Can AI Make Mental Health Care Smarter, Faster, and Fairer for Children?

The journey to effective help for challenges to mental health in children can be long. While there are many effective interventions, clinicians are often unsure which intervention will most likely help. So, there is usually a trial-and-error approach, trying one intervention, then another, and sometimes even another, until the most helpful one is identified. It is a process that even the most dedicated clinicians struggle to navigate, and it can be overwhelming for families. This is why my work looks at how Artificial Intelligence (AI) can support and improve this process.

As a researcher working at the intersection of mental health, artificial intelligence, neuroimaging, and genomics, I’ve spent the past decade asking how we can improve. Can we make mental health interventions more precise and more inclusive? Can we move away from relying on observable behaviours and start using objective biological data to guide understanding, diagnosis, and treatment? And perhaps most importantly, can we ensure these improvements reach all children, not just a few?

I recently moved to King’s College London from the US-based National Institutes of Health to build a team focused on addressing exactly these questions. And while AI can’t fix everything, we’re seeing more and more ways it could meaningfully transform youth mental health care.

From symptoms to biology

Much of our current approach to mental health starts with symptoms—what a young person reports, how they behave, and how others describe them. That’s valuable, but it’s only part of the picture. Mental health conditions like attention-deficit/hyperactivity disorder (ADHD) and obsessive-compulsive disorder (OCD) often vary widely between individuals: this variability isn’t captured by the diagnostic label. That’s where AI can help.

We’ve used machine learning to uncover brain-based subtypes—we sometimes call these “biotypes”—in large datasets of children living with ADHD. These subtypes reflect real differences in brain function that aren’t always visible through clinical observation. And they matter: we’re beginning to find that children with different biotypes may respond differently to different treatments.

For example, one subtype we identified involves disruptions in attention-related brain circuits, and those children tend to respond particularly well to stimulant medications. We’re now asking if other subtypes or biotypes might benefit more from non-stimulant medication or behavioural therapies. The idea is simple: tailor treatment to biology, not just behaviour. But the potential impact—faster recovery, fewer side effects, and less family stress—is enormous.

Predicting what comes next

Another area where AI is proving decisive is predicting outcomes. One of the most common—and essential—questions from children, young people, and families is about what the future might hold. This is a hard question to answer. Even detailed clinical assessments don’t reliably predict whether a child with a diagnosis of ADHD, for example, will continue to have troublesome symptoms into adolescence or will resolve.

That’s changing. In a recent project, we combined genetic and brain imaging data from children diagnosed with ADHD and trained a machine learning model to predict their future outcomes. With over 80% accuracy, we can now predict whether a child will meet diagnostic criteria in adolescence or whether symptoms will subside. That level of insight can be transformative for families and clinicians trying to make informed decisions about support and intervention. Our next critical step is to see if this predictive tool works in other settings.

Mind the gap

But with this exciting future comes a critical challenge: equity.

AI models are only as good as the data they’re trained on. And too often, that data doesn’t fully capture the children who need help the most—those from underrepresented backgrounds, or those with the most severe symptoms who can’t complete some research procedures (such as having an MRI scan).

If we’re not careful, we risk building tools that work well for some groups but leave others behind, reinforcing existing disparities in healthcare.

That’s why equity isn’t an afterthought in our research—it’s a design principle. In the new Pears Maudsley Centre, we’re investing in child-friendly, movement-tolerant brain imaging technologies that work even when kids are anxious or active. We’re also building strong partnerships with schools and communities to recruit participants who reflect the full diversity of the population. And we’re designing our models with inclusion in mind, so they’re more likely to benefit all children, not just the easiest to study.

A broader effort, a shared goal

This work is part of a broader effort across the King’s Maudsley Partnership, where researchers, clinicians, and technologists work together to improve youth mental health care. From digital tools being trialled in clinics to large-scale studies on how social media and cognitive functioning interact, we’re creating a more connected, data-informed understanding of mental health.

We’re also closely aligned with the new Pears Maudsley Centre for Children and Young People, which will bring cutting-edge research, clinical care, and community engagement under one roof. It’s precisely the kind of setting where research like ours can translate into real-world impact.