High-Risk Carotid Lesion Segmentation: Advancing Stroke Risk Detection With Deep Learning

Area of Science:

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

Background:

• Carotid artery disease management is debated for asymptomatic lesions.
• Plaque characteristics may predict events better than stenosis alone.
• AI feasibility for carotid lesion segmentation is investigated.

Purpose of the Study:

• To assess AI software (PRAEVAorta2) for segmenting carotid lesions on CT angiography.
• To compare plaque composition (thrombus, calcium) between symptomatic and asymptomatic patients.
• To evaluate AI segmentation accuracy against manual segmentation.

Main Methods:

• Two segmentation methods: manual and AI-based (PRAEVAorta2).
• Analysis of thrombus, calcium, and lumen volume in 156 patients.
• Evaluation of AI performance using DSC, sensitivity, specificity, and reliability metrics.

Main Results:

• AI segmentation showed strong agreement with manual segmentation for lumen, calcification, and plaque.
• Symptomatic lesions had higher thrombus volume and total lesion volume.
• Asymptomatic lesions showed higher calcification-to-total volume ratios.
• AI demonstrated high intra- and inter-observer reliability.

Conclusions:

• PRAEVAorta2 effectively automates carotid lesion analysis on CT angiography.
• AI aids in identifying high-risk asymptomatic plaques, supporting surgical decisions.
• Plaque composition differs significantly between symptomatic and asymptomatic carotid lesions.