Developing a System to Detect Autism Spectrum Disorder (ASD) in Children from Facial and Behavioral Features Using Multimodal Approach

Autism Spectrum Disorder (ASD) is a brain disorder for which symptoms such as repetitive activities or movements start appearing in early childhood. Early detection of ASD empowers parents to take appropriate actions and proper care of the affected children. Earlier research works have mostly focused on detecting ASD from either facial images or behavioral data. However, combining both types of features would ensure better prediction of ASD. In this research work, we have proposed a system to detect ASD in children by evaluating both facial image and behavioral features. We have developed an integrated dataset that contains both images and behavioral data by combining two existing individual image and behavior datasets. We have designed a multimodal model architecture, trained different CNN models on the new integrated dataset, and evaluated the performance of each model. MobileNetV3Large model has outperformed all the other models with an accuracy of 97.36%. Finally, a simple webapp has been developed that takes an image and other behavioral data as inputs and predicts if the child is a probable ASD patient.