Early Autism Diagnosis Using Facial Images: A Deep Ensemble Model for Feature Fusion

Autism spectrum disorder (ASD) is a neurodevelopmental syndrome that is characterized by a wide variety of unique symptoms that can vary widely from person to person. These manifestations can even be very different from one another. Through the early identification of autism spectrum disorder (ASD) and the provision of proper medical treatment, it is possible to significantly enhance the quality of life of children who have ASD as well as the families of such children. Using static facial characteristics from photos of autistic children as a biomarker to differentiate them from usually developing youngsters is the topic of this article. A combination of the EfficientNet B3, B5, and B7 deep learning models was employed to enhance classification accuracy. For both the autistic and non-autistic groups, the adjusted model achieves an AUC-ROC of 96% and an accuracy of 91.67%.