An accurate diagnosis of ADHD is crucial in bringing clarity and the right support to people who need it, but current diagnosis methods are time-consuming and inconsistent. A new study suggests AI could help.
Researchers in South Korea trained machine learning models to connect characteristics in photos of the fundus at the back of the eye to a professional diagnosis of ADHD (attention deficit hyperactivity disorder).
Of four machine learning models tested in the study, the best achieved a 96.9 percent score for predicting ADHD accurately, based on image analysis alone.
The team found that higher blood vessel density, shape and width of vessels, and certain changes in the eye’s optic disc were key signs someone had the condition.
For several years it’s been thought that brain connectivity changes associated with ADHD could also show up in our eyes. If we can figure out what to look for, this could mean a faster, more reliable method for spotting the disorder.
“Our analysis of retinal fundus photographs demonstrated potential as a noninvasive biomarker for ADHD screening and executive function deficit stratification in the visual attention domain,” write the researchers, led by a team from Yonsei University College of Medicine, in their published paper.
The approach was tested on 323 children and adolescents already diagnosed with ADHD, and another 323 without an ADHD diagnosis, matched by age and sex to the first group.
The researchers found the AI system scored highly across several measures when it came to predicting ADHD. It also performed well at spotting some of the characteristics of the disorder, including impairments in visual selective attention.
Several machine learning techniques to screen for ADHD have been explored recently, from analysis of alternative eye scans to behavioral tests, but this one has a few major drawcards. While not the absolute most accurate method in terms of raw scores, it’s very close, it’s also quick to run and assess, and simple to scale up.
“Notably, earlier high-accuracy models typically relied on a diverse set of variables, each contributing incrementally to differentiating subjects,” write the researchers.
“Our approach simplifies the analysis by focusing exclusively on retinal photographs. This single-source data strategy enhances the clarity and utility of our models.”
Next, the researchers want to try these tests across larger groups of people and wider age ranges. The average age of participants in this study was 9.5 years, and we know ADHD in adults can present quite differently.
There’s also room for improving the scope of the system: for example, those with autism spectrum disorder were excluded from the main part of this study, but further tests showed the AI wasn’t great at distinguishing autism from ADHD.
Recent estimates suggest about 1 in 20 people have ADHD, which can involve struggles with attention, impulses, and hyperactivity. That’s a lot of individuals for whom a quicker, more accurate diagnosis could make a difference.
“Early screening and timely intervention can improve social, familial, and academic functioning in individuals with ADHD,” write the researchers.
The research has been published in npj Digital Medicine.
