Accurate Step Count With Generalizable Deep Learning on Accelerometer Data
Long Luu, Arvind Pillai, Halsey Lea, Rubén Buendía, Faisal M. Khan, Glynn Dennis · 2021 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech) · 2021
Physical Activity (PA) is a globally recognized pillar of general health. However, there is no widely accepted measure to quantify PA. Step count, as one measure of PA, is a well known predictor of long term morbidity and mortality. Although step count is widely available in consumer grade mobile and wearable devices, a lack of methodological stan-dards and clinical validation remains a major impediment to step count being accepted as a valid clinical endpoint. Previous works have mainly focused on developing device-specific step count algorithms in controlled experimental settings, and often employ sensor modalities such as gyroscopes that may not be widely available. This limits step count suitability in clinical scenarios as algorithms might not generalize across devices and patient populations. In this paper, we trained neural network models on publicly available data and directly tested on an independent cohort without further tuning. Herein we assess step count accuracy of neural network architectures trained on accelerometer-only activity signals from one device and applied to accelerometer-only data obtained from a separate subject cohort wearing multiple distinct devices. We found highly accurate step count estimates using accelerometer-only activity signals (96 % - 99 % with the best model). The results demonstrate that it is possible to develop device-agnostic, accelerometer-only algorithms that provide highly accurate step count. Thus it positions step count as a reliable mobility endpoint and a strong candidate for clinical validation.