Why Bad Sleep Hits ADHD 10x Harder Than Everyone Else
Eye Scan Detects ADHD at 89% Accuracy. Now What?
A 17-year-old built a tool that reads your retina for ADHD at 89% accuracy. The hardware it runs on is already sitting in your eye doctor’s office.
RetinaMind, developed by Edward Kang — a high school senior at Bergen County Academies in Hackensack, New Jersey — analyzes standard retinal fundus photographs and accurately diagnoses neurodevelopmental disorders including ADHD and autism 89% of the time. The project won second place and a $175,000 prize at the 2026 Regeneron Science Talent Search, the oldest and most prestigious STEM competition for high school students in the United States. Psychiatric Times covered it as part of their 2026 Regeneron STS research coverage.
This story has a different shape than the Cambridge urine biomarker finding from last week. That one requires lab infrastructure most GP offices don’t have and would need to be built out. RetinaMind runs on fundus cameras — the machines optometrists already use for routine eye exams. They’re in every optometry practice in the country.
That infrastructure gap is the whole story.
TL;DR
The Finding The Reality RetinaMind: 89% accuracy diagnosing ADHD from retinal images Not available in any clinic today — still in research phase Built by 17-year-old Edward Kang, 2026 Regeneron STS finalist AI detects subtle retinal patterns undetectable by human clinicians Fundus cameras already exist in every optometry office No new hardware needed for clinical rollout Faster path than urine biomarker or MRI approaches But “faster” still means years, not months Average US wait for ADHD diagnosis: 12–24 months That bottleneck is specialist access, not diagnostic technology What this means right now: Retinal screening may be the fastest route to non-specialist ADHD diagnosis once it clears clinical validation. That process takes years. Start documenting symptoms and exploring telehealth routes in the meantime.
What the Retina Actually Reveals About ADHD
The retina isn’t just an eye structure. Developmentally, it’s an extension of the central nervous system — retinal tissue and brain tissue share the same embryonic origin. That’s why it keeps showing up in biomarker research for neurological and psychiatric conditions at all.
The differences RetinaMind reads are subtle. Not visible during a standard exam. Not something a clinician catches by looking.
Research on retinal structure in ADHD consistently finds reduced thickness in the retinal nerve fiber layer (RNFL) — the innermost layer, which carries signals from photoreceptors toward the optic nerve. A 2025 study published in npj Digital Medicine analyzed 1,108 retinal images from 646 children and adolescents and found that machine learning models achieved up to 96.9% AUROC accuracy for ADHD screening. The key features: blood vessel density, vessel shape and width, and structural changes around the optic disc.
These differences are sub-perceptual. A clinician cannot look at a retinal image and diagnose ADHD. The model can.
RetinaMind goes one step further than most research implementations. It produces a heat map of the retinal image — highlighting, in red, the specific regions that drove the diagnosis. Not just “ADHD positive” but a visual record of where in the retina the model found its evidence. That kind of interpretability is rare in diagnostic AI, and it matters for clinician trust in the output.
The underlying claim — that ADHD produces measurable structural differences in the retina that machine learning can reliably detect — has now been replicated across independent research teams. That’s not a single study. The signal is real.
Why This Path Is Faster Than Urine or MRI
Three non-behavioral ADHD diagnostic approaches are in various stages of research right now. Where each one sits in the infrastructure problem determines realistic timelines.
Urine biomarker (Cambridge, 2026). The Cambridge study identified N,N-dimethylglycine as a candidate ADHD-specific urinary marker. If clinical trials confirm it, a GP could theoretically order it alongside routine bloodwork. The infrastructure problem: urine metabolite testing for psychiatric biomarkers requires mass spectrometry equipment most GP labs don’t have. Rollout means buildout. A procurement and training wave across thousands of practices, each of which needs to acquire equipment, calibrate it, and integrate it into existing workflows. Even the optimistic scenario plays out over many years.
MRI neuroimaging (research context only). Structural and functional MRI can identify ADHD-associated brain patterns with increasing specificity. The infrastructure problem: MRI costs $400–$2,000 per scan, machines aren’t in most primary care offices, and scan time is significant. Valuable for research and ambiguous diagnostic cases. Not a population-level screening path.
Retinal imaging (RetinaMind and related). Fundus cameras are standard equipment in optometry offices — the standard tool for glaucoma screening and diabetic retinopathy checks. The hardware is already there, already in use, already covered by insurance for optical purposes. Most adults see an optometrist once a year.
The clinical rollout path for retinal ADHD screening doesn’t require a procurement wave. It requires regulatory clearance and integration into existing optometry workflows. That’s still not fast. But it’s a categorically different kind of lift than building new lab infrastructure or expanding MRI access across the healthcare system.
This is the part of the story that actually changes the realistic timeline.
How Does Retinal ADHD Diagnosis Work?
Retinal ADHD screening uses a standard fundus camera to photograph the back of your eye, then runs the image through a machine learning model. The model detects subtle structural differences in the retinal nerve fiber layer, blood vessel patterns, and optic disc — features statistically associated with ADHD but invisible to human observers. No dilation required for most fundus cameras. The exam takes under five minutes and is already routine in optometry practices for other conditions.
That’s the core of it.
Where RetinaMind Stands Clinically
RetinaMind is a research project that won a competition. It’s not an FDA-cleared diagnostic tool. It’s not available at any clinic or optometry practice today.
The path from “impressive research result” to “your optometrist can order this” requires several stops:
- Peer-reviewed publication and independent replication. A competition project needs published data that other teams can independently verify across diverse populations.
- Prospective clinical validation. Accuracy on training datasets needs to hold up in real clinical populations, across different demographics and comorbidity profiles.
- FDA clearance. Any diagnostic software used in clinical decision-making needs De Novo or 510(k) clearance before it can legally be applied in US clinical settings.
- Workflow integration and payor coverage. Optometry practices need to integrate the software, handle the liability questions, and get insurance carriers to cover the indication.
Steps 1 through 4 realistically take three to seven years from where RetinaMind stands now. That’s the honest version of the timeline. Not pessimism — just how diagnostic validation actually works.
The average wait for formal ADHD diagnosis in the US is still 12 to 24 months in 2026. But that gap isn’t a technology problem. It’s a specialist access problem. The diagnostic technology that already exists — behavioral assessment, rating scales, clinical interviews — is sufficient for diagnosis. The bottleneck is the number of clinicians trained to use it for adults, in the places where adults actually are.
RetinaMind’s eventual value is changing who needs to be in that clinical chain. If a routine annual eye exam generates a retinal flag that triggers a referral, that’s a screening pathway that doesn’t require a psychiatrist as the first point of contact. You’d get flagged at the optometrist, referred to a GP or specialist, and arrive at that appointment with objective data already in hand. That’s the long-term shift worth tracking.
What to Do Right Now
The diagnostic infrastructure isn’t changing this year. But your personal timeline can still move.
If you’re undiagnosed and suspect ADHD:
The ASRS-v1.1 Adult ADHD Self-Report Scale is free, takes 10 minutes, and is the validated screener clinicians use as a first-step flag. A positive result is documented starting material you can hand to a GP when requesting a referral. It’s in their language. It signals you’ve engaged methodically.
Start logging specific incidents today — not feelings, but events. “Lost car keys Tuesday, 25 minutes late to a call” is more clinically useful than “I feel scattered.” Behavioral diagnosis depends on documented evidence of functional impairment across settings. A few weeks of consistent logging before an appointment changes what that appointment produces.
The pattern that runs through the AI detection research and the biomarker work is the same: the diagnostic signals have been present all along in people who were never diagnosed. The late-diagnosis gap wasn’t about evidence that didn’t exist — it was about systems that weren’t looking, or weren’t looking in the right way. Your job is to make the evidence visible when someone finally does look.
If you’re currently on a referral waitlist:
Telehealth routes can compress a 12-month wait to 2–4 weeks in many US states. The regulatory environment tightened significantly after 2022 — the providers still operating tend to run more rigorous multi-step evaluations than the early telehealth wave, which means you get a more thorough evaluation, not a worse one. Look for licensed prescribers with explicit adult ADHD specialization, not general mental health platforms that handle ADHD as one condition among many.
While you wait, the evidence-based productivity strategies that work for undiagnosed adults are the same ones that work post-diagnosis. You don’t need a formal diagnosis to start accommodating your brain better. You might need one to access medication, formal workplace accommodations, or ADHD-specific therapy — but the systems work regardless of paperwork status.
On tracking the research:
RetinaMind’s website is the best place to follow the project as it moves toward publication and validation. The 2025 npj Digital Medicine study on retinal ADHD biomarkers is the underlying peer-reviewed science supporting this approach. And the CHADD weekly news digest is worth following for any clinical trial announcements on the retinal pathway.
If retinal ADHD screening reaches clinical availability in the next decade — and the infrastructure argument makes it more plausible than most other diagnostic innovations — it will almost certainly start in optometry offices rather than psychiatry clinics. That’s the referral chain to watch.
The Bottom Line
RetinaMind is a real signal. The retinal biomarker research has been replicated across independent teams. The infrastructure advantage over urine testing or MRI is concrete and specific. A 17-year-old built a version of this tool that works at 89% accuracy — that’s not noise.
The clinical timeline is years, not months. And the 12-to-24-month specialist queue defining ADHD diagnosis in 2026 has nothing to do with diagnostic technology. The technology works. The bottleneck is specialist access.
Track the research. Document your symptoms now. Use the tools that already exist.
The version of ADHD diagnosis where your optometrist catches it at a routine annual eye exam is coming. It’s probably five to ten years out. The work you need to do this week doesn’t change while you wait for it to arrive.
This post reflects publicly available research about ADHD diagnostic approaches. It is not medical advice. Speak with a qualified healthcare provider about your specific situation.
