The barriers to getting a formal ADHD diagnosis are many: cost, time, the availability of qualified clinicians, a general lack of awareness.
Plus, those with inattentive ADHD—which is believed to be more common in women and girls—often struggle with completing long or complex tasks, ironically making it extra difficult to go through the lengthy process of getting diagnosed.
A further complication is the fact that ADHD doesn’t show itself in ways that can be objectively observed on a brain scan, X-ray or MRI. Instead, most cases are diagnosed by medical professionals through clinical interviews or standardized questionnaires.
Despite affecting an estimated 7.2% of the population, including 11.4% of all children, an estimated 80% of those with ADHD never receive a formal diagnosis.
“It usually comes down to a judgment call from a specialist provider,” says Elliot Hill, a biostatistician in the Department of Family Medicine at Duke University. “There’s no, ‘yes, you definitely have ADHD,’ or ‘you definitely have autism, or schizophrenia.’ The brain is too complicated, and our understanding is limited.”
Studies also show that living with undiagnosed ADHD can have a wide range of negative outcomes, including challenges in academic and professional environments, due to distractibility, impulsivity, hyperactivity, poor memory, and deficient time management skills. Without proper coping strategies, treatment or medication, those with ADHD tend to be underemployed and have significantly lower lifetime earnings.
They also suffer from higher rates of divorce, car accidents, substance abuse, unplanned pregnancies, eating disorders—even suicide attempts.
Artificial intelligence, however, is quickly proving itself a game changer in ADHD diagnosis.
The technology can spot patterns in healthcare records and flag those who may be at risk. Though clinicians are still a necessary part of the process, tools that search for symptoms and alert patients or parents are already proving effective.
Searching for needles in a haystack of medical data
One of those tools was developed by Hill and his coresearchers.
Together they developed an AI algorithm to analyze the electronic health records of children aged 9 and younger and spot patterns among those who are diagnosed with ADHD.
“Modern electronic health records have hundreds of thousands of variables that you could include in a model, and you just don’t know what’s important,” Hill says. “All sorts of things that we didn’t realize were important wound up being important.”
Some of the variables the algorithm flagged as potential ADHD indicators were expected, like scholastic delays, behavioral challenges, psychiatric conditions, anxiety, and sleep disorders.
“But then we also found some more oddball stuff, like vitamin D deficiency,” Hill says. “I’m not exactly sure how or if that’s causally related to ADHD, although just recently a paper was published along the same lines.”
Among the individuals whom the model flagged as highly likely to have ADHD, 92% were diagnosed with the condition by a trained clinician.
“Our study was about ‘Is this possible?’ ‘Is it feasible?’ ‘Can we actually do it with an accuracy that would be sufficient for clinical practice?’ And the answer to all those questions wound up being ‘yes,’” Hill says.
“What we really stress is that these tools that we’re building—they’re not the final decider,” he says. “They’re not an AI doctor who’s making the diagnosis. They’re just another tool in the providers’ bag that they can pull out to help evaluate patients more efficiently.”
Why an early diagnosis can be life-changing
Researchers who engaged in a similar experiment at the University of Alberta hope the technology will be used to flag children for an assessment before they begin to struggle in traditional academic environments.
“If someone is diagnosed with ADHD and treated effectively in first grade, rather than waiting until they’re 16, they could perform better in school, which could change the entire direction of their lives,” says Dr. Yang Liu, a research associate at the University of Calgary’s Department of Psychiatry and one of the study’s coauthors. “If you don’t recognize there’s a problem, you could miss that opportunity to intervene and educate the parents, as well as the kids, and provide coping strategies and therapies.”
The machine learning model developed by Dr. Liu and his coresearchers was proven as accurate as clinicians, but he similarly cautions against taking humans out of the equation.
An assistant, not a replacement
Dr. Liu explains that AI is still not 100% accurate, for the very reason such tools are necessary.
If the majority of those with ADHD are not formally diagnosed, the data that’s used to train AI and machine learning models can’t be entirely accurate, as it mischaracterizes those with undiagnosed ADHD as neurotypical.
“AI and technology can help clinicians save time, improve their efficiency and accuracy,” Dr. Liu. “I just want the clinicians to be always cautious when relying on AI until it’s proven to be 100%, which I don’t think is going to happen.”
Instead, he says algorithms can help flag potential ADHD cases that might otherwise go untreated. “It’s not a diagnosis; it’s a tool to pick up high-risk individuals so clinicians can talk to their parents,” Dr. Liu says.
Looking for physical evidence in a virtual world
AI might not be ready to offer reliable ADHD diagnoses on its own. But the technology may be able to finally spot physical markers of the condition when paired with other technologies, like wearables and virtual reality.
In 2018, a group of Finnish researchers designed a VR simulation that put users in a home-like environment and asked them to complete a series of tasks within an allotted time frame. In 2021, the tool was turned into a product called EFSim, which has since been adapted into a web browser game. Now, children ages 8 to 16 can play the video game for 25 minutes—and their parents and clinicians receive a report that breaks down some of the children’s cognitive strengths and challenges, and offers specific recommendations.
“There’s this virtual character that gives you a set of instructions, but there’s also a lot of other things that you can do,” says Juha Salmitaival, the head of the Translational Cognitive Neuroscience Lab at Aalto University in Finland, who cocreated the tool. “During this task performance, we record different behavioral parameters, like planning and task structure, memory-related things, and attention, which we measure in many ways, including eye and head movement.”
EFSim has been cleared for medical use and has been used in hospitals in Europe since 2021. “The diagnosis cannot be given based on a virtual reality game only. But it is something that can be used in the process,” Salmitaival says.
In the future, Salmitaival believes wearable technologies could be paired with AI to make mental health assessments the same way today’s tools count steps and heart rate.
The hope is that these tools will eventually lead to a more standardized assessment in the future. Even if ADHD can’t be picked up on a brain scan, MRI or X-ray, it may be identifiable on an AI-enabled wearable device.
“Objective [measurement] methods are critical, so that we could discriminate between our experiences and our cognitive abilities,” Salmitaival says.
