Heartbeat Detection from Ballistocardiogram using Transformer Network

TL;DR

This paper introduces a transformer-based method for detecting heartbeats from ballistocardiogram signals, enabling accurate, non-invasive long-term heart rate monitoring for health assessment.

Contribution

The study presents a novel transformer network architecture specifically designed for heartbeat detection from BCG signals, demonstrating high accuracy across diverse datasets.

Findings

Segment model achieved correlation of 0.97 for HR.

Method shows promise for in-home cardiovascular monitoring.

Effective across different age groups.

Abstract

Longitudinal monitoring of heart rate (HR) and heart rate variability (HRV) can aid in tracking cardiovascular diseases (CVDs), sleep quality, sleep disorders, and reflect autonomic nervous system activity, stress levels, and overall well-being. These metrics are valuable in both clinical and everyday settings. In this paper, we present a transformer network aimed primarily at detecting the precise timing of heart beats from predicted electrocardiogram (ECG), derived from input Ballistocardiogram (BCG). We compared the performance of segment and subject models across three datasets: a lab dataset with 46 young subjects, an elder dataset with 28 elderly adults, and a combined dataset. The segment model demonstrated superior performance, with correlation coefficients of 0.97 for HR and mean heart beat interval (MHBI) when compared to ground truth. This non-invasive method offers significant potential for long-term, in-home HR and HRV monitoring, aiding in the early indication and prevention of cardiovascular issues.