EEG Based Autism Recognition Model: EEGCM-Net Based Attention Mechanism

Autism spectrum Disorder (ASD) is a neurodevelopmental disorder whose diagnosis currently relies primarily on assessment and diagnosis by experienced clinicians. In this study, we used electroencephalogram (EEG) signals and EEGCM-Net model to detect ASD. EEGCM-Net first realized adaptive weight allocation to different EEG channels through channel attention mechanism, and then extracted time-frequency features by one-dimensional convolution. Subsequently, the multi-head attention mechanism is introduced to focus on the spatio-temporal relationship between different time Windows and channels to better capture signals and generate more discriminative features. In this study, two different experimental paradigms were tested on the ABC-CT public data set, and both were targeted at the same subjects. The experimental results show that the average classification accuracy of the proposed method is 91% and 92% under the two paradigms. Sensitivity is 98%, 99.5%; F1-score was 93.9% and 94.7%; The AUC was 97%,96.3%. These indicators indicate that the method can automatically extract and classify the features of EEG signals, so as to effectively detect ASD.