Classification of Cognitive EEGs via Attention-Based Multi-Scale Dilated Convolution Network

Cognitive ability refers to an individual's psychological capacity and level demonstrated in activities such as thinking, memory, and learning. The transcutaneous electrical nerve stimulation (TENS) is widely regarded as a therapeutic method for improving cognitive ability. Therefore, this paper proposes a framework based on EEG signals to evaluate the effectiveness of TENS therapy in cognitive enhancement. Specifically, this paper uses Continuous Wavelet Transform (CWT) to convert EEG signals into time-frequency diagrams, while Phase Space Reconstruction (PSR) is employed to transform EEGs into phase space matrices. Subsequently, both the time-frequency diagrams and PSR matrices are simultaneously input into the Attention-based Multi-Scale Dilated Convolution Network (A-MSDCNet), leveraging their time-frequency characteristics and nonlinear properties to classify EEGs before and after the application of TENS. The results based on the dataset established by our research institute indicate that the proposed method is superior to existing methods. It achieves optimal performance in various indicators such as average classification accuracy (ACC), standard deviation (STD), Cohen's kappa coefficient (kappa), F1 score (F1), and area under the curve (AUC).