The attention mechanism of the Transformer has the advantage of extracting feature correlation in the long-sequence data and visualizing the model. As time-series data, the spatial and temporal dependencies of the EEG signals between the time points and the different channels contain important information for accurate classification. So far, Transformer-based approaches have not been widely explored in motor-imagery EEG classification and visualization, especially lacking general models based on cross-individual validation. Taking advantage of the Transformer model and the spatial-temporal characteristics of the EEG signals, we designed Transformer-based models for classifications of motor imagery EEG based on the PhysioNet dataset. With 3s EEG data, our models obtained the best classification accuracy of 83.31%, 74.44%, and 64.22% on two-, three-, and four-class motor-imagery tasks in cross-individual validation, which outperformed other state-of-the-art models by 0.88%, 2.11%, and 1.06%. The inclusion of the positional embedding modules in the Transformer could improve the EEG classification performance. Furthermore, the visualization results of attention weights provided insights into the working mechanism of the Transformer-based networks during motor imagery tasks. The topography of the attention weights revealed a pattern of event-related desynchronization (ERD) which was consistent with the results from the spectral analysis of Mu and beta rhythm over the sensorimotor areas. Together, our deep learning methods not only provide novel and powerful tools for classifying and understanding EEG data but also have broad applications for brain-computer interface (BCI) systems.

No abstract

No abstract

Objective: For high-level peripheral nerve injured (PNI) patients with severe sensory dysfunction of upper extremities, identifying the multi-site tactile stimulation is of great importance to provide neurorehabilitation with sensory feedback. In this pilot study, we showed the feasibility of identifying multi-site and multi-intensity tactile stimulation in terms of electroencephalography (EEG). Approach: Three high-level PNI patients and eight non-PNI participants were recruited in this study. Four different sites over the upper arm, forearm, thumb finger and little finger were randomly stimulated at two intensities (both sensory-level) based on the transcutaneous electrical nerve stimulation (TENS). Meanwhile, 64-channel EEG signals were recorded during the passive tactile sense stimulation on each side. Main results: The spatial-spectral distribution of brain oscillations underlying multi-site sensory stimulation showed dominant power attenuation over the somatosensory and prefrontal cortices in both alpha-band (8–12 Hz) and beta-band (13–30 Hz). But there was no significant difference among different stimulation sites in terms of the averaged power spectral density over the region of interest (ROI). By further identifying different stimulation sites using temporal-spectral features, we found the classification accuracies were all above 89% for the affected arm of PNI patients, comparable to that from their intact side and that from the non-PNI group. When the stimulation site-intensity combinations were treated as eight separate classes, the classification accuracies were ranging from 88.89% to 99.30% for the affected side of PNI subjects, similar to that from their non-affected side and that from the non-PNI group. Other performance metrics, including Specificity, Precision, and F1-Score, also showed a sound identification performance for both PNI patients and non-PNI subjects. Significance: These results suggest that reliable brain oscillations could be evoked and identified well, even though induced tactile sense could not be discerned by the PNI patients. This study have implication for facilitating bidirectional neurorehabilitation systems with sensory feedback.