Teen develops AI to identify ADHD & Autism
How a 17-Year-Old Is Using AI to Analyse Retina Images for Earlier Autism and ADHD Screening
Artificial intelligence is usually discussed in terms of what it can automate, improve or create. But some of its most meaningful applications of AI can be found in ways in which it improves people’s lives.
In recent times, a remarkable example of this perspective comes from 17-year-old Edward Kang, a student from Hackensack, New Jersey, who developed an AI research project called RetinaMind. His work explores whether artificial intelligence can analyse images of the retina and identify subtle patterns associated with Autism Spectrum Disorder (ASD) and Attention-Deficit/Hyperactivity Disorder (ADHD).
Kang’s project earned him second place and a $175,000 award at the 2026 Regeneron Science Talent Search, one of the United States’ leading science competitions for high school students.
More importantly, RetinaMind offers an intriguing glimpse into what AI for good can look like: technology being developed not for profit or speed, but to potentially help people receive support at an earlier stage.
Looking for clues in the eye
Let’s look into the story of how this project was formed. At first, the idea of using a retinal image to study conditions such as autism and ADHD might sound surprising.
After all, these are neurodevelopmental conditions, not eye disorders. So why look at the retina?
Part of the answer is in the biological relationship between the eye and the brain. The retina is part of the central nervous system, and the eye and brain develop from related embryonic tissues. Previous research has identified differences in retinal characteristics among people with neurodevelopmental conditions, although these differences can be minute and challenging for humans to pinpoint visually.
That is where AI becomes more than a technology – a lifesaver.
Instead of relying on a person to recognise a pattern in a retinal image, a machine-learning model can examine large amounts of visual data and learn combinations of features that may be difficult for humans to detect.
Kang began exploring this idea after coming across research that investigated retinal images in relation to autism. What started as an attempt to improve an existing model eventually developed into a much broader research project involving AI, neuroscience and cell biology.
How does RetinaMind work?
RetinaMind uses artificial intelligence models trained on retinal images to identify patterns associated with different neurodevelopmental conditions.
One of the technologies Kang used was a Convolutional Neural Network (CNN), a type of deep-learning model particularly suited to analysing images.
The basic idea is relatively straightforward.
A retinal image is provided to the model. The AI analyses visual patterns within the image and produces a prediction. Kang then combined multiple modelling approaches through ensemble learning, allowing different models to contribute to the final result.
The project ultimately achieved an accuracy of approximately 89% across the groups being studied in the research setting. It also produced visual heatmaps using an explainable-AI technique known as Grad-CAM, highlighting areas of the retinal image that influenced the model’s prediction.
That explainability is particularly important.
When AI is used in healthcare, simply producing an answer is not enough. Researchers and clinicians need to understand, as far as possible, why a model reached a particular conclusion.
An AI system that says “this image indicates X” is less useful than one that can also indicate which parts of the image influenced its prediction.
Why earlier screening matters
For families, the process of identifying developmental conditions can be challenging, time-consuming, and involves a good deal of cost.
Autism and ADHD are assessed through behavioural, developmental and clinical information rather than a single simple test. Evaluations can involve multiple professionals and, in some cases, take considerable time.
That creates an important opportunity for AI.
If technologies such as RetinaMind can eventually be validated and developed into reliable screening tools, they could potentially help identify individuals who may benefit from further clinical evaluation.
This does not mean that an AI system should replace doctors, psychologists or other healthcare professionals.
Instead, the goal is to create another tool that can support them in making accurate diagnoses with a much faster turnaround time compared to the existing systems.
Imagine a future in which a quick retinal image could provide an additional piece of information during an assessment. An AI system could flag a potential risk, prompting a clinician to investigate further.
For families, that could mean reaching the next stage of assessment sooner.
And earlier identification can matter because it can open the door to earlier access to appropriate support, interventions and educational resources.
AI that supports people, rather than replaces them
This is one of the most important lessons from AI systems like RetinaMind.
The most valuable use of AI in healthcare may not be about replacing human expertise. It may be about augmenting it.
Doctors bring clinical experience, empathy, context and an understanding of the individual patient. AI brings the ability to process enormous amounts of data and detect patterns that may be difficult to identify manually.
Together, those strengths can complement each other.
RetinaMind itself is still a research-stage project, and its results should not be interpreted as proof that a retinal photograph can independently diagnose autism or ADHD. Experts have also pointed out that retinal differences may not necessarily be specific to these conditions, and that further research and validation are needed.
That caution is not a weakness of the project. It is an essential part of responsible AI development.
From a school project to a scientific breakthrough
Perhaps one of the most inspiring parts of the RetinaMind story is where it began.
Kang was a high school student who encountered an interesting scientific paper and decided to explore the idea further. He taught himself programming and machine learning through online resources and developed multiple versions of his model.
His work eventually went beyond simply building an AI model.
Kang also investigated the biological mechanisms that could potentially explain retinal differences associated with autism. His project included retinal cell models and research into gene expression, creating a bridge between computational AI and biological science.
This combination is particularly significant.
AI can identify patterns, but understanding why those patterns exist can lead to deeper scientific discoveries.
In this case, the technology is not only asking whether an AI model can recognise a signal. It is also contributing to a broader investigation into the relationship between the retina, the brain and neurodevelopmental conditions.
What AI for Good really means
The phrase “AI for Good” can sometimes sound abstract. Something that is just for PR.
RetinaMind makes the idea much more tangible.
AI for good is about applying technological capabilities to challenges that matter to society – healthcare, accessibility, education, environmental protection, disaster response, scientific discovery and many other areas.
Any technology in itself is not automatically “good”. Its impact depends on how it is designed, tested and used.
When AI is developed around a genuine human need, carefully validated and used responsibly, it can become a powerful tool for positive change.
RetinaMind represents this possibility.
Instead of asking AI to generate another piece of content or automate another routine task, Kang used it to explore a difficult scientific problem: Could something as accessible as an image of the eye contain clues that help us identify neurodevelopmental conditions earlier?
The answer is not yet definitive.
But asking that question, and using AI to investigate it, could open the door to new possibilities.
The future of AI-powered healthcare
There is still a long road ahead for RetinaMind.
The model needs further research, validation and testing before technology like this could potentially become part of real-world clinical practice. Researchers will also need to examine how well such systems perform across different populations, imaging conditions and clinical settings.
But that is exactly what makes projects like this valuable.
Innovation does not always arrive as a finished product. Sometimes it begins as a question, a prototype or a school science project.
Edward Kang’s work demonstrates how artificial intelligence can bring together computer science, medicine and biology to explore problems that have a direct impact on people’s lives.
And perhaps that is the bigger message.
The future of AI should not only be about making machines smarter. It should also be about making technology more useful to humanity.
From helping conserve water to supporting medical research, AI has the potential to become a tool for solving some of society’s most difficult challenges.
RetinaMind is one small but fascinating example of that vision.