Detection of alcoholic EEG signals based on whole brain connectivity and convolution neural networks

• Published in: Biomedical signal processing and control Vol. 79; p. 104242
• Main Authors: Shen, Mingkan, Wen, Peng, Song, Bo, Li, Yan
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.01.2023
• Subjects: 3D-CNN; Alcoholism; Continuous wavelet transform; EEG; Mutual information; Whole brain connectivity; Whole brain connectivity; 3D-CNN; EEG; Alcoholism; Mutual information; Continuous wavelet transform
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2022.104242

•Firstly, a deep learning enabled whole brain connectivity analysis method was applied to detect alcoholic EEG signal.•Design a framework of a 3D-CNN, and apply the image classification method to detect EEG signal and get an accuracy of 96.25 ± 3.11 % using leaving-one out training method for all the testing subjects.•Brain rhythms factor was taken into consideration in detecting the alcoholic EEG, and the gamma band (30–40 Hz) was found to be the most significant rhythm.•After the evaluation of all CMI connectivity values, the adjacent connectivities between the left parietal part, the left frontal part, the right temporal part, the right frontal part and the right parietal part were found to be the fuzzy locations in determining alcoholism. Alcoholism is a common complex brain disorder caused by excessive drinking of alcohol and severely affected the basic function of the brain. This paper investigates classification of the alcoholic electroencephalogram (EEG) signals through whole brain connectivity analysis and deep learning methods. The whole brain connectivity analysis is proposed and implemented using mutual information algorithm. Continuous Wavelet transform was applied to extract time–frequency domain information in each selected frequency bands from EEG signal. The 2D and 3D convolutional neural networks (CNN) were used to classify the alcoholic subjects and health control subjects. UCI Alcoholic EEG dataset is employed to evaluate the proposed method, a 96.25 ± 3.11 % accuracy, 0.9806 ± 0.0163 F1-score result in 3D-CNN model was obtained via leaving-one out training method of all the testing subjects.