Design a Novel BCI for Neurorehabilitation Using Concurrent LFP and EEG Features: A Case Study

Design a Novel BCI for Neurorehabilitation Using Concurrent LFP and EEG Features: A Case Study IEEE Transactions on Biomedical Engineering (TBME)
Author(s): Zhao Feng, Yi Sun, Linze Qian, Yu Qi, Yueming Wang, Cuntai Guan, Yu Sun

Brain-computer interfaces (BCIs) have been widely explored in patient-oriented neurorehabilitation training for those with severe motor disabilities, during which BCIs translate patients’ motor intention into instructions for controlling exoskeleton. Of note, unimodal BCI-based neurorehabilitation was exclusively developed via using either invasive or non-invasive brain signals to decode motor intention, a novel BCI that inherits advantages of invasive intracortical and non-invasive scalp brain signals is of great interest for practical neurorehabilitation application. To this end, we developed and validated, for the first time, the performance of a novel hybrid BCI that incorporate both local field potential (LFP) and electroencephalogram (EEG) (named LFP-EEG-BCI) for motor intention decoding during neurorehabilitation.

Specifically, we collected concurrent LFP and EEG signals from a completely paraplegic patient (ASIA impairment scale A) undergoing upper-limb motor imagery (MI) neurorehabilitation training. A common spatial filter approach was utilized to extract multi-frequency task-related power features from both LFP and EEG signals, which were used to decode patient’s motor intention in an unimodal fashion. Then a decision fusion strategy was introduced to obtain the final decoding result. We found that the proposed LFP-EEG-BCI outperformed unimodal BCI (EEG-BCI and LFP-BCI) in motor intention decoding with the highest accuracy of 82% (for four-class MI) during neurorehabilitation. By investigating spatial-spectral patterns of LFP and EEG features, we proved that the LFP-EEG-BCI is capable of extracting motor intention features of multiple scopes. Moreover, a transfer learning (TL) method was developed to further reduce the calibration efforts of training decoding model and improve the user-friendliness of the LFP-EEG-BCI. This work can lead to new insights of developing BCI-based neurorehabilitation systems utilizing concurrent intracortical and scalp brain signals, which are beneficial for clinical applications.   

Access the Full Paper on IEEE Xplore®

Sign-in or become an IEEE member to discover the full contents of the paper.