Deep learning-based multimodal risk stratification for atherosclerosis management

Area of Science:

• Cardiovascular Imaging and Diagnostics
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Atherosclerosis poses a significant risk for cardiovascular events, necessitating precise risk stratification.
• Current risk stratification methods, relying on subjective imaging and clinical scores, lack precision.

Purpose of the Study:

• To develop and validate a multimodal deep learning (DL) model for enhanced atherosclerosis risk stratification.
• To improve the accuracy and efficiency of identifying high-risk plaques compared to traditional methods.

Main Methods:

• A DL model integrating U-Net for segmentation, ResNet for classification, and an attention mechanism was developed.
• The model processed multimodal data including ultrasound, CTA, and clinical variables.
• A 5-fold cross-validation strategy was employed on a split dataset (8:1:1).

Main Results:

• The U-Net achieved a Dice coefficient of 0.88 for lesion segmentation.
• The ResNet model reached 92% classification accuracy and an AUC of 0.97.
• The attention mechanism improved vulnerable plaque detection by 10%, and processing time was reduced by 70%.

Conclusions:

• A multimodal DL model significantly enhances atherosclerosis risk stratification accuracy (92% accuracy, 0.97 AUC).
• The model provides a rapid (70% time reduction) and reliable decision-support tool for clinicians.
• This approach is expected to optimize individualized intervention strategies and improve patient prognosis.