Application of deep learning on classification of disorders of consciousness based on EEG frequency-domain features Classification Method of Disorder of Consciousness

The clinical prognosis diagnosis Of Disorders Of Consciousness (DOC) mainly relies on doctors to use behavioral scales and medical images for visual detection and manual labeling, which is inefficient. To overcome the limitations of traditional diagnostic methods, this paper proposes a deep learning processing and classification method based on EEG data frequency domain features. Firstly, the pre-processed EEG data are traced according to the Desikan-Killiany atlas rules, and then the EEG power spectrum and phase-locked value features in the traceability space are extracted. And input to a two-dimensional convolutional network for classification. The results show that compared with the direct processing of raw data, the data classification after frequency domain feature extraction has higher classification accuracy. It is found that the sliding window parameters when extracting the power spectrum and the frequency parameters when extracting the phase-locked value have a significant impact on the classification results, which provides a research basis for the prognosis diagnosis of DOC based on EEG.