Detecting Autism in Children Through Facial Image Analysis Using Transfer Learning

Abstract

Autism Spectrum Disorder (ASD) is a neurological condition that impacts individuals' ability to interact and communicate with others throughout their lives. Early diagnosis of autism can significantly enhance the effectiveness of treatment. Often described as a “behavioral disorder,” ASD typically presents symptoms within the first two years of a child's life through observable behaviors. However, recent studies have indicated that ASD may also influence certain facial features. Driven by the growing adoption of machine learning in medical diagnostics, we have experimented with several deep learning models that leverage transfer learning to detect autism in children based on facial images. In addition, we introduce a novel latency feature that enhances the model's decision-making process. Our novel method, which integrates an Xception model as the foundation for the AdaBoost algorithm, achieves an accuracy of 91 %, surpassing the performance of previous standalone models.