Filter bank sinc-convolutional network with channel self-attention for high performance motor imagery decoding

Abstract

Objective. Motor Imagery Brain-Computer Interface (MI-BCI) is an active Brain-Computer Interface (BCI) paradigm focusing on the identification of motor intention, which is one of the most important non-invasive BCI paradigms. In MI-BCI studies, deep learning-based methods (especially lightweight networks) have attracted more attention in recent years, but the decoding performance still needs further improving. Approach. To solve this problem, we designed a filter bank structure with sinc-convolutional layers for spatio-temporal feature extraction of MI-EEG in four motor rhythms. The Channel Self-Attention method was introduced for feature selection based on both global and local information, so as to build a model called Filter Bank Sinc-convolutional Network with Channel Self-Attention (FB-Sinc-SCANet) for high performance MI-decoding. Also, we proposed a data augmentation method based on Multivariate Empirical Mode Decomposition (MEMD) to improve the generalization capability of the model. Main results. We performed an intra-subject evaluation experiment on the unseen data of three open MI datasets. The proposed method achieved mean accuracy of 78.20% (4-class scenario) on BCI Competition IV IIa, 87.34% (2-class scenario) on BCI Competition IV IIb, and 72.03% (2-class scenario) on OpenBMI dataset, which are significantly higher than those of compared deep learning-based methods by at least 3.05% (p=0.0469), 3.18% (p=0.0371), and 2.27% (p=0.0024) respectively. Significance. This work provides a new option for deep learning-based MI decoding, which can be employed for building BCI systems for motor rehabilitation.

Explainable cross-task adaptive transfer learning for motor imagery EEG classification

“…The results of FBCSP [21], TCA [65], JDA [65], CORAL [66], Deep ConvNet [23], EEGNet [24], FBCNet [25] obtained under this condition were also reported in [35]. FBMSNet [26], IFNet [27] and FB-Sinc-CSANet [67] all used the entire data in session 2 as evaluation data, the results of the FBCSP-SVM [21], Shallow ConvNet [23], Deep ConvNet [23], EEGNet [24] and FBCNet [25] obtained under this condition were also reported in [26]. MIN2Net [68] used the online data in session 1 and session 2 as the evaluation data, the results of the FBCSP-SVM [21], Deep ConvNet [23], EEGNet-8,2 [24] and Spectral-Spatial CNN [64] obtained under this condition, were also reported in [68].…”

EEGminer: discovering interpretable features of brain activity with learnable filters

“…There are very limited constraints on the model however, such as fully trainable temporal filter functions. Convolutional filters have been constrained by parameterized sinc functions in the audio domain [9], an approach gaining some popularity for EEG decoding [10][11][12][13]. This constrains the model but could reduce trainability, especially in the low signalto-noise ratio regime of the EEG domain.…”

Attention-ProNet: A Prototype Network with Hybrid Attention Mechanisms Applied to Zero Calibration in Rapid Serial Visual Presentation-Based Brain–Computer Interface

“…Following manual and machine searches, RSVP has evolved as a brain-inspired approach for target detection. This represents a novel paradigm in BCI that emerged after steady state visual evoked potential (SSVEP) and motor imagery [4,5].…”