Recognition In Autistic Children Based On Abnormal Face Processing

One of the main manifestations of social interaction disorder in autism spectrum disorder(ASD)is abnormal face processing.Based on eye tracking technology,the paper analyzes the atypical facial scanning mode of ASD from two aspects: manual and automatic segmentation of area of interest(AOI),and carries out feature selection and classification.First of all,we designed an eye movement experiment paradigm and collected eye movement data of 85 autistic children aged 3 to 6 years old and 89 gender-and age-matched typically developing(TD)children.Secondly,in the study of manually dividing AOI,the paper analyzes the native and foreign unfamiliar faces from two aspects: static analysis(statistical analysis and correlation)which is not related to time series,and dynamic analysis(scanning path and Markov chain entropy)which is related to time series.The results show that autistic children have atypical facial scanning patterns,the percentage of fixation time can predict the severity of one aspect of symptoms in autistic children.Then feature selection is performed through independent sample t sample method,with which the four classifiers were trained to establish an effective classification model.The results show that the support vector machine(SVM)classification accuracy is up to 82.22%.Finally,based on the gaze coordinates,the K-means clustering algorithm is used to automatically divide the AOI of native and foreign unfamiliar faces into 64 AOI.The coordinate frequency is used as a feature,and the maximum correlation and minimum redundancy(m RMR)algorithm is used for feature selection and classificatio.AOI with significant difference between ASD children and TD children was calculated.The results show that after feature selection,the SVM classification result is 86.73% at most.Compared with the manual division of interest areas,the classification accuracy and performance are improved.Compared with TD children,ASD children pay more attention to the background area and body area,and pay less attention to the triangle area of the face.From the perspective of clinical application,the paper looks for the eye movement features that can effectively identify ASD children and TD children,and establishes the classification model through feature selection.The results of the study show that different face scanning modes can provide enough information to distinguish ASD children from TD children and provide objective indicators for the auxiliary diagnosis of autism.