Early Detection of Autism Spectrum Disorder in Pediatric Populations using Facial Feature Analysis

Autism Spectrum Disorder commonly known as ASD is a severe neuro-developmental impairment which disrupts the ability to interact both verbally and non-verbally and challenges learning. ASD affects the central nervous system and influences the emotional, physical, and social health of children, worsening with age. Traditional diagnostic procedures based on behavioral observations are time-consuming and financially demanding, relying on the experience of specialists. Thus, the detection of ASD is difficult in children. Early detection can mitigate its severity and long-term effects. Advanced machine learning techniques, including MobileNet, ResNet50, AlexNet, and classifiers like SVM and RF, have been used in many case studies for ASD diagnosis but have limitations in achieving high accuracy and robustness with numerous faces. To approach this, Convolutional Neural Network (CNN) is employed to recognize features from children’s facial visuals to enhance precision. This approach recognizes ASD by analyzing the facial landmarks of children to match the characteristics of ASD. This module can extract more useful image features, avoiding the negative effects of useless features. The dataset contains images of non-autistic and autistic children.