High-School Inventor’s AI Uses the Retina to Diagnose Autism and ADHD

Seventeen-year-old Edward Kang, a senior at Bergen County Academies in Hackensack, New Jersey, created RetinaMind, a convolutional neural-network system that analyzes retinal photographs to predict autism spectrum disorder (ASD) or attention-deficit/hyperactivity disorder (ADHD). The model produces a heat-map highlighting retinal regions influencing its decision and achieves an overall accuracy of about 89 percent. Kang’s work earned second place and a $175,000 award at the 2026 Regeneron Science Talent Search.

Smithsonian Magazine

Kang taught himself programming and machine-learning basics, first replicating a Chinese University of Hong Kong study that linked retinal images to autism. He then added ADHD detection, employing ensemble learning—multiple models receive the same image, predict independently, and their outputs are averaged for a more reliable result. To interpret the model, he applied Grad-CAM, an explainable-AI technique that visualizes which retinal areas drive each prediction. Since late 2024 he has also pursued cell-biology experiments, identifying candidate genes such as ABCA4 that may underlie retinal differences in neurodivergent individuals.

Smithsonian Magazine

Neurodevelopmental pediatrician Paul Lipkin of the Kennedy Krieger Institute praised the concept of earlier screening but warned that retinal markers might reflect broader brain-based conditions rather than ASD or ADHD specifically. Maya Ajmera, president and CEO of Society for Science, highlighted the project’s blend of computational sophistication and biological depth, noting its relevance to families facing lengthy diagnostic waits.

Smithsonian Magazine

Both Lipkin and Kang acknowledge that RetinaMind currently offers only a binary diagnosis—autism or ADHD—without distinguishing severity levels or accounting for overlapping neurodevelopmental disorders. The tool’s reliance on subtle retinal features, which overlap with normal variation, raises concerns about specificity and potential false-positive rates in real-world clinical settings.

Smithsonian Magazine

• “I thought it was fascinating and really unintuitive that you can use something like the eye to understand what’s happening in the brain,” — Edward Kang, high-school senior

Smithsonian Magazine

• “Distinguishing between neurotypical individuals and those with autism is not very difficult, and existing studies have already achieved close to 100 percent accuracy,” — Edward Kang

Smithsonian Magazine

• “One potentially interesting gene I identified is ABCA4, which encodes a protein responsible for detoxifying the retina,” — Edward Kang

Smithsonian Magazine

• “Edward’s project stood out for combining A.I. with lab-based biology, which gave it both computational sophistication and biological depth,” — Maya Ajmera, Society for Science

Smithsonian Magazine

• “Any retinal differences identified may not be specific for these conditions, but instead of some brain-based neurologic condition generally,” — Paul Lipkin, Kennedy Krieger Institute

Smithsonian Magazine