Deep Learning Radiomics Signature from Multicontrast MRI for Automated Identification of Symptomatic Carotid Plaques:...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Disease Research

Background:

• Ischemic stroke is a major global health concern.
• Identifying symptomatic carotid plaques is crucial for stroke risk stratification, but current methods relying on stenosis degree are insufficient.
• Advanced imaging biomarkers are needed for precise risk assessment.

Purpose of the Study:

• To develop and validate a deep learning radiomics (DLR) signature using multicontrast MRI.
• To accurately identify symptomatic carotid atherosclerotic plaques.
• To improve stroke risk stratification beyond traditional stenosis measurements.

Main Methods:

• A retrospective multicenter study enrolled 409 carotid arteries from 355 patients.
• Deep learning and radiomics features were extracted from automatically segmented plaque regions on multicontrast MRI.
• A 3-stage feature selection pipeline optimized the DLR signature, which was used to train machine learning classifiers for symptomatic plaque identification.

Main Results:

• The DLR model achieved high area under the receiver operating characteristic curve (AUROC) values in training (0.975), internal validation (0.933), and external validation (0.881).
• The DLR signature significantly outperformed clinical, radiomics-only, and deep learning-only models.
• Consistent high performance was observed across different stenosis severities, and DLR-defined symptomatic groups showed a higher prevalence of complex plaque types.

Conclusions:

• The developed multicontrast MRI-based DLR signature is a highly accurate and robust tool for automated identification of symptomatic carotid plaques.
• This noninvasive DLR signature holds significant potential for guiding personalized stroke prevention strategies.
• The DLR approach enhances the ability to stratify stroke risk in patients with carotid atherosclerosis.