Long-range correlation analysis of high frequency prefrontal electroencephalogram oscillations for dynamic emotion recognition. (February 2022)

• Record Type: Journal Article
• Authors: Gao, Zhilin; Cui, Xingran; Wan, Wang; Zheng, Wenming; Gu, Zhongze
• Is Part Of: Biomedical signal processing and control. Volume 72(2022)Part A
• Publication Date: 2022-02
• Subjects: Emotion recognition -- High frequency EEG -- Multi-order detrended fluctuation analysis -- Emotion accumulation; 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.2021.103291
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A novel multi-order detrended fluctuation analysis (MODFA) method was proposed. High frequency EEG in prefrontal region were analyzed to classify six emotions. MODFA achieved the best performance in binary-, ternary-, and six-classification. Cumulative effects of negative emotions along with the stimulus were observed. Abstract: Numerous previous studies have proved the enormous potential of high frequency EEG in emotion recognition, however, the current existing EEG analytic methods are not so effective when dealing with high frequency oscillations. Therefore, a novel refined-detrended fluctuation analysis method multi-order detrended fluctuation analysis (MODFA) was proposed. The best fitting order is selected according to the inflection point in the dependence degree curve of high frequency EEG and multi-order polynomial. MODFA measures the power-law long-range correlation of high frequency nonlinear signals. Prefrontal EEG signals were recorded during six emotion-inducing tasks ( neutral, fear, sad, happy, anger, and disgust ). To confirm the susceptibility and efficiency of MODFA indices, including hurst-exponent MODFA-h1 and intercept MODFA-a1, on emotion recognition, we compared MODFA with original detrended fluctuation analysis, as well as the conventionally used fuzzy entropy (FE) and power spectral density (PSD) on high frequency EEG oscillations (62.50–93.75 Hz). The results showed that MODFA achieved the best performance in binary emotion classification (positive and negative, accuracy = 96.81%), ternary classification (neutral, positive, and negative, accuracy = 76.39%), and six-classification (accuracy = 42.17%). Moreover, along with inducing time, the cumulative effects of the four negative emotions (fear, sad, anger, and disgust) were observed by MODFA-a1, FE, and PSD, which demonstrated that the accumulation of negative emotions are associated with the prefrontal lobe and could be measured via high frequency gamma rhythms. These findings indicated the nonlinear dynamics of high frequency brain activity during emotion induction, and the prefrontal EEG-based emotion recognition might have great application prospect in real-life practice.