Autism Spectrum Disorder (ASD) is a neurodiverse disorder that impacts children’s development, presenting a range of symptoms that persist into adulthood. Recognizing ASD in children is challenging due to the diversity of associated disorders. We propose a hybrid method for detecting autism in children using MRI screening. Our model leverages hybrid deep-learning techniques to address these challenges. In particular, a blend of Convolutional Neural Networks (CNN) and Graph Attention Networks (GAT) is used for model interpretation, allowing us to identify critical brain regions and connections that contribute to the classification. Experiments conducted using data accessible to the public show that the proposed framework achieves an 88% classification accuracy between ASD and Non-ASD cases, outperforming state-of-the-art methods. This indicates the proposed Hybrid CNN-GAT model’s strong classification ability and generalizability, offering a powerful tool for identifying brain disorders.