Enhancing Autism Diagnosis Using Federated Learning and Multimodal Data Fusion

Autism spectrum disorder is a neurodevelopmental condition that is complex by nature, and therefore proper diagnosis at the right time is essential to manage it. The traditional approaches to diagnose autism have been mainly stand-alone data sources like neuroimaging, genetic analyses, or clinical observations alone, whichcannot reflect the multifaceted nature of ASD. It brings forth a novel framework in leveraging federated learning combined with multimodal data fusion toward neuroimaging, clinical, and genetic data in diagnosis of ASD. Since it uses federated learning, the model is collaboratively trainable without having to share private data related to patients across institutes while ensuring compliance with the need for privacy. Our results show that thismodel is better than a central and single-modality-based system in terms of precision, recall, accuracy, and F1-score. So, the potential of such techniques ishigh in improving diagnostic ASD systems while keeping all personal data confidential. Finally, with this paper, conclusions related to future directions.