Utilizing Circadian Heart Rate Variability Features and Machine Learning for Estimating Left Ventricular Ejection Fraction Levels in Hypertensive Patients: A Composite Multiscale Entropy Analysis.
Early identification of left ventricular ejection fraction (LVEF) levels during the progression of hypertension is essential to prevent cardiac deterioration. However, achieving a non-invasive, cost-effective, and definitive assessment is challenging. It has prompted us to develop a comprehensive machine learning framework for the automatic quantitative estimation of LVEF levels from electrocardiography (ECG) signals.
We enrolled 200 hypertensive patients from Zhongshan City, Guangdong Province, China, from 1 November 2022 to 1 January 2025. Participants underwent 24 h Holter monitoring and echocardiography for LVEF estimation. We developed a comprehensive machine learning framework that initiated with preprocessed ECG signal in one-hour intervals to extract CMSE-based heart rate variability (HRV) features, then utilized machine learning models such as linear regression (LR), Support Vector Machines (SVMs), and random forests (RFs) with recursive feature elimination for optimal LVEF estimation.
The LR model, notably during early night interval (20:00-21:00), achieved a RMSE of 4.61% and a MAE of 3.74%, highlighting its superiority. Compared with other similar studies, key CMSE parameters (Scales 1, 5, Slope 1-5, and Area 1-5) can effectively enhance regression models' estimation performance.
Our findings suggest that CMSE-derived circadian HRV features from Holter ECG could serve as a non-invasive, cost-effective, and interpretable solution for LVEF assessment in community settings. From a machine learning interpretable perspective, the proposed method emphasized CMSE's clinical potential in capturing autonomic dynamics and cardiac function fluctuations.
We enrolled 200 hypertensive patients from Zhongshan City, Guangdong Province, China, from 1 November 2022 to 1 January 2025. Participants underwent 24 h Holter monitoring and echocardiography for LVEF estimation. We developed a comprehensive machine learning framework that initiated with preprocessed ECG signal in one-hour intervals to extract CMSE-based heart rate variability (HRV) features, then utilized machine learning models such as linear regression (LR), Support Vector Machines (SVMs), and random forests (RFs) with recursive feature elimination for optimal LVEF estimation.
The LR model, notably during early night interval (20:00-21:00), achieved a RMSE of 4.61% and a MAE of 3.74%, highlighting its superiority. Compared with other similar studies, key CMSE parameters (Scales 1, 5, Slope 1-5, and Area 1-5) can effectively enhance regression models' estimation performance.
Our findings suggest that CMSE-derived circadian HRV features from Holter ECG could serve as a non-invasive, cost-effective, and interpretable solution for LVEF assessment in community settings. From a machine learning interpretable perspective, the proposed method emphasized CMSE's clinical potential in capturing autonomic dynamics and cardiac function fluctuations.
Authors
Zhang Zhang, Pan Pan, Yang Yang, Huang Huang, Yin Yin, Lin Lin, Huang Huang, Ding Ding, Zou Zou, Zheng Zheng, Zhang Zhang
View on Pubmed