Early Autism Spectrum Disorder Screening Based on Integrated Neural Networks and Scales

Quantitative checklist for autism in toddlers (QCHAT) is a more general screening tool for early autism spectrum disorder (ASD), but it is mostly manually determined in the assessment, and many complex situations cannot be judged. In recent years, deep neural network (DNN) architectures have been proven to perform well in dealing with complex problems, based on the idea of Bootstrap Aggregation (Bagging), a Multi-Channel Bagging Attention Tabular Network (MBAT-Net) for ASD screen is proposed and attempts to identify ASD patients from QCHAT datasets in this paper. Two datasets were used in this study: QCHAT and QCHAT-10 to train and evaluate the model performance, and finally achieved 97.20% accuracy on QCHAT and 100% on QCHAT-10. The proposed framework is then compared with existing research methods, and the results showed that it effectively improved the early screening accuracy of ASD, and could provide a more reliable automated tool for early ASD diagnosis.