ASDvit: Enhancing autism spectrum disorder classification using vision transformer models based on static features of facial images

Abstract

This study embarks on an exploratory journey into autism spectrum disorder (ASD), a multifaceted neurological developmental disorder with a spectrum of manifestations. Recognizing the transformative impact of early diagnosis and tailored medical interventions on the lives of children diagnosed with ASD and their families, The intersection of early diagnosis and tailored medical intervention can substantially enhance the quality of life for children diagnosed with ASD and their families. This study embarks on an innovative approach to augmenting the diagnostic process, specifically through the analysis of static features extracted from facial photographs of autistic children. By employing Vision Transformers (ViT) enhanced with Squeeze-and-Excitation (SE) blocks, our research delves into the potential of facial features as a biomarker for distinguishing autistic children from their typically developing counterparts. The fusion of ViT with SE mechanisms aims to amplify the model's sensitivity toward the subtle yet diagnostically crucial facial cues associated with ASD. Through comprehensive experimentation on a curated dataset, categorized into “autistic” and “non-autistic” groups, our approach demonstrates remarkable proficiency in identifying ASD, thereby opening new avenues for employing facial image analysis as a scalable biomarker in ASD diagnosis.