Retinal Photographs Improve the Diagnosis of Autism Spectrum Disorder

Autism spectrum disorder (ASD) is a neurological and developmental condition that affects behavior, communication, and learning, often requiring lifelong support.Diagnosis by the age of two years can significantly reduce symptom severity and enhance cognitive, language, and social skills.However, current diagnostic methods rely heavily on subjective behavioral observations, rendering them prone to inaccuracies, stressful for caregivers, and time-consuming.To address this issue, this study introduces a novel and objective diagnostic system that utilizes retinal (fundus) photographs in conjunction with machine learning.The fast gradient sign method (FGSM), originally developed as an adversarial perturbation technique, was applied in this study to evaluate the robustness of convolutional neural networks in classifying ASD from retinal images.This robustness test also resulted in modest performance improvements across all tested models, surpassing baseline performances.These findings could aid the development of efficient, accurate, and non-invasive tools for early ASD detection and intervention, thereby significantly benefiting individuals with ASD and their families.Future studies should investigate additional adversarial methods and incorporate larger and more diverse datasets.