Prediction of 3D ground reaction forces during gait based on accelerometer data

Abstract

Introduction: The aim of this study was to predict 3D ground reaction force signals based on accelerometer data during gait, using a feed-forward neural network (MLP). Methods: Seventeen healthy subjects were instructed to walk at a self-selected speed with a 3D accelerometer attached to the distal and anterior part of the shank. A force plate was embedded into the middle of the walkway. MLP neural networks with one hidden layer and three output layers were selected to simulate the anteroposterior (AP), vertical (Vert) and mediolateral (ML) ground reaction forces (GRF). The input layer was composed of fourteen inputs obtained from accelerometer signals, selected based on previous studies. Principal component analysis (PCA) was used to compare the simulated and collected curves. The Pearson correlation coefficient and the mean absolute deviation (MAD) between signals were calculated. Results: PCA identified small, but significant differences between collected and simulated signals in the loading response phases of AP and ML GRF, while Vert did not show differences. The correlation between the simulated and collected signals was high (AP: 0.97; Vert: 0.98; ML: 0.80). MAD was 1.8%BW for AP, 4.5%BW for Vert and 1.4%BW for ML. Conclusion: This study confirmed that multilayer perceptron neural network can predict the highly non-linear relationship of shank acceleration parameters and ground reaction forces, as well as other studies have done using plantar pressure devices. The greater advantages of this device are the low cost and the possibility of use outside the laboratory environment.

Joint Torque Prediction via Hybrid Neuromusculoskeletal Modelling during Gait Using Statistical Ground Reaction Estimates: An Exploratory Study

“…Minimum gradient was used as stopping criteria to enhance the model generalisation. These designs align to those of previous studies for similar applications [ 31 , 55 ]. For every subject, each of the four trials was tested in turn.…”

“…Features selected by NCA were fed into separate ANN feedforward models for GRF and COP estimation in each direction. All the models were trained with one hidden layer [ 31 , 55 ]. The model weights and biases were updated through Levenberg–Marquardt backpropagation [ 56 ].…”

“…These outstanding challenges have motivated researchers to estimate GRF and COP through machine learning. Specifically, marker trajectories collected with motion capture systems and kinematics measured by such wearable sensors as inertial measurement units are employed to drive the statistical models for the estimation of the two gait parameters [ 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 ]. Despite that past literature has shown great potential of this approach to be used in real-world environments, there is still room for improvement in the design and application of statistical models incorporating GRF and COP estimation.…”

Prediction of Lower Limb Kinetics and Kinematics during Walking by a Single IMU on the Lower Back Using Machine Learning

“…The errors of the vertical and A-P GRF were comparable to ours (Table 8). Recently, without a great loss of prediction accuracy, the number of IMUs were reduced to two, attached at each shank of the subject [11]. To complement the reduced sensor information, a feed-forward neural network was used to produce the best estimate of the GRFs.…”

“…To complement the reduced sensor information, a feed-forward neural network was used to produce the best estimate of the GRFs. In this recent study, however, the training datasets included part of the data from the test subject [11], which may have decreased the prediction accuracy when LOO validation was performed. Because the GRF pattern could differ between subjects [39,40], we used a conservative validation method, by excluding the test subjects' data from the training dataset, and following the LOO validation method to guarantee generalization.…”