Research On The Difference Of Emotional Cognition Of Children With Autism Spectrum Disorder And Typical Developing Children Based On Physiological Signal

Autism spectrum disorder(ASD,hereafter autism)is a neurodevelopmental disorder,and the analysis of physiological signals of autistic children can be used for objective evaluation of their cognitive development.Because emotional recognition ability is of great importance to individual social development,the emotional recognition defect of children with autism has an extremely negative impact on their social interaction ability,thus hindering their social development.According to related studies,the abnormal emotional cognition of autism can assist early screening assessment.Physiological signals can reflect children's emotional cognitive states.Non-invasive studies of electroencephalogram signals and eye movement data can observe the cognitive activity of the brain from multiple perspectives.The experiment recruited 80 children,40 autistic and 40 typically developing children,and we analyses the difference of testees' electroencephalograms and eye movement data.We propose Convolutional Neural Networks-Long Short-Term Memory(CNN-LSTM)and Multilayer Perceptron(MLP)mixed models to improve the accuracy of autism screening.The specific research content is as follows:The study used a two-factor mixed method experimental design.The type of children in the experiment is an inter-subject variable.The emotion category is an in-subject variable.The electroencephalogram and eye movement indicators of children while they are being induced by videos are dependent variables.Explore the differences in performance and data of the two types of children induced by different emotional videos.This paper analyses testees' electroencephalograms of several encephalic regions,including frontal,temporal and occipital lobes,along with their emotional performance metrics and cognitive level to investigate the difference in autistic children and typically developing children's performance and electroencephalograms while they are being induced by videos designed to produce positive or negative emotions.Based on the emotional performance indicators obtained from brainwave power spectrums and cognitive questions,we analyzed the children's performance under two emotional stimuli.Our results show that autistic children's electroencephalograms generally show very high power in the low frequency range(delta,theta)and high frequency range(beta,gamma)but low power in the alpha band.Moreover,different encephalic regions show different features of the power spectrum.By comparing the testees' emotional performance metrics based on their electroencephalograms' power spectrums,as well as cognitive levels based on answering questions under two kinds of emotional stimulation,we found that autistic children's emotional changes are far less than those of typically developing children.We also analyzed the eye tracking of the two types of children.And we explored the differences in eye movement processing between autistic children and typical developing children.The results showed that the analysis results of the five eye movement indicators,such as the number of fixation points,average fixation time,average saccade distance,fixation time ratio,and social priority score,all showed that the main effect of child type was significant.Children with autism are more interested in video materials under positive emotions.Based on the differential analysis of the above two types of physiological signals,this study selects indicators with significant differences between subjects for in-depth analysis.We propose a mixed model of CNN-LSTM and MLP for feature extraction and data fusion of the children's physiological signals.Then we complete the classification of two types of children.In this model,the electroencephalogram features extracted by CNN-LSTM and the eye movement features extracted by MLP are merged into the feature layer to complete the classification.The experimental result shows that the CNN-LSTM model is used to classify the two groups of children based on electroencephalogram signals,with an accuracy rate of 69.9%.Using MLP to classify two groups of children based on eye movement data,the accuracy rate reached 67.1%.After the fusion of the two,the classification effect has improved,and the accuracy has reached 75.2%.Compared with selecting only a single data classification,the accuracy rate has improved by about 5%.Based on the abnormal emotion recognition of autistic children,this article uses a two-factor mixed experimental design method to analyze,explain and compare the characteristics of the two groups of children.We construct a hybrid model based on CNN-LSTM and MLP to perform data fusion and classification on physiological signals,and obtain a better classification result.This research provides an effective objective basis for the auxiliary diagnosis of autistic children.