Convolution Neural Network for Indentifying Autism Spectrum Disorder in Children

Autism spectrum disorder (ASD) is a neurodevelopment condition characterized by irregularities in brain development that affect face morphology. Children with autism have significant differences in the configuration of their facial characteristics compared to ordinarily developing children. This study proposes a novel deep-learning approach for detecting autism through facial cues, employing a convolutional neural network with a multi-head attention mechanism (CNNMHA), thereby facilitating the diagnostic process for families and physicians. It primarily employs two phases: preprocessing and classification. During this preparation phase, the suggested system executes image resizing, Gaussian filtering for denoising, and normalization to improve image quality and suitability for the given classification task. During the classification phases, the proposed system employs CNNMHA to extract features from the preprocessed picture and categorizes the images as depicting autistic or non-autistic children. The data were acquired from Kaggle, and the initial computational findings indicate that the suggested approach surpassed current methodologies. The technology can assist medical practitioners in corroborating their first screening results to identify children with ASD.