Comparative Study of AI vs. Traditional Diagnostic Methods for Autism Spectrum Disorder: Demonstrating Real-World Superiority Through Multi-Site Clinical Validation
Posted: Dec 09, 2023
Abstract
The diagnostic assessment of autism spectrum disorder has traditionally relied
on clinician-administered observational tools and caregiver interviews, approaches
that while valuable face significant limitations in standardization, accessibility, and
scalability. This comprehensive comparative study evaluates the performance of artificial intelligence diagnostic systems against traditional assessment methods across
multiple clinical sites and diverse patient populations. We conducted a prospective multi-center trial involving 2,840 children aged 18-96 months across 28 clinical sites, comparing three AI diagnostic approaches—multimodal deep learning,
computer vision analysis, and natural language processing—against gold-standard
traditional methods including the Autism Diagnostic Observation Schedule-Second
Edition (ADOS-2) and clinical expert diagnosis. The AI systems demonstrated
significantly superior performance, with the multimodal deep learning approach
achieving 94.7% diagnostic accuracy compared to 87.3% for ADOS-2 and 85.1% for
clinical expert diagnosis. The AI methods reduced average diagnostic time from 186
minutes to 47 minutes while maintaining higher inter-rater reliability (Cohen’s =
0.92 vs 0.76) and demonstrating better consistency across demographic subgroups.
Crucially, AI systems identified 89.2% of cases missed by initial traditional assessment while maintaining specificity above 93% across all validation cohorts. The
implementation of AI diagnostics increased early intervention access by 42% and
reduced diagnostic disparities in underserved populations by 67%. These findings
provide compelling evidence for the real-world superiority of AI-assisted autism diagnosis, offering substantial improvements in accuracy, efficiency, accessibility, and
equity that address critical limitations of current diagnostic practices
Downloads: 98
Abstract Views: 1196
Rank: 34678