Diagnosing post-traumatic stress disorder in children is notoriously difficult. Many people, especially those with limited communication skills and emotional awareness, have a hard time explaining what they are feeling. Researchers at the University of South Florida are working to address these gaps and improve patient outcomes by combining childhood trauma with artificial intelligence expertise.
A multidisciplinary team led by USF Social Work School professor Alison Salloum and Shaun Canavan, an associate professor at Bellini University of Artificial Intelligence, Cybersecurity and Computing, are building a system that can provide clinicians with objective, cost-effective tools to help them track the recovery of children and adolescents.
Research published in Pattern recognition charactersthe first type that incorporates context-conscious PTSD classification while fully preserving participants' privacy.
Traditionally, the diagnosis of PTSD in children relies on subjective clinical interviews and self-report questionnaires. This can be limited by cognitive development, language skills, avoidance behavior, or emotional suppression.
This really started when I realized how intense the child's expression had become during the trauma interview. Even when they didn't say much, you could see what they were going through in their faces. That's when I told Sean about whether AI would help detect it in a structured way. ”
Alison Salloum, Professor at USF School of Social Work
Specializing in facial analysis and emotional recognition, Canavan reuses existing tools in his lab to build a new system that prioritizes patient privacy. The technology removes identification and analyzes only identified data, such as head poses, gazes, and facial landmarks such as eyes and mouth.
“That's what makes our approach unique,” Canavan said. “We don't use live videos. We completely remove subject identification, keep only data on facial movements, and consider whether the child is talking to a parent or clinician.”
The team created a dataset from 18 sessions with the child, sharing emotional experiences. With each video containing over 100 minutes of videos and approximately 185,000 frames per child, Canavan's AI model extracted various subtle facial muscle movements associated with emotional expression.
The findings revealed that different patterns are detectable in facial movements in children with PTSD. The researchers also found that facial expressions during clinician-led interviews were more evident than parent-child conversations. This is consistent with existing psychological research showing that children may be more emotionally expressive towards therapist, and may avoid sharing distress with parents due to shame and their cognitive abilities.
“That's where AI can provide valuable supplements,” Salloum said. “It strengthens the tools rather than replacing clinicians. Ultimately, the system can be used to provide real-time feedback to practitioners during treatment sessions, helping to monitor progress without repeating repeated, potentially painful interviews.”
The team hopes to expand their research to further explore potential biases from gender, culture, age, and especially preschoolers. There, verbal communication is limited and diagnosis is almost entirely dependent on parental observation.
Although this research is still in its early stages, Salloum and Canavan feel that potential applications are widespread. Many current participants had complex clinical photographs that reflected co-occurrence conditions such as depression, ADHD, anxiety, and real-world cases, and provided promises for system accuracy.
“This kind of data is extremely rare in AI systems and I am proud to have conducted such ethically sound research. It is important when working with vulnerable subjects,” Canavan said. “Now, we can provide clinicians with information and objective insights from this software.”
When validated in large trials, USF's approach can redefine how PTSD in children is diagnosed and tracked to bring mental health care in the future using routine tools such as video and AI.
sauce:
University of South Florida
Journal Reference:
Aathreya, S. , et al. (2025). Multimodal, context-based dataset for children with post-traumatic stress disorder. Pattern recognition characters. doi.org/10.1016/j.patrec.2025.05.003.
