An Adaptive EEG Feature Extraction Method Based on Stacked Denoising Autoencoder for Mental Fatigue Connectivity

Abstract

Mental fatigue is a common psychobiological state elected by prolonged cognitive activities. Although, the performance and the disadvantage of the mental fatigue have been well known, its connectivity among the multiareas of the brain has not been thoroughly studied yet. This is important for the clarification of the mental fatigue mechanism. However, the common method of connectivity analysis based on EEG cannot get rid of the interference from strong noise. In this paper, an adaptive feature extraction model based on stacked denoising autoencoder has been proposed. The signal to noise ratio of the extracted feature has been analyzed. Compared with principal component analysis, the proposed method can significantly improve the signal to noise ratio and suppress the noise interference. The proposed method has been applied on the analysis of mental fatigue connectivity. The causal connectivity among the frontal, motor, parietal, and visual areas under the awake, fatigue, and sleep deprivation conditions has been analyzed, and different patterns of connectivity between conditions have been revealed. The connectivity direction under awake condition and sleep deprivation condition is opposite. Moreover, there is a complex and bidirectional connectivity relationship, from the anterior areas to the posterior areas and from the posterior areas to the anterior areas, under fatigue condition. These results imply that there are different brain patterns on the three conditions. This study provides an effective method for EEG analysis. It may be favorable to disclose the underlying mechanism of mental fatigue by connectivity analysis.