Prognostic Model Development for Continuous Carotid Intima-Media Thickness: A Graph-Driven Self-Supervised Learning...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiovascular Research

Background:

• Cardiovascular disease (CVD) is a major global health issue.
• Carotid intima-media thickness (cIMT) is a key biomarker for atherosclerosis and cardiovascular risk.
• Current cIMT measurement methods (ultrasound) have accessibility limitations, especially for stroke survivors.

Purpose of the Study:

• To develop a prognostic learning model for estimating cIMT without imaging data.
• To enable precise quantification of atherosclerosis severity.
• To address limitations of existing tabular data models that only classify risk presence/absence.

Main Methods:

• Constructed a patient similarity graph using demographic and clinical features.
• Developed a graph-guided self-supervised learning (Self-SL) framework.
• Learned informative representations encoding local and global graph information.

Main Results:

• The model effectively estimates cIMT, quantifying atherosclerosis severity without imaging.
• Achieved up to 93.22% average Mean Squared Error (MSE) reduction on the UK Biobank cohort.
• Outperformed conventional learning models in prediction accuracy.

Conclusions:

• Graph similarity effectively captures latent clinical patterns for cIMT prediction.
• The Self-SL framework provides accurate atherosclerosis assessment, enhancing accessibility.
• This approach offers a privacy-preserving and efficient alternative for CVD risk assessment.