University of South Florida researchers are using artificial intelligence to identify signs of post-traumatic stress disorder in children.
It's challenging to recognize signs of PTSD in children because it can be hard for them to share their stories, said Alison Salloum, a professor in the USF School of Social Work, who is leading the study along with Shaun Canavan, an associate professor in the Bellini College for Artificial Intelligence, Cybersecurity and Computing.
One of the main symptoms of PTSD is avoidance, which is not wanting to talk about what happened, Salloum said.
She said parents will come in and say that their child has changed — they could be getting in fights with friends and siblings, having angry outbursts, not expressing positive feelings, and pushing people away.
When conducting online clinical trials to find psychosocial treatments for children, Salloum began to notice their expressions as they discussed their experiences.
“It can be very distressing for a child to have to talk about what happened,” she said. “So to be able to have a tool that we could use to recognize facial expressions would be incredibly helpful.”
ALSO READ: FIU researchers use AI to detect heart disease early with '95% accuracy'
Trial involves younger children and adolescents
She came up with the idea of using AI and asked Canavan for help. He has experience with research on facial expressions in a wide range of applications, including in the medical domain.
“It was a really nice fit for the work that I've already been doing, and to move forward with this workflow,” he said.
When the AI model was ready for trial, the team analyzed 18 children. Two of them were between age 7 and 11, and 16 were between 13 and 17.
“We did have more adolescents than younger children,” Salloum said. “They are obviously much more engaged in technology, and so they really liked the idea of using AI.”
Canavan said the AI system looks for muscle movements on people’s faces — whether they are smiling, neutral, looking at a parent or therapist, or looking down.
“So the AI does comparisons between the teenagers or the children that have PTSD and those that don't, and it finds the differences between those two groups,” he said.
Privacy important with AI tools
A crucial aspect of the study was the deidentification of the children and teenagers.
The AI system converted the videos into something that was no longer readable by humans, Canavan said.
“You can look at them, but it looks like noise,” he said. “It doesn't look like anything that a human would say is a person.”
It is essential to protect children's privacy when developing AI tools. So, when Canavan explained his idea to Salloum, she was all in.
“We certainly want to protect children's privacy and make sure that the videos are not being used in any other way other than to detect PTSD or not,” Salloum said.
The AI model is only a pilot study within a research setting and not yet ready for clinical use, Canavan said.
Still, he said the goal is to refine the model and make an actual system or app that clinicians can use.
Although the study initially focused on older children and teens, Salloum said she would be interested in adapting it for use with preschoolers.
“If we can use this as another tool, and they're not having to constantly be assessed verbally, I think it would be really useful for the field,” she said.
