Advancing Autism Spectrum Disorder Diagnosis With Multimodal Data: A Survey

Autism Spectrum Disorder (ASD) is a complicated neurodevelopmental condition characterised by a variety of behavioural and communication challenges, marking timely and accurate diagnosis and intervention essential. Despite their widespread occurrence, conventional diagnostic methods like clinical observations and standardised questionnaires are hindered by issues like subjectivity, practitioner variability, and long evaluation durations and existing works using AI for diagnosis falters as they fail to capture range of symptoms, generalizability across cohorts and a lot more challenges. This survey compiles a comprehensive set of ASD datasets, including the limited multimodal collections, and outlines clear modality specific preprocessing steps that ensures data quality. By reviewing prior multimodal fusion surveys and highlighting breakthroughs across individual modalities with the main fusion studies, it establishes a unified view of the diagnostic approach and the fusion techniques used in ASD research. Furthermore, the survey explains the advantages of multimodal fusion, summarises the major limitations reported across existing work, and introduces a simple taxonomy with practical solutions that address these challenges and guide future development.