论文类型：

期刊论文

论文编号：

107885

发表刊物：

Biomedical Signal Processing and Control

收录刊物：

SCI

卷号：

108

关键字：

Atrial fibrillation Ballistocardiogram Temporal–spectral feature fusion Cross-band attention

DOI码：

10.1016/j.bspc.2025.107885

发表时间：

2025-04-29

影响因子：

4.9

摘要：

Atrial fibrillation (AF) is one of the most prevalent cardiac arrhythmias worldwide and significantly increases the risk of ischemic stroke and heart failure, among other cardiovascular conditions. Given the serious health risks posed by atrial fibrillation, particularly in older adults, early screening is essential to prevent disease progression. The usually transient nature of early-stage atrial fibrillation makes continuous monitoring necessary. However, traditional diagnostic methods, such as Holter monitors, require strict operational standards and high levels of patient compliance, making extended home monitoring impractical. The unobtrusive data collection manner of ballistocardiogram (BCG) makes it a feasible option for long-term home cardiovascular health monitoring. However, the variability and diversity in ballistocardiogram data have obstructed its applications. In this study, we introduce FabosNet, a frequency-aware system designed for bedside overnight screening of atrial fibrillation using ballistocardiogram signals. By employing polyvinylidene fluoride (PVDF) film sensors placed on a mattress, our system allows for the unobtrusive collection of continuous ballistocardiogram data. Our classification algorithm leverages multiscale temporal–spectral feature fusion and cross-frequency band attention to detect atrial fibrillation from short segments of ballistocardiogram signals. Testing on a dataset of 70 subjects reveals that FabosNet achieves an accuracy between 95.26% and 96.78% for various segment lengths (5 to 30 s). Compared with other state-of-the-art methods, FabosNet effectively distinguishes between atrial fibrillation patients and those with occasional arrhythmias, highlighting its potential for early screening.