Cortical signals analysis to recognize intralimb mobility using modified RNN and various EEG quantities
Maged S Al-Quraishi1, Wooi Haw Tan2, Irraivan Elamvazuthi3
1Interdisciplinary Research Center for Smart Mobility and Logistics (IRC-SML), King Fahd University of Petroleum & Minerals (KFUPM), Dhahran, 31261, Saudi Arabia.
Heliyon
|May 10, 2024
Summary
Deep learning models like GRU and LSTM show superior accuracy in recognizing foot movements from EEG signals compared to traditional methods. This breakthrough enhances Brain-Computer Interface development for foot rehabilitation and physical therapy.
Area of Science:
- Neuroscience and Biomedical Engineering
- Signal Processing and Machine Learning
Background:
- Electroencephalogram (EEG) signals are vital for predicting sensorimotor activities but struggle with recognizing subtle intralimb movements like foot dorsiflexion/plantar flexion.
- Accurate identification of intralimb movements is crucial for developing effective Brain-Computer Interface (BCI) devices for motor rehabilitation.
Purpose of the Study:
- To investigate the efficacy of various EEG signal features in recognizing intralimb foot movements.
- To develop and evaluate deep learning models for enhanced intralimb movement detection in BCI applications for foot rehabilitation.
Main Methods:
- Collected EEG data from 22 participants using 21 electrodes over the motor cortex, alongside EMG for ankle movement onset.
- Analyzed slow cortical potentials (SCP) and sensorimotor rhythms (SMR) in alpha and beta bands, extracting features like Autoregressive, variance, waveform length, standard deviation, and permutation entropy.
- Developed and compared modified Recurrent Neural Networks (RNNs) including Long Short-term Memory (LSTM) and Gated Recurrent Unit (GRU) against traditional classifiers (SVM, kNN) for movement recognition.
Main Results:
- GRU and LSTM models significantly outperformed conventional machine learning algorithms in recognizing intralimb movements from EEG signal features.
- LSTM achieved accuracies of 98.87% (within-subject) and 87.38% (across-subjects).
- GRU achieved accuracies of 99.18% (within-subject) and 86.44% (across-subjects), demonstrating high performance in movement recognition.
Conclusions:
- Deep learning models, specifically GRU and LSTM, offer superior potential for identifying intralimb movements using EEG signals compared to standard machine learning techniques.
- These findings pave the way for advanced BCI devices in foot rehabilitation, improving physical therapy outcomes.
- The study highlights a promising new direction for enhancing motor rehabilitation technologies through sophisticated EEG signal analysis.
