eduzhai > Applied Sciences > Engineering >

GCNs-Net A Graph Convolutional Neural Network Approach for Decoding Time-resolved EEG Motor Imagery Signals

  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Towards developing effective and efficient brain-computer interface (BCI)systems, precise decoding of brain activity measured by electroencephalogram(EEG), is highly demanded. Traditional works classify EEG signals withoutconsidering the topological relationship among electrodes. However,neuroscience research has increasingly emphasized network patterns of braindynamics. Thus, the Euclidean structure of electrodes might not adequatelyreflect the interaction between signals. To fill the gap, a novel deep learningframework based on the graph convolutional neural networks (GCNs) was presentedto enhance the decoding performance of raw EEG signals during different typesof motor imagery (MI) tasks while cooperating with the functional topologicalrelationship of electrodes. Based on the absolute Pearson s matrix of overallsignals, the graph Laplacian of EEG electrodes was built up. The GCNs-Netconstructed by graph convolutional layers learns the generalized features. Thefollowed pooling layers reduce dimensionality, and the fully-connected softmaxlayer derives the final prediction. The introduced approach has been shown toconverge for both personalized and group-wise predictions. It has achieved thehighest averaged accuracy, 93.056 and 88.57 (PhysioNet Dataset), 96.24 and80.89 (High Gamma Dataset), at the subject and group level, respectively,compared with existing studies, which suggests adaptability and robustness toindividual variability. Moreover, the performance was stably reproducible amongrepetitive experiments for cross-validation. To conclude, the GCNs-Net filtersEEG signals based on the functional topological relationship, which manages todecode relevant features for brain motor imagery.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...