Training state and performance evaluation of working memory based on task-related EEG. (May 2019)

• Record Type: Journal Article
• Authors: Wang, Hong; Hua, Chengcheng; Wang, Qiaoxiu; Fu, Qiang; Fetlework, Tenssay
• Is Part Of: Biomedical signal processing and control. Volume 51(2019)
• Page Start: 296
• Page End: 308
• Publication Date: 2019-05
• Subjects: Working memory (WM) -- EEG -- Functional brain network (FBN) -- Phase synchronization -- Stacked auto-encoder (SAE); Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2019.03.002
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: We developed a system based on the EEG to evaluate working memory ability (training state and performance) of human. We applied PLV and PLI to build functional brain network based on EEG, and used graph-theoretic indexes as EEG features to measure working memory ability. Compared with the correlation between the EEG features and working memory performance, the correlation between the EEG features and working memory training was stronger. We used a deep learning algorithm named as stacked auto-encoder to predict of the performance in the WM task based on the EEG features, and the MSE was 144.24. Abstract: The working memory (WM) refers to the information maintaining and manipulation during a short period, and it is corresponding to human ability in many tasks. The correlation between EEG features and the training state of the subjects or their performance in WM tasks had been investigated by many researches. However, there was no research done on the comparison between the training and performance to investigate which one is more correlated with the EEG features and adequately developed practical application of this correlation by now. This paper used phase synchronization methods to build functional brain networks (FBN) of the subjects based on their task-related EEG. Based on this, we investigated the correlation of the global and local features of the FBNs and applied Quadratic Discriminant, Cosine KNN and stacked auto-encoder (SAE) to evaluate the performance and the training state. The accuracy of training state detection was 98.7%, while the accuracy of performance prediction (predict if the score>79) was 81.2% and the MSE of the score prediction was 144.24. The results suggested that the training state is more reliance to the FBN features than performance. The method had the potential to be extended to other fields to assess WM ability or proficiency of people.