(P118) EXTRACTING HEART RATE FROM BALLISTOCARDIOGRAPH SIGNALS COLLECTED FROM NON-CONTACT SENSORS WITHOUT ELECTROCARDIOGRAM GATING

Background: Remote heart monitoring is crucial for better management of heart disease. We have developed a system that can acquire a continuous ballistocardiograph (BCG), a micro-vibratory signal generated by cardiac ejection and subsequent circulation, without requiring skin contact or user interaction. This system empowers remote, unobtrusive monitoring of cardiac status, which may serve patients with physical and/or cognitive limitations. The mechanical vibrations of BCG can be converted to physiologic properties of the circulation. Here we present an algorithm to extract heart rate (HR) from the BCG recording, without requiring an electrocardiographic (ECG) signal for gating.

METHODS AND RESULTS: BCG was collected from 50 healthy subjects. Subjects were asked to stand on a weight scale and perform the Valsalva Maneuver (VM). We evaluated the efficacy of measuring HR in two regions: a stable region with steady HR and a dynamic region of changing HR due to the VM. After cleaning BCG via a bandpass filter, we used a root-mean-square (RMS) of the BCG signal to locate segments of low noise in the stable region. Segments that fell below the threshold set using the RMS of BCG were labelled as clean regions. Four 10-second clean regions were determined per subject. The dynamic region was further divided into regions of breath-hold (BH) and post-release (PR). A 10-second segment was extracted for both conditions. Beat-by-beat intervals of BCG were calculated for each segment based on detected peaks via thresholding. Ten-second averages of the intervals were calculated and validated against ECG.

The predicted HR showed a correlation coefficient of r=0.92 (p <.001) in the clean region, r=0.66 (p <.001) in the BH region, and r=0.78 (p <.001) in the PR region against the ECG standard (Figure 1). A Bland-Altman test showed that 80% of clean, 40% of BH, 58% of PR data points fell within acceptable limits of agreement (± 5 bpm deviation) with ECG (Figure 2).

Conclusion: This study showcases a modality that enables HR acquisition without any attachments. In the clean, static HR region, the algorithm performs more reliably. However, it is less effective in the dynamic HR regions (BH and PR) due to movement artifacts. Future efforts involve improving BCG peak detection sensitivity for better HR correlation. Collecting data with a more stable posture (e.g. supine, seated) could also improve signal quality. This project lays the groundwork for an unobtrusive heart estimation tool and shows potential to monitor relevant conditions such as arrhythmia.