Component-specific radiomics of carotid plaque: Characterizing calcium, necrotic core, and fibrous tissue for enhanced risk prediction

Carotid atherosclerosis is a major cause of ischemic stroke. Predicting the risk of adverse cerebrovascular events in atherosclerotic patients is crucial to prevent serious outcomes. Recent advances have redefined the approach to stratify patient risk, emphasizing plaque characterization over the traditional measurement of stenosis severity alone. Radiomics-based models have shown promises in this context; however, radiomics has been applied only to the full atherosclerotic plaque, without an in-depth analysis of its components. In this work, we propose a novel component-specific radiomic approach to detect symptomatic and asymptomatic patients from computed tomography angiography (CTA) images by separately assessing calcified, necrotic core and fibrous tissue regions. The dataset included 129 patients, 53 of which symptomatic, undergoing CTA prior to carotid endarterectomy. Plaque components were semi-automatedly segmented and, for each, radiomic features were extracted and a machine learning model was implemented to estimate patient risk. Component-specific models were combined via weighted average to produce an overall patient risk estimate. For comparison, the same methodology was applied to full plaques, as commonly performed. The component-specific models reached balanced accuracies of 0.785, 0.623 and 0.734 for calcium, necrotic core and fibrous tissue, respectively. Each component contributed differently to the patient risk, with calcium and fibrous tissue emerging as key players. The integration of the component-specific contributions further enhanced stratification performance (balanced accuracy 0.848) and surpassed the conventional radiomic approach (balanced accuracy 0.774). This methodology may support the detection of early signs of adverse cerebrovascular events and the development of more individualized preventive and therapeutic strategies.