A Comprehensive Survey on Deep Learning Approaches for Autism Spectrum Disorder Detection Using Multimodal Data

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by impairments in social communication and restricted, repetitive behaviors. Recent advances in deep learning (DL) have significantly improved our ability to detect and understand ASD by learning discriminative patterns from diverse biomedical and behavioral data modalities such as MRI, EEG, and eye-tracking. This survey provides a comprehensive overview of DL-based ASD detection methods, focusing on multimodal data integration, architectural innovations, and evaluation methodologies. Through a detailed comparative analysis of some representative studies, we identify trends, strengths, and persistent limitations, highlighting the shift from conventional CNNs toward explainable, transformer-based, and federated architectures.