Autism Spectrum Disorder Classification on Facial Images by using Deep Learning Models

Abstract

Autism Spectrum Disorder (ASD) is a developmental disorder identified by difficulties with behaviour, speech, and social interaction. Effective intervention requires early diagnosis, but present diagnostic techniques frequently depend on expert knowledge and personal evaluations, which may not be available in all areas.In the field of medical imaging and diagnosis, deep learning techniques have recently acquired popularity, offering an effective means for automated and objective examination.In order to automatically detect ASD via facial image analysis, this study investigates the application of deep learning techniques, more especially Convolutional Neural Networks (CNNs). VGG16, VGG19 and AlexNet are three popular CNN architectures that we use to separate images into autistic and non-autistic groups.The models were trained, verified, and tested to identify characteristic facial traits linked to ASD using a publicly available dataset of 2936 facial imagery. These models' comparative performance analysis shows that VGG16 and VGG19 identify mild facial defects associated with ASD with superior accuracy and specificity. Despite having a quicker computing speed, AlexNet performed less well overall. This study shows how deep learning techniques can help with early detection of ASD and provide a scalable solution to address inequalities in access to clinical expertise.