A non-invasive prediction model for coronary artery stenosis severity based on multimodal data

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiovascular Diagnostics

Background:

• Current coronary artery disease (CAD) diagnosis relies on invasive coronary angiography, which carries procedural risks.
• Accurate assessment of stenosis severity is crucial for effective CAD management and treatment planning.
• There is a need for reliable, non-invasive methods to diagnose and stratify CAD risk.

Purpose of the Study:

• To develop a transformer-based multimodal prediction model for non-invasive assessment of coronary artery stenosis severity.
• To integrate heterogeneous biomarkers, including facial morphometrics, cardiovascular waveforms, and biochemical indicators, for precision risk stratification.
• To establish an interpretable framework for enhanced CAD diagnosis.

Main Methods:

• A transformer-based architecture with residual modules and adaptive weighting was employed.
• Multimodal data (facial features, lip/tongue images, pulse/pressure waves, lab indicators) were collected from 488 CAD patients.
• The model was trained and validated on internal and external datasets to predict stenosis severity.

Main Results:

• The model achieved over 90% accuracy in assessing coronary artery stenosis risk on the training dataset.
• External validation demonstrated robustness with 85% accuracy on a real-world validation set.
• Integration of multimodal data and advanced architecture significantly improved predictive performance.

Conclusions:

• A non-invasive, transformer-based multimodal model for CAD stenosis severity assessment was successfully developed.
• The model presents a clinically viable alternative to invasive diagnostic procedures.
• Multimodal data integration holds significant potential for improving CAD diagnosis and patient care.