Smooth and accurate predictions of joint contact force time-series in gait using over parameterised deep neural networks

The aging process is commonly accompanied by a general or specific loss of muscle mass, force and/or function that inevitably impact on a person’s quality of life. To date, various clinical tests and assessments are routinely performed to evaluate the biomechanical status of an individual, to support and inform the clinical management and decision-making process (e.g., to design a tailored rehabilitation program). However, these assessments (e.g., gait analysis or strength measures on a dynamometer) are typically conducted independently from one another or at different time points, providing clinicians with valuable yet fragmented information. We hereby describe a comprehensive protocol that combines both in vivo measurements (maximal voluntary isometric contraction test, superimposed neuromuscular electrical stimulation, electromyography, gait analysis, magnetic resonance imaging, and clinical measures) and in silico methods (musculoskeletal modeling and simulations) to enable the full characterization of an individual from the biomechanical standpoint. The protocol, which requires approximately 4 h and 30 min to be completed in all its parts, was tested on twenty healthy young participants and five elderlies, as a proof of concept. The implemented data processing and elaboration procedures allowing for the extraction of several biomechanical parameters (including muscle volumes and cross-sectional areas, muscle activation and co-contraction levels) are thoroughly described to enable replication. The main parameters extracted are reported as mean and standard deviation across the two populations, to highlight the potential of the proposed approach and show some preliminary findings (which were in agreement with previous literature).

Section: Discussionmentioning

confidence: 99%

Towards a comprehensive biomechanical assessment of the elderly combining in vivo data and in silico methods

Davico,

Labanca,

Gennarelli

et al. 2024

“…Future study shall consider a well-designed experimental setup that match the real badminton court under training and match conditions. Second, only the discrete and key datapoints were applied to train and test the intelligent statistical models, without considering the time-varying features; thus, other machine learning algorithms, deep learning algorithms, and convolutional neural networks (such as long short-term memory, LSTM) may be utilized for the monitoring and prediction of loading accumulation ( Shao et al, 2022 ; Liew et al, 2023 ).…”

Section: Discussionmentioning

confidence: 99%

Intelligent prediction of lower extremity loadings during badminton lunge footwork in a lab-simulated court

Jiang

Mei

et al. 2023

Self Cite

Introduction: Playing badminton has been reported with extensive health benefits, while main injuries were documented in the lower extremity. This study was aimed to investigate and predict the knee- and ankle-joint loadings of athletes who play badminton, with “gold standard” facilities. The axial impact acceleration from wearables would be used to predict joint moments and contact forces during sub-maximal and maximal lunge footwork.Methods: A total of 25 badminton athletes participated in this study, following a previously established protocol of motion capture and musculoskeletal modelling techniques with the integration of a wearable inertial magnetic unit (IMU). We developed a principal component analysis (PCA) statistical model to extract features in the loading parameters and a multivariate partial least square regression (PLSR) machine learning model to correlate easily collected variables, such as the stance time, approaching velocity, and peak accelerations, with knee and ankle loading parameters (moments and contact forces).Results: The key variances of joint loadings were observed from statistical principal component analysis modelling. The promising accuracy of the partial least square regression model using input parameters was observed with a prediction accuracy of 94.52%, while further sensitivity analysis found a single variable from the ankle inertial magnetic unit that could predict an acceptable range (93%) of patterns and magnitudes of the knee and ankle loadings.Conclusion: The attachment of this single inertial magnetic unit sensor could be used to record and predict loading accumulation and distribution, and placement would exhibit less influence on the motions of the lower extremity. The intelligent prediction of loading patterns and accumulation could be integrated to design training and competition schemes in badminton or other court sports in a scientific manner, thus preventing fatigue, reducing loading-accumulation-related injury, and maximizing athletic performance.

“…In term of model prediction, a recent study reported that including data on walking and running was no better than including walking or running alone (B. X. Liew et al, 2023 ). Nevertheless, a more diverse dataset in machine learning can benefit the model by improving generalization, reducing bias, and enhancing robustness to variations in individual characteristics.…”

Section: Discussionmentioning

confidence: 99%

Performance of multiple neural networks in predicting lower limb joint moments using wearable sensors

Altai,

Boukhennoufa,

Zhai

et al. 2023

Self Cite

Joint moment measurements represent an objective biomechemical parameter in joint health assessment. Inverse dynamics based on 3D motion capture data is the current 'gold standard’ to estimate joint moments. Recently, machine learning combined with data measured by wearable technologies such electromyography (EMG), inertial measurement units (IMU), and electrogoniometers (GON) has been used to enable fast, easy, and low-cost measurements of joint moments. This study investigates the ability of various deep neural networks to predict lower limb joint moments merely from IMU sensors. The performance of five different deep neural networks (InceptionTimePlus, eXplainable convolutional neural network (XCM), XCMplus, Recurrent neural network (RNNplus), and Time Series Transformer (TSTPlus)) were tested to predict hip, knee, ankle, and subtalar moments using acceleration and gyroscope measurements of four IMU sensors at the trunk, thigh, shank, and foot. Multiple locomotion modes were considered including level-ground walking, treadmill walking, stair ascent, stair descent, ramp ascent, and ramp descent. We show that XCM can accurately predict lower limb joint moments using data of only four IMUs with RMSE of 0.046 ± 0.013 Nm/kg compared to 0.064 ± 0.003 Nm/kg on average for the other architectures. We found that hip, knee, and ankle joint moments predictions had a comparable RMSE with an average of 0.069 Nm/kg, while subtalar joint moments had the lowest RMSE of 0.033 Nm/kg. The real-time feedback that can be derived from the proposed method can be highly valuable for sports scientists and physiotherapists to gain insights into biomechanics, technique, and form to develop personalized training and rehabilitation programs.