A Non-Invasive Prediction Model for Coronary Artery Stenosis Severity Based on Multimodal Data

Jiyu Zhang¹Jiatuo Xu^1*Liping Tu¹Tao Jiang¹Yu Wang¹Jijie Xu²

• ¹Shanghai University of Traditional Chinese Medicine, Shanghai, China
• ²Shanghai Baoshan Hospital of Integrated Traditional Chinese and Western Medicine, Shanghai, China

The use of AI-based intelligent disease diagnosis has become a research hotspot in recent years. This study aims to develop a Transformer-based multimodal prediction model for the non-invasive assessment of coronary artery stenosis severity. We hypothesize that by integrating multiple data modalities, the transformer architecture can provide an efficient and accurate assessment of coronary artery stenosis, offering a viable auxiliary option to traditional invasive diagnostic methods. The proposed model incorporates Transformer architecture with residual modules and an adaptive weighting mechanism. Multimodal data, including facial features, tongue images, pulse and pressure wave amplitudes, and laboratory indicators from 488 patients with coronary artery disease (CAD), are collected, preprocessed, and analyzed to predict stenosis severity. Model performance is evaluated through both internal and external validation datasets to assess accuracy and generalizability. Experimental results show that the model achieves over 90% accuracy on the training dataset, demonstrating high performance in risk assessment. On external validation, the model reaches 85% accuracy, further confirming its potential for clinical application. The fusion of multimodal data and the advanced architecture enhances model performance, improving prediction accuracy and robustness. In conclusion, this study presents a Transformer-based multimodal non-invasive prediction model, offering a promising alternative for assessing coronary artery stenosis severity with significant clinical applicability.