An artificial intelligence tool can detect autism spectrum disorder with 100-percent accuracy, just by scanning images of children’s eyes, according to a new study.
If confirmed, this would be a major breakthrough for detecting the condition. But multiple autism experts told DailyMail.com that the number is unrealistic, and the result is probably ‘too good to be true.’
Autism affects an estimated 1 in 36 children in the US, but many children remain undiagnosed until later in childhood, depriving them of potential therapies.
If a technological solution could help cut down on long waits for autism specialists or other obstacles to diagnosis, it could benefit millions of families.
A new AI tool can detect autism with 100% accuracy from retinal scans, its inventors say. Autism experts are not convinced, saying the results are ‘too good to be true’
Autism is a condition involving altered brain development, and the optic nerve connects the retina to the brain in a very short path.
So it stands to reason that brain differences could be reflected in the eyes.
Dozens of news outlets picked up on the news of the AI tool, developed by a team of researchers at Yonsei University in Seoul.
But experts say it is too soon to trust these findings, and that the research raises multiple red flags – starting with that 100 percent accuracy figure.
‘There is clearly something wrong here,’ Fred Shic, an autism researcher at Yale School of Medicine, told DailyMail.com. Shic researches eye tracking and imaging techniques in autistic children.
There is no way that this test is more accurate than doctors, he told DailyMail.com. ‘That reliability is not 100 percent, even amongst the best clinicians in the world.’
Other autism experts share Shic’s skepticism.
‘It just seems too good to be true,’ Cathy Lord, distinguished professor of psychiatry at the University of California Los Angeles, told DailyMail.com.
Lord is the co-creator of the Autism Diagnostic Observation Schedule-Second Edition (ADOS-2), the gold-standard clinical tool used to assess the children in the new study.
She said she hopes other researchers will try to replicate the findings – performing the experiment again and comparing results with this one.
‘It seems worth trying to replicate but I’m very skeptical,’ she added.
DailyMail.com has reached out to the study’s authors and will update this story if we receive a response.
The study in question involved 958 children: 479 with autism and 479 without autism.
Both groups had the same split of boys and girls – 82 percent boys and 18 percent girls – which lines up with the 4:1 sex ratio found in most countries.
The researchers fed images of children’s retinas to train the algorithm, excluding children with other psychiatric conditions that could complicate or confuse the results.
Specialists screened the children with the ADOS-2 to confirm that they had autism and to assess how pronounced their autism traits were.
A deep neural network was trained to use iris scans to differentiate between the children with and without autism. It also learned how to connect autism trait severity to the retinal scans.
Retinal scans have been found effective at screening for some conditions, like Alzheimer’s
When the AI tool was tested on a separate set of the children than the one it was trained on, it accurately detected children’s diagnosis 100 percent of the time, according to the study, which was published in JAMA Network Open.
Furthermore, it could determine autism severity with about 74 percent accuracy based on retina scans alone.
The idea of scanning the retina to detect autism is ‘intriguing and promising,’ Geri Dawson, director of the Duke Autism Clinic, told DailyMail.com. ‘Changes in the retina have also been used to predict Alzheimer’s Disease.’
Multiple studies have shown that autistic and neurotypical people have significant differences in the nerves of the retina.
But further work is necessary to tell whether the differences between the two groups of children are due to autism or some other factor.
‘The differences might be an indicator of brain changes more broadly associated with cognitive disability, for example,’ said Dawson. ‘Also, the authors note that many of the autistic participants were taking medications that could have affected the retina.’
The average intelligence quotient (IQ) of the autism group was 70, right at the border for a diagnosis of intellectual disability.
The researchers did not report IQ scores for the non-autistic children, though, so this unaccounted-for factor adds complicates the study, Thomas Frazier, professor of psychology at John Carroll University, told DailyMail.com.
‘This makes the comparison to [typically developing children] even less realistic for clinical purposes,’ he said.
But even that would not account for 100-percent accuracy, Lord said.
‘If it was just IQ you’d still expect less perfect results.’
Multiple research teams are working on smartphone- or tablet-based apps to detect autism, but these apps focus on social attention, not on retinal scans
The AI model itself could also be the problem, several experts said.
Something besides the retina that gives away a child’s diagnosis could somehow be included on the images, nudging the AI to its unrealistic 100 percent figure, Shic said.
‘This could be as simple as a word describing the source of data from specialized [autism] clinics,’ he said. It could also include subtle changes in image quality between the two groups.
At an acoustical engineering conference in 2013, Shri Narayanan’s lab participated in a challenge where they developed a method for identifying autistic children from voice recordings.
They achieved very strong results, Narayanan, University Professor and Nikias Chair in Engineering at the University of Southern California, told DailyMail.com
But they turned out to be caused by a hidden factor: sound quality.
Their system’s performance was actually due to the difference between noisy mainline classrooms and quiet special-education classrooms – a difference reflected in the voice recordings – ‘which incorrectly provided the seemingly right answer,’ he said.
‘While the promise of being able to screen and diagnose a clinical condition with data and new (AI) tools is exciting and can be impactful, it needs to be done with extreme care and caution,’ said Narayanan.
There are already some apps under development that are meant to diagnose autism.
They tend to track where a child is looking, as opposed to the actual structure of their eyes.
Because social communication issues make up one of autism’s core traits, researchers have attempted to screen children for autism based on whether they pay more attention to objects than people.
One such app, under development by a team at Duke University, predicted autism diagnosis with 90-percent accuracy in a 2021 study.
Using eye-tracking software and smartphone cameras, this app showed toddlers videos of people talking and playing with toys, discerning whether toddlers were paying more attention to the toys or the people.
But any technology used to diagnose or treat a condition still must undergo testing and clearance by the Food and Drug Administration, and few studies of new techniques end up making it that far.