Predicting Cardiovascular Events in Hemodialysis Patients Based on the Fusion of Physicochemical Indicators and Tongue Images:a prospective and multicenter study

Abstract

Cardiovascular events represent the predominant cause of mortality among patients undergoing hemodialysis, and the precise prediction of such events holds significant clinical significance. Conventional predictive models, which predominantly rely on biochemical and physiological parameters, have already exhibited robust performance. Recent investigations have commenced exploring the utilization of facial or tongue images for the risk assessment of cardiovascular diseases. Nevertheless, to the present date, no research has specifically probed the utility of tongue images for predicting cardiovascular events in hemodialysis patients. Inspired by the traditional Chinese medicine (TCM) principle that “the heart opens into the tongue,” we employed machine learning techniques to extract quantitative features from tongue images for the prediction of cardiovascular events in this high risk population. Furthermore, we integrated these tongue derived features with well established biochemical and clinical variables to develop a multimodal predictive model intended to improve the accuracy of risk stratification. The effectiveness and feasibility of this approach were evaluated via a prospective, multicenter cohort clinical study with standardized data collection. Our findings illustrate a novel paradigm that synergistically integrates TCM theory with artificial intelligence for cardiovascular risk prediction in hemodialysis patients, presenting a promising supplementary strategy to current clinical practice.