04 Design-Driven Federated Learning Framework for Multimodal Neurodevelopmental Assessment: A Human-Centered Approach to Pediatric Mental Health Screening

Abstract

Pediatric mental health screening faces critical challenges in accessibility, cultural sensitivity, and privacy protection, limiting early intervention opportunities for neurodevelopmental conditions.  Here  we  show  that  integrating   human-centered design principles with federated learning creates a transformative framework for multimodal neurodevelopmental assessment. Our Design-Driven Federated Learning (DDFL)  framework  combines  adaptive  user  interfaces,  cultural  intelligence  systems,   and privacy-by-design architectures to enable collaborative analysis across diverse healthcare institutions while maintaining data sovereignty. Through comprehensive evaluation across five international pediatric healthcare institutions spanning different cultural contexts, we demonstrate substantial improvements in diagnostic accuracy (12.3-18.2% across ADHD, ASD, and learning disabilities), user experience metrics (33% improvement in completion rates), and cross-cultural performance consistency  (86.9- 91.2% accuracy  across  all  cultural  groups).  The  framework  successfully  balances privacy protection with diagnostic effectiveness, achieving 85.2% accuracy under strict privacy constraints while baseline methods reach only 76.3%. These findings establish design innovation as a fundamental driver of medical  AI  effectiveness,  providing  a scalable solution for global pediatric mental health screening that  prioritizes  both technical excellence and human factors.