Autism Detection in Children by Features Extraction and Classification Using a Deep Learning Model

Autism Spectrum Disorder (ASD) is a very complex nervous growth syndrome that can manifest in different forms. Timely diagnosis and suitable medical intervention can significantly enhance the daily functioning of children with ASD and their families. This study investigates the potential of utilizing static facial characteristics extracted from photos of children suffering from ASD in comparison to normally developing children as a biomarker to distinguish between them. To achieve this, we utilized a publicly accessible dataset containing facial images of both ASD and non-ASD controls, which were classified into autistic and non-autistic categories. Our study used binary image classification to differentiate between ASD and non-ASD pictures. The ResNet50 is used as a pre-trained model for feature extraction, and the last few layers are modified to fit the binary classification task. The performance of the model has produced exceptional results on the testing set.