Related Experiment Video
Updated: Jun 3, 2025
10:14
Author Spotlight: Enhancing Neurorehabilitation Through EEG, Motor Imagery, and Virtual Reality
Published on: May 10, 2024
885
Enhancing Deep-Learning Classification for Remote Motor Imagery Rehabilitation Using Multi-Subject Transfer Learning
Joharah Khabti1,2, Saad AlAhmadi1,2, Adel Soudani1,2
1College of Computer and Information Sciences (CCIS), King Saud University, Riyadh 11543, Saudi Arabia.
Sensors (Basel, Switzerland)
|January 8, 2025
Summary
This study introduces a multi-subject transfer learning framework for remote motor imagery (MI) training using brain-computer interfaces (BCIs). The approach enhances accuracy and efficiency for flexible rehabilitation in an IoT environment.
Area of Science:
- Neuroscience
- Computer Science
- Rehabilitation Engineering
Background:
- Electroencephalogram (EEG)-based brain-computer interfaces (BCIs) show promise for motor rehabilitation via motor imagery (MI) tasks.
- Current MI training necessitates physical attendance, limiting rehabilitation flexibility.
- Remote MI training faces challenges in accurate task recognition, computation, and communication costs, especially with complex EEG data and subject-dependent variations.
Purpose of the Study:
- To propose a multi-subject transfer-learning approach for an efficient motor imagery training framework.
- To develop an IoT architecture integrating cloud/edge computing for enhanced system efficiency and reduced network resource usage.
- To improve the accuracy and efficiency of remote MI training for rehabilitation.
Main Methods:
- Implemented a multi-subject transfer-learning approach within an IoT architecture featuring cloud/edge computing.
- Utilized deep learning classification (with and without channel selection) in the cloud.
- Applied multi-subject transfer-learning classification at the edge node, experimenting with various transfer-learning strategies.
Main Results:
- The proposed framework significantly enhanced average accuracy in both multi-subject and single-subject transfer-learning classification.
- Three-subject transfer learning achieved up to 79.77% accuracy (FCNNA model without channel selection).
- Transfer learning improved average accuracy by up to 6.55% (two-subject) and 12.19% (single-subject) compared to non-transfer learning methods.
Conclusions:
- The developed framework offers a viable solution for remote motor imagery rehabilitation.
- It provides accurate motor imagery task recognition while optimizing computational and communication resource usage.
- This approach facilitates flexible and efficient rehabilitation through advanced BCI and IoT integration.
Keywords:
brain–computer interface (BCI)deep learning (DL)edge computingelectroencephalogram (EEG)internet of things (IoT)motor imagery (MI)transfer learning (TL)
