Using AI and Machine Learning to Improve Mental Health Care

Over the last few years, Patrick Romani, PhD, BCBA-D, began noticing a troubling trend in mental health. Kids with neurodevelopmental disorders, such as autism, were being admitted to Children’s Hospital Colorado’s neuropsychiatric special care inpatient unit with high-acuity cases and increasingly severe behaviors. Often, their families would bring them to the emergency department (ED) in crisis, seeking a lifeline.

Current care standards see these kids admitted from the ED to inpatient care, which can be stressful for both the child and their family members.

“I think that a lot of families are going without support, and so they try to maintain at home, and eventually they’re not able to any longer,” Dr. Romani says. “That’s really where we’re trying to intervene — with those families that are hanging on.”

To do this, Dr. Romani reached out to Sidney D’Mello, PhD, and Bobby Moulder, colleagues at the University of Colorado Boulder who focus on artificial intelligence (AI) and machine learning. Working together, their goal is to map the subtle physical signs that a child might be about to engage in severe behaviors, such as hitting, biting and kicking themselves or others.

“We are using biosensors to get a sense of kids’ motion — the way they move their arms, their torsos, their bodies, their heart rate, as well as even facial expression changes or things like that,” he explains. “We’re relating that to when they exhibit severe behavior here in the hospital. We are collecting all this biological data to then be able to develop a model that predicts when these things will happen.”

The team is using a tool that takes videos of kids who are working with staff and color codes each body part so that researchers can see how they move in relation to each other and what movements might be precursors to behavioral events.

While Dr. Romani is currently analyzing data from 61 children and plans to develop further studies, the goal is to one day create an app that utilizes a database of information to predict when a child might act out with precision, giving parents time to use evidence-based strategies to intervene. One day, the team hopes that using tools like this, families can avoid the need for emergency mental health care altogether.