EEG-based anxiety emotion classification using an optimized convolutional neural network and transformer

Anxiety disorder is a prevalent mental health problem across all age groups. Although deep learning has shown remarkable success in numerous medical domains, the classification of anxiety emotion using deep learning remains challenging and underexplored. In this paper, we propose a hybrid model based on an optimized convolutional neural network and Transformer, named MSDSTT. The optimized convolutional neural network extracts local multi-scale spatio-temporal features from EEG signals. The temporal convolutional network(TCN)-Transformer parallel module is used further to extract the deeply local and global spatio-temporal features of EEG signals. Ablation studies have validated the capabilities of the proposed optimized modules to enhance the model's feature representation capabilities. The experimental results show that the proposed MSDSTT model achieves an accuracy of 87.8%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\%$$\end{document} and 86.9%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\%$$\end{document} on the binary classification and quad classification tasks of the DASPS dataset, respectively, indicating its superior performance in anxiety emotion classification compared to other methods.