TY  - EJOU
AU  - Rasheed, Faiza 
AU  - Zheng, Jinchuan 
AU  - Lizama, Luis Eduardo Cofré 
AU  - Martin, Suzanne 
AU  - Tse, Kwong Ming 

TI  - Multi-Surface Knee Joint Kinematics Estimation Using Hybrid Stacked Long Short-Term Memory-Multilayer Perceptron Network
T2  - Computer Modeling in Engineering \& Sciences

PY  - 
VL  - 
IS  - 
SN  - 1526-1506

AB  - The estimation of lower limb joint kinematics has great potential to be used for different applications, for instance, gait analysis, diagnosis of any joint injury, diagnosis of any other lower limb injury, prosthesis control, etc. The natural and controlled interaction between human and lower limb prosthesis is very important. Hybrid Stacked Long Short-Term Memory-Multilayer Perceptron (HS-LSTM-MLP) network is hypothesized to estimate knee joint angle individually over five different surfaces: level ground, ramp ascent, ramp descent, stair ascent, and stair descent. The spatial and temporal information extracted from reflective markers is used as input features, and the calculated knee joint angles are used as the target variable. The standard performance metrics for such regression tasks, including coefficient of determination (R<sup>2</sup>), mean squared error (MSE), and root mean square error (RMSE), are employed here for proposed model evaluation. The proposed HS-LSTM-MLP model is compared with the state-of-the-art Bidirectional Long Short-Term Memory (BiLSTM), Long Short-Term Memory (LSTM), and Multilayer Perceptron (MLP) neural networks to evaluate its efficiency for knee joint angle estimation. The HS-LSTM-MLP model outperforms BiLSTM, LSTM, and MLP with R<sup>2</sup> (above 0.9513, 0.9435, 0.9340, and 0.9111, respectively) and RMSE (below 4.6395°, 5.1667°, 6.2338°, and 7.6487°, respectively). The estimation results are simulated in MATLAB for a proof-of-concept for potential application in transfemoral powered prostheses. The proposed model HS-LSTM-MLP exhibits significant potential for precise and reliable knee joint angle estimation and to enable amputees in the future to walk over different surfaces with more efficient knee joint control and greater stability.
KW  - Knee joint angle estimation; knee joint kinematics prediction; multilayer perceptron; hybrid stacked long short-term memory-multilayer perceptron; prosthesis assistance

DO  - 10.32604/cmes.2026.086130