Work

Heart Rate Estimation Using Finger-Worn Accelerometers

Year: 2025
Type: dissertation
Abstract: The purpose of this thesis is to examine the possibility of estimating heart rate with an integrated accelerometer using a finger-worn ring. In modern wearable devices heart rate monitoring is a highly common feature, with most using photoplethysmography (PPG) for this purpose. However, a clear challenge with PPG sensors is their relatively high power consumption, which is problematic for small devices with limited power supplies. This study presents a more power-efficient alternative method for heart rate estimation based on ballistocardiography (BCG). The objectives of this study are to determine whether heart rate estimation is possible from the finger using only accelerometers and whether this approach could be implemented into a commercial smart ring device to track sleep and heart rate. To address these objectives, a custom-made ring was developed, containing an integrated accelerometer capable of recording 3-axis movement data. This ring was used to create a dataset containing BCG data from 23 subjects, with corresponding 3-lead ECG recordings used as the reference signal. Subsequently, an autocorrelation-based algorithm was developed to estimate heart rate from the BCG recordings using 10-second windows for analysis. To study the possibility of integrating this solution into a commercial smart ring, more accurate estimations were attempted by filtering out signal windows with poor quality. This was accomplished by calculating several statistical metric values for each signal portion and setting threshold values that the metrics had to satisfy. Signal portions that failed to meet these criteria were excluded from heart rate estimation. When comparing the heart rate estimations produced by the proposed algorithm to the reference ECG values without using signal quality-based filtration, an average mean absolute error of 1.88 bpm (beats per minute) was obtained. With the statistical signal quality filtration, this value improved to 1.40 bpm. Without the signal quality assessment, estimations were within ±5 bpm 93.3 % of the time. When filtration was applied, this increased to 96.2 %. The results demonstrate that this method has clear potential for implementation in commercial devices for heart rate estimation. This method performs especially well in scenarios with minimal motion artifacts, such as during sleep. Modern wearable devices already use accelerometers for sleep tracking, and implementing this method could eliminate the need for additional sensors.
Author: Nasri, Omar
Language: English
Open Access status: green
OpenAlex ID: W6991613483
Cites: 66
Cited by: 1
Related to: 10