Predicting gait events from tibial acceleration in rearfoot running: A structured machine learning approach

Abstract

Background: Gait event detection of the initial contact and toe off is essential for running gait analysis, allowing the derivation of parameters such as stance time. Heuristic-based methods exist to estimate these key gait events from tibial accelerometry. However, these methods are tailored to very specific acceleration profiles, which may offer complications when dealing with larger data sets and inherent biological variability. Research question: Can a structured machine learning approach achieve a more accurate prediction of running gait event timings from tibial accelerometry, compared to the previously utilised heuristic approaches? Methods: Force-based event detection acted as the criterion measure in order to assess the accuracy, repeatability and sensitivity of the predicted gait events. 3D tibial acceleration and ground reaction force data from 93 rearfoot runners were captured. A heuristic method and two structured machine learning methods were employed to derive initial contact, toe off and stance time from tibial acceleration signals. Results: Both a structured perceptron model (median absolute error of stance time estimation: 10.00 ± 8.73 ms) and a structured recurrent neural network model (median absolute error of stance time estimation: 6.50 ± 5.74 ms) significantly outperformed the existing heuristic approach (median absolute error of stance time estimation: 11.25 ± 9.52 ms). Thus, results indicate that a structured recurrent neural network machine learning model offers the most accurate and consistent estimation of the gait events and its derived stance time during level overground running. Significance: The machine learning methods seem less affected by intra-and inter-subject variation within the data, allowing for accurate and efficient automated data output during rearfoot overground running. Furthermore offering possibilities for real-time monitoring and biofeedback during prolonged measurements, even outside the laboratory.

Recent Machine Learning Progress in Lower Limb Running Biomechanics With Wearable Technology: A Systematic Review

“…Three types of IMUs sensors are utilized, including commercial sensors, custom-built sensors, and virtual sensors. All studies contain acceleration data with the range of accelerometers from ±6 to ±50 g. Gyroscope data were not contained in nine articles (Watari et al, 2018a , b ; Dixon et al, 2019 ; Komaris et al, 2019 ; Tan et al, 2019 ; Derie et al, 2020 ; Matijevich et al, 2020 ; Johnson et al, 2021 ; Robberechts et al, 2021 ) and for most of the studies, magnetometer data were not incorporated in the IMU sensor (Watari et al, 2018a , b ; Zrenner et al, 2018 ; Dixon et al, 2019 ; Komaris et al, 2019 ; Stetter et al, 2019 , 2020 ; Tan et al, 2019 ; Derie et al, 2020 ; Koska and Maiwald, 2020 ; Matijevich et al, 2020 ; Young et al, 2020 ; Hernandez et al, 2021 ; Johnson et al, 2021 ; Rapp et al, 2021 ; Robberechts et al, 2021 ).…”

“…Different types of outdoor terrain (Dixon et al, 2019 ), inclinations of the running surface (Ahamed et al, 2019 ), and environmental weather conditions (Ahamed et al, 2018 ) were detected and classified in three studies. The accuracy of gait event and spatiotemporal parameter detections was also tested (Zrenner et al, 2018 ; Tan et al, 2019 ; Liu et al, 2020 ; Robberechts et al, 2021 ). Two studies aimed at the running pattern or level classification (Clermont et al, 2019a ; Liu et al, 2020 ).…”

“…Discrete biomechanical variables measured from wearable sensors were also selected as input (Watari et al, 2018a , b ; Clermont et al, 2019a ). Three studies extracted the statistical features in raw sensor signals as the attributes (Derie et al, 2020 ; Matijevich et al, 2020 ; Robberechts et al, 2021 ). The predicted outcomes of the regression algorithms were compared with the ground truth reference data, which was measured from the optical motion capture system and force plate.…”

Predicting vertical ground reaction forces from 3D accelerometry using reservoir computers leads to accurate gait event detection

“…Using machine learning for activity recognition and gait phase recognition based on gait features extracted from biomechanical data, which may be measured with wearable sensors 68,69 has become increasingly popular. The most common techniques to classify gait events or predict GRF waveforms using machine learning are hidden Markov models 26,30,33,34,37,43,[70][71][72][73] , neural networks such as deep neural networks (more than 1 hidden layer) 25,35,36,44,47,52 ; feed-forward neural networks 48,50,53,[56][57][58]74,75 ; long short-term models 24,28,38,54,74,75 ; convolutional neural networks 29,55 ; support vector machines 42,44,76 ;…”

“…(multilayer) perceptron models 28,42,49,51,77 , as well as random forest classifiers 36,42,44 , K-nearest neighbours 42,54,78 , and other types of machine learning using, e.g., Bayesian models 31,32,73,76 , Gaussian mixture model 41 , and principal component analysis 39,40,51,73 . Echo state networks have the great advantage of low computational costs while still showing excellent performance.…”