A framework for motor imagery with LSTM neural network. (May 2022)

• Record Type: Journal Article
• Authors: Xu, Fangzhou; Xu, Xiaoyan; Sun, Yanan; Li, Jincheng; Dong, Gege; Wang, Yuandong; Li, Han; Wang, Lei; Zhang, Yingchun; Pang, Shaopeng; Yin, Sen
• Is Part Of: Computer methods and programs in biomedicine. Volume 218(2022)
• Publication Date: 2022-05
• Subjects: Brain-computer interface (BCI) -- Motor imagery (MI) -- Long short-term memory (LSTM); Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2022.106692
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: We design an effective fusion model to solve the recognition based on motor imagery tasks. This model makes full use of the spatial-temporal information obtained from different channels to significantly improve the recognition performance. Specifically, the long short term memory(LSTM) recurrent neural network captures the temporal dependencies of the raw electroencephalogram(EEG)/electrocorticogram(ECoG) signals while still retaining some of the spatial information. Then, the fully connected layer can fuse and transform the deep representations from the LSTM layer. The input is the raw EEG/ECoG data vectors that eliminate both the data preprocessing and feature engineering operations. This input reduces the time-consumption of building the model, and it also reduces the dependence on professional expertise. We extensively evaluate the performance of our framework on two public datasets. The high levels of accuracy on the EEG (99%) and ECoG (100%) datasets demonstrate the consistent applicability of our proposed model. Meanwhile, the high accuracy shows that our research method is superior to the state of the art models. Abstract: Background and Objective: How to learn robust representations from brain activities and to improve algorithm performance are the most significant issues for brain-computer interface systems. Methods: This study introduces a long short-term memory recurrent neural network to decode the multichannel electroencephalogram or electrocorticogram for implementing an effective motor imagery-based brain-computer interface system. The unique information processing mechanism of the long short-term memory network characterizes spatio-temporal dynamics in time sequences. This study evaluates the proposed method using publically available electroencephalogram/electrocorticogram datasets. Results: The decoded features coupled with a gradient boosting classifier could obtain high recognition accuracies of 99% for electroencephalogram and 100% for electrocorticogram, respectively. Conclusions: The results demonstrated that the proposed model can estimate robust spatial-temporal features and obtain significant performance improvement for motor imagery-based brain-computer interface systems. Further, the proposed method is of low computational complexity.