Decoding EEG for Spinal Cord Injury Rehabilitation: Integrating Source Localization with Residual CNN

In this paper, we present a novel approach to classify motor tasks using electroencephalogram (EEG) data from participants with cervical spinal cord injury (SCI). Our method integrates advanced source localization techniques, including Minimum Norm Estimation (MNE), dipole fitting, and beamforming, with a customized Residual Convolutional Neural Network (ResNet-CNN) architecture. By leveraging these techniques, we enhance the spatial resolution and accuracy of EEG signal classification, which is crucial for developing assistive devices and neurorehabilitation strategies for individuals with SCI. We focus on six specific hand movements: pronation, supination, palmar grasp, lateral grasp, hand open, and a rest condition. The proposed approach achieved classification accuracies of 81.35 % with Dipole fitting, 83.34 % with MNE, and 83.36 % with Beamforming, outperforming state-of-the-art methods in the sensor domain which reported accuracies ranging from <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">${8 0. 1 1 \%}$</tex> to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">${8 0. 7 5 \%}$</tex>. Our ResNet-CNN model with source localization demonstrated F1-scores between <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">${7 5. 1 6 \%}$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">${8 4. 7 2 \%}$</tex>, highlighting the importance of accurate spatial mapping in EEG analysis. The findings of this study underscore the potential of integrating source localization with deep learning to improve brain-computer interface (BCI) applications for individuals with SCI. This approach offers a promising direction for enhancing rehabilitation and daily assistance technologies.