Nightbeat: Heart Rate Estimation From a Wrist-Worn Accelerometer During Sleep

TL;DR

The first dataset of wrist-worn accelerometer recordings and electrocardiogram references in uncontrolled at-home settings is presented to investigate the recent promise of IMU-only HR estimation via ballistocardiograms and introduces a frequency-based method to extract HR via curve tracing from IMU recordings while rejecting motion artifacts.

Abstract

Today's fitness bands and smartwatches typically track heart rates (HR) using optical sensors. Large behavioral studies such as the UK Biobank use activity trackers without such optical sensors and thus lack HR data, which could reveal valuable health trends for the wider population. In this paper, we present the first dataset of wrist-worn accelerometer recordings and electrocardiogram references in uncontrolled at-home settings to investigate the recent promise of IMU-only HR estimation via ballistocardiograms. Our recordings are from 42 patients during the night, totaling 310 hours. We also introduce a frequency-based method to extract HR via curve tracing from IMU recordings while rejecting motion artifacts. Using our dataset, we analyze existing baselines and show that our method achieves a mean absolute error of 0.88 bpm -- 76% better than previous approaches. Our results validate the potential of IMU-only HR estimation as a key indicator of cardiac activity in existing longitudinal studies to discover novel health insights. Our dataset, Nightbeat-DB, and our source code are available on GitHub: https://github.com/eth-siplab/Nightbeat.

Figures (5)

• Figure 1: Our dataset comprises continuous motion signals from a wrist-worn 3-axis accelerometer (Axivity AX3, 100 Hz) and corresponding ECG signals (movisens EcgMove 4) from 42 patients and nights. From the ballistocardiogram (BCG) captured by the vibrations reaching the wrist-based sensor, our signal processing method estimates the patient's heart rate using a combination of filters and heuristics for pre-processing,a short-term frequency analysis to identify HR curve, and a filter stage to precisely select the peak of the BCG wave for inter-beat interval detection. We combine the HR curve with the detected inter-beat intervals and a 5-minute median smooth to make a prediction for every 20-second window.
• Figure 2: We detect motion artifacts during 3-minute windows based on the energy of the STFT of the Nightbeat signal over time. If the energy lies 5 standard deviations ($\sigma$; estimated robustly via the IQR) above the median, we flag the segment as a motion artifact.
• Figure 3: Peak detection (purple crosses) from the Nightbeat signal (blue) for an exemplary 15-second signal. The smoothed and filtered signal is displayed in orange, the height threshold for peak detection in green.
• Figure 4: Comparison of Prediction and ground-truth HR over time for P11 of the Nightbeat-DB dataset. Using our Nightbeat HR estimation method, we achieve an MAE of 0.41 and a correlation of 0.98.
• Figure 5: Predictions against ground-truth HR for all participants of the Nightbeat-DB dataset---colored by participants. Across all participants, our Nightbeat Hr estimation method achieves a correlation of 0.94.