Deep Learning Techniques in Estimating Ankle Joint Power Using Wearable IMUs

Abstract

Estimating ankle joint power can be used to identify gait abnormities, which is usually achieved by employing a complicated biomechanical model using heavy equipment settings. This paper demonstrates deep learning approaches to estimate ankle joint power from two Inertial Measurement Unit (IMU) sensors attached at foot and shank. The purpose of this study was to investigate deep learning models in estimating ankle joint power in practical scenarios, in terms of variance in walking speeds, reduced number of extracted features and inter-subject model adaption. IMU data was collected from nine healthy participants during five walking trials at different speeds on a force-plate-instrumented treadmill while an optical motion tracker was used as ground truth. Three state-of-the-art deep neural architectures, namely Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN) and, fusion of CNN and LSTM (CNN-LSTM), were developed, trained, and evaluated in predicting ankle joint power by extracting few simple, meaningful features. The proposed architectures were found efficient and promising with higher estimation accuracies (correlation coefficient, R > 0.92 and adjusted R-squared value > 83%) and lower errors (mean squared error < 0.06, and mean absolute error < 0.13) in interparticipant evaluations. Performance evaluations among the three deep regressors showed that LSTM performed comparatively better. Lower standard deviations in mean squared error (0.029) and adjusted R-squared value (5.5%) proved the proposed model's efficiency for all participants.

Inertial Motion Capture-Based Wearable Systems for Estimation of Joint Kinetics: A Systematic Review

“…Among these, the most evaluated activity was walking, which is a fundamental human movement. In 29 of the 46 selected papers, level walking was included in the experimental motions [ 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 , 54 , 55 , 56 , 57 , 58 , 59 , 60 ], and 23 of these discussed only walking [ 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 , 54 ]. While most of thes...…”

“…The most commonly used machine learning technique was neural networks, which were applied in nine studies. Among these, feedforward neural networks (FFNNs) were used in five papers [ 46 , 48 , 50 , 57 , 60 ], the convolutional neural network (CNN) was used in [ 59 ], and the remaining three papers [ 50 , 52 , 54 ] compared different techniques, including FFNN, CNN, and the long short-term memory (LSTM) network (see Table 4 ). Jiang et al [ 45 ] used random forest regression to randomly select a part of the training dataset, and Matijevich et al [ 67 ] trained the model using gradient-boosted decision trees based on an ensemble of decision trees.…”

“…Dorschky et al [ 59 ] used the AP and vertical components of the accelerometer signal and the ML component of the gyroscope signal based on a 2D biomechanical model. Barua et al [ 52 ] used the L2 norm and average extracted from the sensor signal as the feature data instead of using each axial signal, in order to be independent of the sensor attachment orientation for the convenience in practice. Iwama et al [ 53 ] used the peak-to-peak acceleration in the ML direction based on the observation that the peak of the knee adduction moment occurs together with the peak of the ML acceleration immediately after heel contact.…”

“…As a result, the case of using both IMU and PS data for machine learning showed the highest accuracy, which was superior to IDM (NRMSE: IDM = 5.2%, MLM = 2.6%). In three studies [ 50 , 52 , 54 ], the estimation performances of different neural networks were compared. In [ 54 ], three neural networks (multilayer perceptron, CNN, and LSTM) for estimating the 3D joint kinetics of the lower body were compared.…”

Computer-Aided Ankle Ligament Injury Diagnosis from Magnetic Resonance Images Using Machine Learning Techniques

“…In view of this, several studies have been carried out with the aim of proposing Computer Aided Design (CAD)-based medical diagnosis systems for various fields of medicine such as the diagnosis of Parkinson’s disease [ 23 ] and the Alzheimer’s disease [ 24 ], recognition and detection of atrial fibrillation [ 25 , 26 ], virtual nasal endoscopy system [ 27 ], lung nodule detection [ 28 ], oral cancer classification [ 29 ], breast cancer detection [ 30 , 31 ], as well as ankle prostheses and diseases [ 32 , 33 , 34 ], and many more [ 35 , 36 , 37 , 38 ].…”