High-risk carotid plaque detection using a novel radio-morphometric analysis of the carotid bifurcation.
Authors
Affiliations (8)
Affiliations (8)
- Department of Electronics, Information and Bioengineering, Politecnico di Milano, Via Ponzio 34/5, Milan, 20133, Italy.
- Department of Chemistry, Material and Chemical Engineering, Politecnico di Milano, P.zza Leonardo da Vinci, 32, Milan, 20133, Italy.
- Department of Clinical Sciences and Community Health, University of Milan, Via Festa del Perdono, 7, Milan, 20122, Italy.
- Foundation IRCCS Ca' Granda Ospedale Maggiore Policlinico, via Francesco Sforza 28, Milan, 20122, Italy.
- Foundation IRCCS Ca' Granda Ospedale Maggiore Policlinico, Via Francesco Sforza 28, Milan, 20122, Italy.
- Electronics, Information and Bioengineering, Politecnico di Milano, Via Ponzio 34/5, Milano, 20133, Italy.
- Politecnico di Milano, Piazza Leonardo da Vinci 32, Milano, 20133, Italy.
- Department of Electronics, Information and Bioengineering, Politecnico di Milano, Via Ponzio 34/5, Milan, Lombardy, 20133, Italy.
Abstract
Current surgical eligibility criteria for the treatment of carotid artery disease, primary relying on stenosis and image plaque characteristics, are suboptimal, highlighting the unmet need for improved strategies focused on recognition of high-risk carotid plaques. Despite the suggestion of various vulnerability biomarkers (clinical, biological, image, morphometric and biomechanical) a comprehensive approach integrating multiple markers is lacking. To this aim, this study proposes a novel integrative machine learning approach for the detection of symptomatic and asymptomatic patients, based on clinical data, plaque radiomics and morphometric features of the carotid bifurcation. 
Approach. The study included 129 patients (53 of which symptomatic) who underwent computed tomography angiography (CTA) prior to carotid endarterectomy. Following image preprocessing and segmentation, CTA-based radiomic features were extracted and a morphometric analysis of the carotid centerline and bifurcation was performed. Machine learning models were implemented to stratify symptomatic and asymptomatic patients using clinical characteristics, radiomic features and morphometric markers alone and in combination. Moreover, radiomic-clinical, radiomic-morphometric, clinical-morphometric and radiomic-clinical-morphometric models were developed by combining the predicted probabilities of the individual models. 
Main results. The radiomic model was the strongest standalone predictor (AUC=0.903, balanced accuracy=0.847), significantly outperforming both the clinical and morphometric models (AUC=0.611 and 0.567, balanced accuracy=0.642 and 0.648, respectively). Integrating radiomics with clinical and morphometric features also demonstrated excellent stratification ability (AUC=0.883), with the highest balanced accuracy (=0.884), associated with the correct identification of all the 12 symptomatic patients, and 21 over 27 asymptomatic ones.
Significance. This study introduced the first CTA-based approach integrating clinical characteristics, plaque radiomics, and carotid bifurcation morphometric features to predict cerebrovascular events. The findings highlight the value of integrating multiple sources to capture subtle and complementary determinants of cerebrovascular risk, supporting the potential for automated risk stratification and personalized decision-making in patients with carotid artery disease.