Predicting continuous ground reaction forces from accelerometers during uphill and downhill running: a recurrent neural network solution

• Published in: PeerJ (San Francisco, CA) Vol. 10; p. e12752
• Main Authors: Alcantara, Ryan S., Edwards, W. Brent, Millet, Guillaume Y., Grabowski, Alena M.
• Format: Journal Article
• Language: English
• Published: United States PeerJ. Ltd 04.01.2022; PeerJ, Inc; PeerJ Inc
• Subjects: Accelerometers; Accelerometry; Accuracy; Biomechanics; Computational linguistics; Data Mining and Machine Learning; Exercise Test; Fitness equipment; GRF; Humans; IMU; Kinematics; Laboratories; Language processing; LSTM; Machine learning; Movement; Natural language interfaces; Neural networks; Neural Networks, Computer; RNN; Running; Sacrum; Somatotropin releasing hormone; Variables; Wearable computers; United States > US; Biomechanics; IMU; LSTM; RNN; Machine learning; Biofeedback; GRF
• ISSN: 2167-8359; 2167-8359
• DOI: 10.7717/peerj.12752

Ground reaction forces (GRFs) are important for understanding human movement, but their measurement is generally limited to a laboratory environment. Previous studies have used neural networks to predict GRF waveforms during running from wearable device data, but these predictions are limited to the stance phase of level-ground running. A method of predicting the normal (perpendicular to running surface) GRF waveform using wearable devices across a range of running speeds and slopes could allow researchers and clinicians to predict kinetic and kinematic variables outside the laboratory environment. We sought to develop a recurrent neural network capable of predicting continuous normal (perpendicular to surface) GRFs across a range of running speeds and slopes from accelerometer data. Nineteen subjects ran on a force-measuring treadmill at five slopes (0°, ±5°, ±10°) and three speeds (2.5, 3.33, 4.17 m/s) per slope with sacral- and shoe-mounted accelerometers. We then trained a recurrent neural network to predict normal GRF waveforms frame-by-frame. The predicted versus measured GRF waveforms had an average ± SD RMSE of 0.16 ± 0.04 BW and relative RMSE of 6.4 ± 1.5% across all conditions and subjects. The recurrent neural network predicted continuous normal GRF waveforms across a range of running speeds and slopes with greater accuracy than neural networks implemented in previous studies. This approach may facilitate predictions of biomechanical variables outside the laboratory in near real-time and improves the accuracy of quantifying and monitoring external forces experienced by the body when running.