New AI Technology Could Detect Autism Earlier Across All Ages—Here's Why It Matters

A new artificial intelligence approach could make autism screening faster, fairer, and more accurate across children, teenagers, and adults—without compromising patient privacy. Researchers from Prince Sultan University and the University of Engineering and Technology developed a federated learning system that achieves 97.2% accuracy in detecting autism spectrum disorder (ASD) in children, 89.5% in adolescents, and 86.8% in adults. Unlike traditional AI systems that require centralizing sensitive patient data in one location, this method trains models across multiple institutions while keeping personal health information completely private.

Why Current Autism Screening Falls Short

Detecting autism spectrum disorder has always been challenging because the condition manifests differently depending on a person's age and individual traits. Children show different behavioral patterns than teenagers or adults, yet most screening tools are built on data from just one age group. This creates a significant problem: a tool trained primarily on children's data often fails to accurately identify autism in teenagers or adults. Additionally, traditional machine learning approaches require pooling all patient data into a central database, raising serious privacy concerns that can discourage hospitals and clinics from participating in research.

How Federated Learning Changes the Game

The new system uses a technology called federated learning, which works like a collaborative team project where each member keeps their own work private. Instead of sending patient data to a central location, each hospital or clinic trains its own AI model using its local patient information. These individual models then share only their learnings—not the raw data—with other institutions. The system combines insights from all participating centers to create a more robust, accurate screening tool. This approach protects patient privacy while building a stronger AI system that learns from diverse populations.

What Makes This System More Fair and Accurate?

The researchers tested several personalized federated learning algorithms and compared them to traditional centralized approaches. The results were striking: customized federated learning methods outperformed standard machine learning techniques like support vector machines, random forests, and decision trees. The system achieved significantly higher accuracy across all three age groups, with the strongest performance in children at 97.2% accuracy. Beyond raw accuracy numbers, the system also demonstrated improved fairness—meaning it works equally well for different populations rather than performing better for some groups and worse for others.

The preprocessing steps the researchers used to prepare data included handling missing values, converting categorical information into usable formats, scaling measurements to comparable ranges, selecting the most important features for detection, and addressing class imbalance—a common problem where one outcome (like having autism) is much rarer than the other in training data. These technical refinements ensure the AI learns from clean, well-organized information.

Steps to Implementing Better Autism Screening in Your Community

• Advocate for Multi-Institutional Collaboration: Encourage your local pediatric clinics and hospitals to participate in federated learning networks that allow them to contribute to autism screening research without sharing sensitive patient data.
• Ask About Privacy-Preserving AI Tools: When your child is screened for developmental concerns, ask whether your healthcare provider uses privacy-safe AI systems that don't require centralizing your family's medical information.
• Support Age-Appropriate Screening: Ensure that screening tools used in your community are validated across different age groups—children, adolescents, and adults—rather than relying on single-age-group models.
• Promote Equitable Access: Work with school districts and community health centers to implement screening systems that have been tested for fairness across diverse populations and age ranges.

Why This Matters for Families Right Now

Earlier and more accurate autism detection can be transformative. When children are identified sooner, they can access behavioral therapies, educational support, and other interventions that significantly improve outcomes. The same applies to teenagers and adults who may have gone undiagnosed for years. However, current screening tools often miss cases because they're optimized for one age group or because they're trained on limited, homogeneous data. This new federated learning approach addresses both problems simultaneously.

The system also solves a practical barrier that has slowed autism research: privacy concerns. Many healthcare institutions are reluctant to share patient data, even for research that could help millions of people. Federated learning removes this barrier by allowing institutions to participate without ever exposing individual patient records. This means more diverse data can be included in training, leading to screening tools that work better for children from different backgrounds, socioeconomic situations, and geographic regions.

What's Next for Autism Screening?

The researchers demonstrated that their system is computationally efficient and practical to deploy in real-world clinical settings. This means the technology isn't just theoretically sound—it can actually be implemented in hospitals, clinics, and schools without requiring expensive infrastructure upgrades. As more institutions adopt federated learning approaches, autism screening could become faster, more accurate, and more equitable across age groups and populations. The research opens the door to a future where developmental screening is both ethically responsible and scientifically advanced.