Robust heart rate estimation using wrist-based PPG signals in the presence of intense physical activities

Abstract

Heart rate tracking from a wrist-type photoplethysmogram (PPG) signal during intensive physical activities is a challenge that is attracting more attention thanks to the introduction of wrist-worn wearable computers. Commonly-used motion artifact rejection methods coupled with simple periodicity-based heart rate estimation techniques are incapable of achieving satisfactory heart rate tracking performance during intense activities. In this paper, we propose a two-stage solution. Firstly, we introduce an improved spectral subtraction method to reject the spectral components of motion artifacts. Secondly, instead of using heuristic mechanisms, we formalize the spectral peaks selection process as the shortest path search problem and validate its effectiveness. Analysis on the experimental results based on a published database shows that: (1) Our proposed method outperforms three other comparable methods with regards to heart rate estimation error. (2) The proposed method is a promising candidate for both offline cardiac health analysis and online heart rate tracking in daily life, even during intensive physical motions.

Investigating sources of inaccuracy in wearable optical heart rate sensors

“…Our analyses reinforces the growing body of research demonstrating that wearable devices have higher error during activity than at rest. 8,11,35,38,39 We further demonstrate that the directionality of the HR error is dependent on the activity type. Reliable HR data across all activity types and levels is key to enabling digital biomarker development and to supporting clinical research studies that involve physiologic monitoring during physical activity or exercise interventions.…”

“…[35][36][37] Several studies have demonstrated that HR measurements from wearable devices are often less accurate during physical activity or cyclic wrist motions. 8,11,35,38,39 Several research groups and manufacturers have identified that cyclical motion can affect accuracy of HR in wearable sensors. 9,10,15 The cyclical motion challenge has been described as a "signal crossover" effect wherein the optical HR sensors on wearables tend to lock on to the periodic signal stemming from the repetitive motion (e.g., walking and jogging) and mistake that motion as the cardiovascular cycle.…”

Accuracy in wrist-worn, sensor-based measurements of heart rate and energy expenditure in a diverse cohort

“…Skin tone at the wrist was rated independently by two of the investigators using the Von Luschan Chromatic scale , and the average rating was then transformed to the Fitzpatrick skin tone scale (1)(2)(3)(4)(5)(6). 11 Maximal oxygen uptake (VO2max) was measured by incremental tests in running (n = 32) or cycling (n = 6) to volitional exhaustion, or estimated from the submaximal cycling stages (n = 22) using the Åstrand nomogram (Åstrand 1960).…”

“…Microelectromechanical systems such as accelerometers and Light Emitting Diode (LED)based physiological monitoring have been available for decades. [2][3][4][5][6][7] More recent improvements in battery longevity and miniaturization of the processing hardware to turn raw signals in real time into interpretable data led to the commercial development of wrist worn devices for physiological monitoring. Such devices can provide data directly back to the owner and place estimates of heart rate (HR) and energy expenditure (EE) within a consumer model of health and fitness.…”

How Nonlinear-Type Time-Frequency Analysis Can Help in Sensing Instantaneous Heart Rate and Instantaneous Respiratory Rate from Photoplethysmography in a Reliable Way

“…For the IHR, methods include time-frequency (TF) analyses (Gil et al, 2010; Mullan et al, 2015; Wu et al, 2016), adaptive filtering (Yousefi et al, 2014; Khan et al, 2015; Murthy et al, 2015; Schack et al, 2015; Mashhadi et al, 2016), Kalman filter (Frigo et al, 2015), sparse spectrum reconstruction (Zhang, 2015), blind source separation (Wedekind et al, 2015), a Bayesian approach (D'souza et al, 2015; Sun and Zhang, 2015), correntropy spectral density (CSD) (Garde et al, 2014), empirical mode decomposition (EMD) (Zhang et al, 2015), model fitting (Wadehn et al, 2015), deep learning (Jindal, 2016), fusion approaches (Temko, 2015; Zhu S. et al, 2015), etc. For the IRR, efforts include TF analysis (Chon et al, 2009; Orini et al, 2011; Dehkordi et al, 2015), sparse signal reconstruction (Zong and Jafari, 2015; Zhang and Ding, 2016), neural network (Johansson, 2003), modified multi-scale principal component analysis (Madhav et al, 2013), independent component analysis (Zhou et al, 2006), time-varying autoregressive regression (Lee and Chon, 2010b,a), fusion approaches (Karlen et al, 2013; Cernat et al, 2015), pulse-width variability (Lazaro et al, 2013; Cernat et al, 2014), CSD (Pelaez-Coca et al, 2013; Garde et al, 2014), EMD (Garde et al, 2013), a Bayesian approach (Pimentel et al, 2015; Zhu T. et al, 2015), etc. While the above algorithms focus on either IHR or IRR, only a few ad-hoc algorithms are considered to extract simultaneously the IHR and IRR, like (Garde et al, 2013, 2014).…”