A team of researchers at the University of South Florida is breaking new ground in the diagnosis and treatment of post-traumatic stress disorder (PTSD) in children. Led by Alison Salloum, professor in the School of Social Work, and Shaun Canavan, associate professor in the Bellini College for Artificial Intelligence, Cybersecurity and Computing, the interdisciplinary team has developed a pioneering artificial intelligence system that analyzes facial expressions to help clinicians identify PTSD in young patients—without compromising their privacy.
Diagnosing PTSD in children is typically based on subjective methods like clinical interviews and self-reported questionnaires, both of which can be limited by a child’s developmental stage, language skills, or emotional avoidance. Salloum’s clinical experience sparked the idea for a more objective solution. “Some children’s facial expressions became incredibly intense during trauma interviews,” Salloum said. “Even when they weren’t verbalizing much, their faces revealed so much. That’s when I asked Shaun if AI could help us analyze that.”
Canavan, whose expertise lies in facial analysis and emotion recognition, adapted tools from his lab to create a privacy-first system that processes de-identified video data—tracking head movement, eye gaze, and facial landmarks without retaining raw footage or identifying information. “We don’t use raw video at all,” Canavan emphasized. “The AI analyzes only facial movement data, factoring in the context of whether the child is speaking with a clinician or a parent.”
The team’s peer-reviewed study, published in Science Direct, is the first to use context-aware PTSD classification technology while fully preserving participant anonymity. Using data from 18 therapy sessions with children, the researchers analyzed over 100 minutes of footage per participant, totaling hundreds of thousands of video frames. The AI system identified subtle yet consistent facial muscle movements associated with PTSD, especially during clinician-led interviews, which proved more revealing than parent-child conversations.
“This isn’t about replacing clinicians,” Salloum explained. “It’s about giving them an extra layer of insight. We envision this system as a real-time tool during therapy sessions, helping monitor a child’s progress without subjecting them to repeated, potentially traumatic evaluations.”
The research team plans to expand their study to examine potential biases across gender, culture, and age—particularly among preschool-aged children, where verbal communication is limited and diagnosis often relies on parent observations.
While still in the early stages, the project has already demonstrated promise, particularly given the complex clinical profiles of many participants, who also presented with co-occurring conditions like depression, ADHD, or anxiety.
“High-quality data like this is rare in AI,” Canavan said. “We’re proud to have designed a study that was not only innovative but ethically sound. This tool could ultimately help clinicians make better-informed decisions and improve mental health care for some of the most vulnerable populations.”
If validated in larger-scale trials, the USF team’s system could mark a transformative shift in how PTSD is diagnosed and tracked in children—leveraging AI and video technology to bring greater accuracy and empathy into pediatric mental health care.
