Identification of Autism Spectrum Disorder in Children from Facial Features Using Deep Learning

Autism spectrum disorder (ASD) is a complex neurological developmental condition that presents with a range of symptoms. Early diagnosis along with proper medical care can significantly enhance the daily quality of life for parents and children with ASD. The purpose of this study is to determine whether it is possible to distinguish autistic children from usually developing kids using biomarkers derived from face traits retrieved from their images. The study employs Convolutional Neural Network (CNN) models, specifically VGG19, Densenet121, and InceptionV3, for the extraction of features. Additionally, a Deep Neural Network (DNN) model is utilized as a binary classifier to accurately discern autism. The investigation utilizes a publicly accessible dataset comprising facial images of children diagnosed with autism, alongside control subjects classified into both autistic and non-autistic categories. With a 96.66% accuracy, 96.25% Precision, 94.75% Recall and 95.50% F1 Score, Densenet121 performed better than the other models that were examined. When applied to groups with and without autism, InceptionV3 consistently produced prediction scores of 95.33%