EEG Feature Extraction And Classification Of Autism Spectrum Disorder Children

Autism spectrum disorder is a heterogeneous neurological developmental disorder that involves social,emotional,cognitive,and behavioral abnormalities.With the increasing number of patients,autism populations have attracted more and more attention at home and abroad.However,the pathogenesis is still not clear.Clinical diagnosis mainly depends on behavior observation and assessment scales,which are subjective and inaccurate.There is an(EEG),as a non-invasive neuroimaging tool,is mainly used to measure the neurophysiological changes associated with synaptic activity in the cerebral cortex and has been widely used in the study of neurological disorders.This paper mainly foucus on the feature extraction of EEG and classification of children with autism based on the analyses of EEG,including power spectral density,information entropy and connectivity.Through the recognition of autistic children,the effectiveness of the extracted EEG features is validated.First of all,51 children with autism(3-7 years old)and 50 age-matched normal children were recruited.A 5-minute resting state EEG was obtained from each subject with eyes opened in a quite environment.After screening,the two groups consisted of 45 children with autism and 44 normal children,respectively.Then the EEG signal is preprocessed to remove artifacts such as power frequency,EMG,myoelectricity and others.Three methods were used to extract features from EEG: power spectral density,information entropy,and connectivity.Two features were derived from the power spectral density: the relative power and the fast-slow wave ratio.The information entropy mainly includes approximate entropy,sample entropy,permutation entropy,wavelet entropy and multi-scale entropy methods,while the connectivity mainly analyzes the coherence and phase synchronization of EEG signals.For the features extracted by each method,a support vector machine(SVM)was used as classifier,and the classification accuracy of each feature is compared.Then feature selection is performed through a permutation test method to find the best feature subset,with which the SVM classifier were trained to establish an effective classification model.The results showed that the power spectrum of autistic children were different from controls siginificantly.The relative power of the delta band in the autism group is significantly increased and the alpha frequency is significantly reduced.The classification accuracy in the whole brain is 70.33% and 73.00%,respectively.In addition,the approximate entropy,sample entropy and wavelet entropy of the autism group is significantly lower than that of the normal control group;wavelet entropy is the most effective method for measuring the complexity of autism EEG signals,with a classification accuracy of 70.80%.The classification accuracy of approximate entropy,sample entropy,and permutation entropy were 58.40%,65.20%,and 68.50%,respectively.The connectivity method showed that the coherence and the phase synchronization index of the autism group in the delta band significantly increased;the classification accuracy of the coherence and phase synchronization index were 72.80% and 71.90%,respectively.After integrating all the features from the mothods and conducting feature selection process,13 features were selected as the most discriminative subset,and the final classification accuracy reached 91.01%.The current study explored the potential features of EEG that could identify children with autism and established a classification model based the features we found.From the perspective of clinical application,we provided another objective and effective method for the clinical diagnosis of autism.We hope to contribute to the early detection of children with autism in the future.