Early Autism Screening in Children Using Facial Recognition

Facial data of children aged 2 to 10 years was collected from three institutions in a large city; an autism rehabilitation center, a rehabilitation hospital, and an inclusive kindergarten. The dataset comprised facial data of 65 children diagnosed with autism, and 47 children with typical developmental. We employed the VGG-16 model to develop a facial feature recognition-based early screening system, which involves feature extraction and image processing of eyes, eyebrows, noses, and mouths. The data was processed and categorized using a Convolutional Neural Network (CNN) model, and the accuracy of this algorithm was validated. Cross-testing with the public database Kaggle and our dataset demonstrated an accuracy rate of up to 94% for the current training set. This indicates that the model trained by our system is proficient in classifying children's facial data and maintains high precision on our database.