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

Abstract

With the emergence of wearable technology and machine learning approaches, gait monitoring in real-time is attracting interest from the sports biomechanics community. This study presents a systematic review of machine learning approaches in running biomechanics using wearable sensors. Electronic databases were retrieved in PubMed, Web of Science, SPORTDiscus, Scopus, IEEE Xplore, and ScienceDirect. A total of 4,068 articles were identified via electronic databases. Twenty-four articles that met the eligibility criteria after article screening were included in this systematic review. The range of quality scores of the included studies is from 0.78 to 1.00, with 40% of articles recruiting participant numbers between 20 and 50. The number of inertial measurement unit (IMU) placed on the lower limbs varied from 1 to 5, mainly in the pelvis, thigh, distal tibia, and foot. Deep learning algorithms occupied 57% of total machine learning approaches. Convolutional neural networks (CNN) were the most frequently used deep learning algorithm. However, the validation process for machine learning models was lacking in some studies and should be given more attention in future research. The deep learning model combining multiple CNN and recurrent neural networks (RNN) was observed to extract different running features from the wearable sensors and presents a growing trend in running biomechanics.

Predicting Coordination Variability of Selected Lower Extremity Couplings during a Cutting Movement: An Investigation of Deep Neural Networks with the LSTM Structure

“…For predicting the coordination variability of the three CM directions, we constructed a deep-learning architecture with three layers of LSTM. Following the recommendation of previous research [ 63 , 64 ], we utilized the accelerometers’ raw acceleration data as feature data, which considerably aided our research ( Figure 5 , Figure 6 and Figure 7 ). The model predicts the coordination variability of three couplings in three CM directions with accuracy.…”

Rethinking running biomechanics: a critical review of ground reaction forces, tibial bone loading, and the role of wearable sensors

“…The impact shock has been discussed linked with the incidence of chronic overuse injuries ( Hennig et al, 1993 ). Given the advances of wearable technology in the past twenty decades, trial-axis acceleration and angular velocity could be measured from accelerometer and gyroscope in a single inertial sensor ( Afaq et al, 2020 ; Xiang et al, 2022d ; Xiang et al, 2022e ; Mason et al, 2023 ; Xiang et al, 2024 ; Yamane et al, 2024 ). This made segment acceleration measurements easier and more convenient, shifting the question to: Can we use portable and affordable inertial sensors to evaluate external loading rather than the force plate, which is conventionally embedded in the floor in a gait lab and is cost-prohibitive ( Sheerin et al, 2019 ; Hutabarat et al, 2021 ; Xiang et al, 2022e )?…”

“…One of the most significant advancements in biomechanics facilitated by wearable sensors is their capability to enable data-driven approaches, offering portable and innovative solution ( Halilaj et al, 2018 ; Gholami et al, 2020 ; Hernandez et al, 2021 ; Xiang et al, 2022e ; Mason et al, 2023 ; Xiang et al, 2023 ). Notably, the prediction of GRF metrics from inertial sensors using deep learning algorithms has shown high accuracy, as evidenced in studies ( Ngoh et al, 2018 ; Johnson W. R. et al, 2020 ; Tan et al, 2020 ).…”

Continuous estimation of kinetic variables while walking using an accelerometer and artificial neural networks

“…Most researchers investigated particular ankle, knee, and hip moment in sagittal plane [12], hip moment in sagittal and frontal plane [29], hip moment in frontal plane [30], knee moment in sagittal [31] or frontal plane [13], ankle moment in frontal and sagittal planes [32], and lower limb joints moment in all three planes [18,33]. The previous studies utilized a few subjects and a recent review article by Xiang et al has shown to utilize more subjects (more than 20) [34] and we utilized 73 subjects. We estimated the power, which has been little done in previous research.…”