| Autism spectrum disorder(ASD)is a broad neurodevelopmental disorder characterized by social communication disorders,verbal and non-verbal communication deficits,narrow interests,and repetitive and rigid behaviors.ASD individuals have social maladjustments or lifelong disorders,and they cannot take care of themselves,which places economic and mental burden on society and their families.Crrently,the cause of ASD is not clear,and there are no drugs that can cure it.Clinical control studies have shown that early identification can dramatically improve the language ability,cognitive ability and behavioral habits of children with ASD.Therefore,the early identification of ASD is of great significance,and more and more researchers conduct research on early identification techniques for ASD.Infants and young children's social skills,game skills,language and cognitive abilities showed great differentiation.Combining with the regulation of growth and development of infants and young children,researchers have developed a variety of early identification tools.Most of these tools use standard questionnaires completed by parents or other caregivers,and collect the daily behavior habits and cognitive ability of children in the strict clinical environment.However,there are some problems in the application of these tools,which are mainly reflected in the following aspects:(1)Professional assessors are required to participate in the process of assessment,while experienced assessors are scarce;(2)The evaluation process is time-consuming and inefficient;(3)The evaluation process is affected by various subjective and objective factors,and the objectivity of the evaluation results needs to be improved.Following the development of information technologies,such as mobile Internet,smart sensors,and cloud computing,artificial intelligence technology are increasingly used in the fields of medicine and education.Currently,a lot of data for ASD diagnosis are being generated.Historical data should be used during diagnosis and treatment as a basis of judgment for identifying children with ASD,which is expected to be an accurate,rapid and intelligent identification method.In addition,a large number of studies have shown that children with ASD have empathy deficits,and multi-modal data under empathy conditions are likely to provide more information about the characteristics of ASD,which is conducive to improving the accuracy of identifying ASD.Therefore,this paper proposes a identification method for ASD that integrates multi-modal data under the condition of empathy,which mainly solves the following questions:(1)which stage of empathy is the main defect of ASD in theempathy process?What data can characterize these defects?(2)In the process of intelligent identification method for ASD,how to extract the effective features of various data and whether there is complementary information among the data that can improve the identification accuracy?(3)Is the traditional random forest classification method applicable in the process of children classification and how to improve it?(4)For the multi-mode multi-source asynchronous data,how to make full use of its complementary information for data fusion and improve the identification accuracy of ASD?Facing the empathy defect of ASD,this paper analysis the characteristics of multimodal data under the condition of empathy,and we hypothesize that behavioral data and cognitive data in the condition of empathy can be used for early identification of ASD.Then,using this hypothesis we propose an intelligent multimodal framework for identifying children with ASD.Finally,the hypothesis is verified and the validity of our proposed method is verified.Our research content includes four aspects,(1)Study on data of physiological,behavioral and cognitive of children with ASD in the condition of empathy;(2)Identification of children with ASD based on behavioral or cognitive data;(3)A hybrid multimodal data fusion framework based on weighted random forest;(4)Consistency validation of the results of our proposed identification method and the traditional method.Our contributions are given in the following,(1)Data on physiological,behavioral and cognitive of children with ASD and typical developing children were analyzed in this paper under the condition of empathy.We found that the empathy defect of children with ASD is mainly reflected in their cognitive empathy and facial expression imitation.The empathy process of children with ASD has a normal bottom-up emotional sharing process but an abnormal top-down cognitive adjustment process.Therefore,we propose that behavioral data and cognitive data in the condition of empathy can be used for early identification of ASD.(2)In this paper,the data of eye fixation,facial expression and social cognitive were used to identify ASD from TD children,and their different discriminative powers for detection ASD were explored.The results showed that eye fixation data,facial expression data and social cognitive data were useful indicators for identifying children with ASD,and they have important complementary characteristics,and combining the complementary information may improve the classification accuracy.(3)We proposed an improved RF algorithm based on weighted decision trees.We evaluated the classification abilities of the decision trees according to the mutual information and assigned weights for each decision tree.Weighted voting was conducted according to the prediction weight of each decision tree to improve the accuracy of classification.(4)This paper proposed a hybrid fusion method based on the data source and time synchronization,and the hybrid fusion process was divided into two levels.The data with synchronization and different source were fused in the first level for feature fusion,and then the results were fused in the second level for decision fusion.The hybrid fusion method makes full use of the complementary information of different data,and ensures the flexibility and objectivity of the decision-making. |
PDF Full Text Request