Khairul Anam, Naufal Ainur Rizal, Zamroni Ilyas, Cries Avian, Aris Zainul Muttaqin, Mochamad Edoward Ramadhan, Dwiretno Istiyadi Swasono
Purpose: Individuals with hand amputations often seek prosthetic devices that closely mimic the function of natural limbs and interact with them as if they were a natural part of their bodies. Achieving this level of integration requires implementing a control system that estimates joint angles using surface electromyography (EMG), which offers simultaneous and proportional control. However, accurately predicting the movements of kinematic joints, which have multiple degrees of freedom, is complex and challenging. Methods: This study evaluated the effectiveness of a potential solution and assessed the efficacy of long short-term memory (LSTM) algorithms in decoding multi-degree-of-freedom finger joints. Two different testing scenarios were conducted to evaluate the LSTM model. In the first scenario, the model was tested using publicly available data that involved feature extraction and dimensionality reduction techniques. The second scenario focused on assessing the traditional LSTM model’s efficacy using raw electromyography (EMG) data without employing feature extraction or dimensionality reduction methods. Results: In the first scenario, the LSTM model showcased exceptional performance, achieving an R-squared (R2) value of 0.939. This was accomplished using a mean normalized power method for feature extraction and independent component analysis to reduce dimensionality. On the other hand, in the second scenario, the experimental results indicated a validation performance with an R2 value of 0.811 ± 0.051, while the testing performance had an R2 value of 0.658 ± 0.155. The later result deviates from the performance observed in the preceding scenario. Additionally, the findings demonstrated that the LSTM model could predict the target angle with a normalized root mean square error (NRMSE) of about 0.2. Conclusion: The real-time experiment shows the efficacy of the LSTM model in predicting finger joint movements. Future research will investigate the real-time implementation of edge computing to develop a portable system for many rehabilitation applications. © The Author(s), under exclusive licence to The Brazilian Society of Biomedical Engineering 2024.
Department of Electrical Engineering, University of Jember, Jember, Indonesia; Universitas Muhammadiyah Jember, Jember, Indonesia; Department of Electrical Engineering, Brawijaya University, Malang, Indonesia; Department of Mechanical Engineering, University of Jember, Jember, Indonesia; Department of Computer Engineering, University of Jember, Jember, Indonesia
