17-Year-Old Researcher Develops AI Tool to Screen for Autism and ADHD Using Retinal Images

Edward Kang, a 17-year-old student from Hackensack, New Jersey, developed RetinaMind, an experimental AI screening system that analyzes retinal images for patterns associated with autism spectrum disorder (ASD) and attention-deficit/hyperactivity disorder (ADHD). His project won second place and $175,000 in the 2026 Regeneron Science Talent Search.

Why Researchers Are Looking at the Retina

The retina is more than the light-sensitive tissue at the back of the eye.

Because it develops from the same embryonic tissue as the brain and forms part of the central nervous system, researchers have been investigating whether changes in the retina could provide measurable clues about neurological and neurodevelopmental conditions.

That possibility caught the attention of Edward Kang while he was still a teenager.

After encountering earlier research linking retinal images with autism, Kang began investigating whether artificial intelligence could identify subtle patterns that might be difficult to recognize through conventional visual inspection.

His resulting project, RetinaMind, combines machine learning with biological research to investigate whether an eye-based screening approach could eventually contribute to neurodevelopmental research.

How RetinaMind Uses Retinal Images

Kang trained AI models using retinal images from a large public dataset.

The models were designed to distinguish patterns associated with three groups: people with autism spectrum disorder, people with ADHD and neurotypical individuals.

Rather than simply producing a binary result, RetinaMind generates confidence estimates for the different categories. It also produces a heat-map visualization indicating regions of the retinal image that contributed to the model’s prediction.

This type of visualization is important in medical AI research because researchers need to understand what a model is responding to rather than relying entirely on an unexplained prediction.

Kang combined multiple machine-learning approaches and examined the factors influencing their predictions as part of the project. The official Regeneron Science Talent Search finalist profile describes the system as a screening tool rather than an established clinical diagnostic system.

The Model Reported About 89% Accuracy

In Kang’s testing, RetinaMind achieved an accuracy rate of approximately 89%.

That result attracted significant attention, but the number needs to be interpreted carefully.

An accuracy figure from a research dataset does not mean that the system can diagnose autism or ADHD with 89% certainty in a clinical setting.

Performance can change when an AI model is tested on different populations, different imaging equipment or data collected at other institutions. Clinical validation would require substantially broader testing before the technology could be considered for routine medical use.

The project is therefore best understood as an AI screening research prototype, not a replacement for clinical assessment.

From an AI Model to Biological Investigation

Kang’s project went beyond training a computer vision model.

He also created retinal cell models to investigate biological changes associated with the conditions.

According to Rutgers University, his research explored gene changes through retinal cell models and subsequently validated findings in a second cell model.

This second component gives the project a different dimension.

The AI system searches for patterns in retinal images, while the biological experiments investigate possible mechanisms that could help explain why those patterns might occur.

That combination connects computational neuroscience with experimental biology.

Why an Eye Scan Could Matter for Neurodevelopmental Research

Autism and ADHD are currently assessed through clinical and behavioral evaluation rather than a single retinal or imaging test.

That means researchers continue to investigate whether measurable biological signals could eventually complement existing assessments.

A retinal screening system could be particularly interesting because retinal imaging is already used in ophthalmology and can capture detailed structural information without requiring invasive procedures.

But identifying an association is not the same as establishing a diagnostic biomarker.

For RetinaMind to become clinically useful, researchers would need to determine whether its predictions remain reliable across different ages, ethnic and demographic groups, clinical settings, retinal imaging devices and individuals with overlapping or unrelated neurological conditions.

Those questions remain open.

Rutgers Experience Helped Connect the Technology to Clinical Research

Kang’s work also developed alongside his experience at the Rutgers Center for Autism Research, Education and Services (RUCARES) within the Rutgers Brain Health Institute.

During his senior experience program, he observed autism research and treatment in a clinical environment.

Rutgers reported that this experience helped connect his computational work with the practical complexities involved in autism research and care.

That connection is significant because medical AI development requires more than a high-performing algorithm.

Researchers also need to understand how a tool would fit into clinical workflows, what information clinicians need, how results should be communicated and what happens when an algorithm is uncertain.

A $175,000 Science Talent Search Recognition

Kang’s research earned second place and a $175,000 award in the 2026 Regeneron Science Talent Search.

The competition recognized 40 finalists in 2026, with more than $1.8 million awarded across the finalist group.

The first-place award went to Connor Hill for his mathematical classification of noble polyhedra, while Kang received the second-place award for RetinaMind.

Kang was a senior at Bergen County Academies in Hackensack, New Jersey, and is scheduled to attend the Massachusetts Institute of Technology.

What RetinaMind Could — and Cannot Yet — Do

The most important distinction is between screening research and diagnosis.

RetinaMind demonstrates that an AI model can identify patterns in retinal images associated with neurodevelopmental conditions. It does not establish that autism or ADHD can currently be diagnosed from a retinal image alone.

The project also does not demonstrate that retinal imaging can replace behavioral evaluation or specialist assessment.

Instead, it provides a research direction: if retinal characteristics contain reproducible signals associated with neurodevelopmental differences, AI may be able to help researchers identify and study those signals.

Further validation would be necessary to determine whether such signals are sufficiently specific, robust and clinically meaningful.

The Bigger Shift: AI Is Becoming a Tool for Biological Discovery

RetinaMind reflects a broader trend in biomedical research.

Machine learning is increasingly being used not only to analyze medical records but also to search images for patterns, identify potential biomarkers and generate hypotheses that can then be investigated experimentally.

In Kang’s case, the process moves in both directions.

AI searches retinal images for statistical patterns, while biological experiments explore possible explanations for those patterns.

That combination could become increasingly important as researchers attempt to connect large medical datasets with underlying biological mechanisms.

RetinaMind is not yet a clinical diagnostic test, but it demonstrates how a high-school research project can combine artificial intelligence, retinal neuroscience and experimental biology to investigate a difficult medical question.

The next challenge is no longer simply whether AI can recognize a pattern.

It is whether researchers can prove that the pattern is biologically meaningful, reproducible and useful in real clinical settings.

FAQs

What is RetinaMind AI?

RetinaMind is an experimental AI screening prototype developed by Edward Kang that analyzes retinal images for patterns associated with autism spectrum disorder and ADHD.

Who developed RetinaMind?

RetinaMind was developed by Edward Kang, a 17-year-old student from Bergen County Academies in Hackensack, New Jersey.

How accurate is RetinaMind?

Kang reported an accuracy of approximately 89% in his research testing. This result should not be interpreted as clinical diagnostic accuracy because the system remains a research-stage prototype.

Can retinal scans currently diagnose autism or ADHD?

No. RetinaMind is an experimental screening system, not an approved clinical diagnostic test. Autism and ADHD diagnosis still requires appropriate clinical assessment.

What prize did Edward Kang win?

Kang received second place and $175,000 in the 2026 Regeneron Science Talent Search for his RetinaMind research.