment.

Conclusions

We derived a scalable ML-enabled measure of RCA tortuosity from coronary angiography and found associations with sex, hypertension, diabetes, and presence and severity of CAD.