Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by atypical brain development, often reflected in distinct facial features. This research proposes a deep learning-based diagnostic system leveraging Convolutional Neural Networks (CNNs) for early ASD detection through facial analysis. By integrating social media and biomedical facial image data, the system employs transfer learning with three optimized CNN architectures—ResNet18, EfficientNetB1, and EfficientNetB2—implemented within a Django-based web application. The dataset, consisting of 2,940 facial images sourced from Kaggle, is used to train and evaluate these models using key performance metrics such as sensitivity, specificity, and accuracy. The results highlight the effectiveness of CNN models in identifying subtle facial cues associated with ASD, facilitating early diagnosis and intervention. This approach aims to provide an accessible, automated, and reliable tool to assist healthcare professionals and communities in recognizing ASD at an early developmental stage.

Keywords: Autism Spectrum Disorder, Deep Learning, CNN, ResNet18, EfficientNetB1, EfficientNetB2, Transfer Learning, Facial Recognition, Early Diagnosis, Django.