A Transformer-Based Approach Combining Deep Learning Network and Spatial-Temporal Information for Raw EEG Classification

Abstract

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.

A Detailed Analysis & Parametric Comparison Of Eeg Processing Models

“…Results of the experiments show that the proposed work provides a trustworthy system with a high average accuracy of 94.4% and a training time of 0.6 s for the MAHNOB-HCI database. The Transformer's feature correlation extraction and display method uses long-sequence data [19]. For precise classification, EEG data across time points and channels must exhibit spatial and temporal correlations.…”

“…Work in [18] developed a therapy for the prevention and treatment of epilepsy by using a solution that was based on an in-depth learning technique that was constructed on a cloud platform. A deep network-based coding technique was developed for the analysis of epileptic EEG data by both [19] and [20]. While the bulk of these studies have concentrated on regular data.…”

EEG Conformer: Convolutional Transformer for EEG Decoding and Visualization

“…Lately, attention-based Transformer models have made waves in natural language and image processing due to the inherent perception of global dependencies [20]. Transformers also emerge in EEG decoding and achieve good performance, by leveraging long-term temporal relationships [21], [22]. However, such models ignore learning local features, which are also necessary for EEG decoding.…”

Improved Differential Evolution with Stacked Auto Encoder for EEG Motor Imagery Classification

“…The solutions obtained by the filter coefficients are dependent on the initial parameters (9,10) . Transformed based approaches have been used to classify the EEG signals with the spatial-temporal characteristics of EEG, with good percentage of accuracy but there is change of improving the accuracy even more (11) .…”