On the use of power-based connectivity between EEG and sEMG signals for three-weight classification during object manipulation tasks

Brain-machine interfaces (BMIs) have been used for motor rehabilitation of complex movements, such as those based on object manipulation. However, task identification during these movements remains a challenge in the scientific community. Recent research has suggested that corticomuscular connectivity may enhance the BMIs’ performance in task identification. Therefore, this study presents an algorithm that uses power-based connectivity (PBC) as a descriptor to improve the classification of three different weights during object manipulation which was compared with power spectral density (PSD) benchmark algorithm.

Signals from three electroencephalography (EEG) and five surface electromyography (sEMG) electrodes were analyzed using Welch’s estimator to determine the PSD features and then correlated using Spearman’s correlation. The performance was evaluated using four classifiers that are widely applied in brain-computer interfaces (BCIs). Furthermore, different frequency bands and the influence of EEG and sEMG channels on object weight identification were evaluated using accuracy, F-score, and computational cost metrics.

The proposed algorithm significantly outperforms ( \(p<\) $p<$ 0.05) the standard method based on PSD, with a difference in accuracy of 19.15% and F-score of 10.40% and obtaining a computational cost of 6 s less.

These findings demonstrate the promising potential of the PBC method for object weight identification in complex tasks. The implementation of such algorithms can lead to significant improvements in the effectiveness of BMIs for object manipulation, with potential benefits for rehabilitation and other applications.