A Computational Method for Early Diagnosis of Autism Spectrum Disorder in Children Using Deep Learning Approach

Autism Spectrum Disorder (ASD) is a complicated neurological disease that causes difficulty with social interaction and communication.Proper diagnosis at an early stage is essential for successful intervention.This paper introduces a novel deep learning model called the Hybrid Vision Transformer model for detecting Autism Spectrum Disorder "HyVit-ASD" designed to classify ASD based on facial feature images.The HyVit-ASD model combines the strengths of convolutional neural networks (CNNs) and vision transformers (ViTs) to form a strong and efficient classification framework.The HyVit-ASD model has three main components: a convolutional block (ConvBlock), a transformer block with re-attention (TransformerBlock), and a classification layer.The ConvBlock uses a pre-trained ResNet for initial feature extraction, reducing the feature map's channels and resizing them for further processing.The TransformerBlock integrates a pre-trained ViT with a custom ReAttention mechanism, enhancing the model's focus on critical facial regions through attention scores and residual connections.The final stage involves global average pooling and a linear layer for classification likelihood.Our proposed model achieved 91% accuracy, 94% sensitivity, 88% specificity, and a 95.67% AUC.The combination of CNN and ViT architectures and the innovative ReAttention mechanism make HyVit-ASD a pioneering approach in deep learning for ASD classification, contributing significantly to early diagnosis and intervention strategies.