Real-Time Prediction of Joint Forces by Motion Capture and Machine Learning

Abstract

Conventional biomechanical modelling approaches involve the solution of large systems of equations that encode the complex mathematical representation of human motion and skeletal structure. To improve stability and computational speed, being a common bottleneck in current approaches, we apply machine learning to train surrogate models and to predict in near real-time, previously calculated medial and lateral knee contact forces (KCFs) of 54 young and elderly participants during treadmill walking in a speed range of 3 to 7 km/h. Predictions are obtained by fusing optical motion capture and musculoskeletal modeling-derived kinematic and force variables, into regression models using artificial neural networks (ANNs) and support vector regression (SVR). Training schemes included either data from all subjects (LeaveTrialsOut) or only from a portion of them (LeaveSubjectsOut), in combination with inclusion of ground reaction forces (GRFs) in the dataset or not. Results identify ANNs as the best-performing predictor of KCFs, both in terms of Pearson R (0.89–0.98 for LeaveTrialsOut and 0.45–0.85 for LeaveSubjectsOut) and percentage normalized root mean square error (0.67–2.35 for LeaveTrialsOut and 1.6–5.39 for LeaveSubjectsOut). When GRFs were omitted from the dataset, no substantial decrease in prediction power of both models was observed. Our findings showcase the strength of ANNs to predict simultaneously multi-component KCF during walking at different speeds—even in the absence of GRFs—particularly applicable in real-time applications that make use of knee loading conditions to guide and treat patients.

Prediction of Tool Forces in Manual Grinding Using Consumer-Grade Sensors and Machine Learning

“…Other influences such as the wear of the grinding disc [1], state of charge of the battery in hand-held cordless grinders, and lower stiffness of the tools complicate the prediction. The comparison with the prediction of body forces and movements such as knee joint forces (KJF) with one or more IMUs shows a similar approach (leave-one-out cross-validation and classical and deep learning approaches) for regression [23][24][25][26][27][28]. Similar accuracies are obtained, for example r = 0.60 to r = 0.94 for vertical KJF, r = 0.64 to r = 0.90 for anterior-posterior KJF, r = 0.25 to r = 0.60 medial-lateral KJF [23]) despite a different system being investigated, and more scatter is caused by the grinding machine.…”

“…In comparison with SVM, linear regression, Gaussian kernel, random forests, and decision trees, GPR showed superiority in a preliminary study in terms of accuracy and stability of results. Due to the variety of algorithms available for regression, it cannot be excluded that other methods such as neural networks, which have been successfully used for similar problems [23,24,[35][36][37], would have led to even better accuracy. However, especially for deep learning methods with a very large number of parameters, a large data set must be available to demonstrate the transferability of the results.…”

“…In other fields, determining forces with machine learning methods based on measurement data from consumer-grade sensors, e.g., from an inertial measurement unit (IMU), is a common problem. Thus, body forces and motions, such as joint forces [23,24], joint torques [25], locomotor intentions [26], center of mass-center of pressure inclination angle [27], and the ground reaction force [28] have been predicted using one or more IMUs. The results indicate that the prediction of the motion and forces of a user is in principle also possible in manual grinding with an IMU.…”

A comparison of machine learning models’ accuracy in predicting lower-limb joints’ kinematics, kinetics, and muscle forces from wearable sensors

“…While previous studies concentrated on joint range of motion in the sagittal plane, our study additionally included the pelvis in all plane, hip int/ext rotation and abd/add and ankle inv/eversion. Fewer studies were conducted to investigate the performance of different ML models for joint kinetics estimation [12,[20][21][22] compared to joint kinematics. All previous studies would focus on a speci c lower-limb joint kinetics (e.g.…”

Classification of Walking Environments Using Deep Learning Approach Based on Surface EMG Sensors Only

“…The output layer then classified the input data as being from one of the five walking environments. A Softmax cross-entropy with logits was employed as a loss function, and Adaptive Moment Estimation (Adam) was used as an optimization algorithm to minimize the loss function [ 38 , 39 , 40 , 41 ].…”