International Journal of Science, Engineering and TechnologyAuthors: C. Sai Kalyani Deepthi, N. Jyoshnavi, N. Varshini, K. Harshitha, M. Ankitha
Abstract: Autism Spectrum Disorder (ASD) is a complex condition that affects social communication linked to repetitive behaviours. Early detection of ASD is crucial because timely treatment can significantly improve developmental outcomes. Traditional screening methods are often slow, subjective, and depends on single data sources like behavioural questionnaires or facial analysis. To solve these problems, a novel framework Cross-Modal Dynamic Hypergraph Attention Network (CDHAN) has been proposed for ASD identification. This framework combines facial image data with behavioural screening information to capture complex interactions between different data types. This model employs dynamic hypergraph structure to capture the higher-order interactions and uses attention mechanisms to focus on important features like eye-gaze direction, facial asymmetry, emotional expressiveness, response latency, and social interaction cues, resulting in reliable and easy-to-understand predictions. Extensive testing shows that CDHAN outperforms than previous models in accuracy of 97.21%, precision of 95.73%, recall of 96.38%, and F1 Score of 95.47%, specificity of 95.84% while keeping a low error rate of 4% and achieving high generalization across various datasets. By providing an automated, scalable, and clinically useful approach and allows for quicker and more reliable ASD screening and early therapies that can improve children's development.
DOI: http://doi.org/10.5281/zenodo.20391418
