Radiomics of Plaque‐Free Carotid Ultrasound Detects Severe Coronary Artery Calcification
Youngjun Lee, Miran Lee, Seokkyoon Hong, Rachel K. Surowiec, Jeonggyu KangObjectives
To assess whether radiomics analysis of carotid ultrasound (CUS) can identify texture features associated with severe coronary artery calcification, even in individuals without carotid plaque.
Methods
This study included 105 participants with coronary artery calcium score (CACS) ≥400 and no carotid plaque, matched by age and sex to 105 controls with CACS = 0. B‐mode CUS images of the bilateral distal common carotid arteries (CCA) were analyzed, with 1‐cm longitudinal regions of interest extending from the lumen to the adventitia. Radiomic features were extracted from each frame, filtered by variance and correlation, and ranked using bootstrap‐based XGBoost feature importance (FI) to identify reproducible features for evaluation.
Results
Among 700 extracted features, the final retained features were reproducible: 8 (right) and 11 (left) for CACS = 0, and 7 (right) and 10 (left) for CACS ≥400 (all p < .05). Group‐specific features, observed only in the CACS 0 or CACS ≥400 group, included 90th Percentile (CACS = 0: right distal CCA, FI = 0.030; left distal CCA, FI = 0.025) and Run Entropy (CACS ≥400: right distal CCA, FI = 0.046; left distal CCA, FI = 0.038). Shared features such as Long Run Emphasis, Dependence Non‐Uniformity, and Entropy were consistently observed across both groups and sides (FI = 0.023–0.029), with Dependence Non‐Uniformity consistent in the left distal CCA across both groups.
Conclusions
Although no plaque was detected on CUS, radiomics can identify ultrasound texture patterns associated with severe coronary calcification. This approach may improve detection of high‐risk individuals who would otherwise be classified as low‐risk by CUS alone.