EEG and EMG Induced Pain-Sensitive Learning Controller forRobotic Knee Rehabilitation Using Deep Learning

Mukherjee, Prithwijit and Roy, Anisha Halder and Konar, Amit and Nagar, Atulya K (2025) EEG and EMG Induced Pain-Sensitive Learning Controller forRobotic Knee Rehabilitation Using Deep Learning. IEEE Access. p. 1. ISSN 2169-3536

Text EEG_and_EMG_Induced_Pain-Sensitive_Learning_Controller_for_Robotic_Knee.pdf - Published Version Available under License Creative Commons Attribution. Download (3MB)

Abstract

Knee pain is a problem of common interest for its increasing prevalence in adult population. It can severely impair mobility and diminish the quality of life of affected individuals. Traditional physiotherapy depends heavily on the expertise of the therapists and is often costly and sometimesinaccessible to many patients. Robotic rehabilitation, under this circumstance, is analternative solution for personalized rehabilitation therapy to these patients.Thisstudyaims at classifying knee pain-levels from the acquired EEG and EMG signals of experimental subjects using an extended Long Short-Term Memory (LSTM) model, and also proposes a novel learning controller that learns the speed-setting of the motors depending on the degree of the classified pain-levels in knee-bending exercises. A complete stand-alone robotic physiotherapeutic system is developed to undertake knee-bending exercises on patients suffering from knee pains. The proposed system learns the pain-levels of experimental subjects at different angular position of knee bending, stores them in the matrix of a Learning controller, and utilizes the learnt experience to modulate motor speeds at different angular knee-bending for the same subject. The proposed system involves 2 classifiers, one to determine subjective intension for left/right leg selection for physiotherapy, and the other to classify pain-levels into 11 classes during knee-bending using P-1000 event related potential and EMG signals. The proposed LSTM-based classifier achieves an accuracy of 95.37% ± 2.53% in knee pain-level classification.The original contribution of the paper includes: i) extension of LSTM classifier architecture by one novel attention module, ii) inclusion of a learning controller with its thorough stability analysis, iii) design and development of the complete stand-alone robotic rehabilitative system for automatic physiotherapy. Feedback taken from healthy people with mild knee pain and clinically designated arthritic patients confirms the superiority of the new proposal over the conventional physiotherapy with respect to individual liking, convenience, trustworthiness and safety.

Item Type:	Article
Additional Information and Comments:	Copyright 2025 The Authors. This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
Faculty / Department:	Faculty of Human and Digital Sciences > School of Computer Science and the Environment
SWORD Depositor:	RISE Symplectic
Depositing User:	RISE Symplectic
Date Deposited:	10 Nov 2025 13:59
Last Modified:	10 Nov 2025 13:59
URI:	https://hira.hope.ac.uk/id/eprint/4781

Actions (login required)

View Item