Researchers at the University Medical Center Groningen (UMCG) use artificial intelligence (AI) to spot subtle changes in everyday speech that may reveal early signs of psychosis. Their research, featured by science journalist Maya Schutte on NEMO Kennislink, shows how language and AI can work together to improve mental health care, including research linked to the TRUSTING project.
Subtle language changes reveal early warning signs
Psychiatrist and linguist Dr. Janna de Boer and neuroscientist Dr. Sanne Schuite-Koops from UMCG found that people with psychosis often show small but meaningful changes in their speech. They may use shorter sentences, make unusual word associations, or speak in a disorganised way.
Although these patterns are difficult for humans to notice, AI models can detect them with higher precision.
By analysing short, natural conversations, the researchers trained an AI system to tell the difference between people with and without psychosis. The model reached an accuracy rate of 85%. It looked at pauses, grammatical complexity, and sentence structure, small details that can signal early cognitive disorganisation.
From language research to digital monitoring
After these promising results, the UMCG team began developing a mobile app that uses weekly speech samples to track changes over time. The app could help clinicians monitor relapse risk and act sooner if needed. Importantly, all decisions remain in the hands of healthcare professionals.
“Normally, psychiatrists see patients every few months. With this app, we can respond much earlier if someone shows signs of relapse,” said Dr. Sanne Schuite-Koops.
The app is now being tested as part of a European research collaboration. It must meet strict standards under the EU AI Act, covering privacy, transparency, and data protection. The researchers stress that AI supports clinical care, it does not replace it. Patients always remain at the centre of the process.
Broader potential for mental health
According to Dr. Janna de Boer, language-based AI could also help detect depression, anxiety, or dementia. Because it analyses natural speech, this approach offers a non-invasive and accessible way to support early diagnosis and intervention.
This work reflects the values of the TRUSTING project, which promotes transparent, responsible, and trustworthy AI in healthcare. By combining linguistic science, data ethics, and advanced technology, researchers are paving the way for safer and more supportive digital health solutions.
Project reference and disclaimer
This news item refers to research by the University Medical Center Groningen (UMCG), covered in NEMO Kennislink (August 2025) by Maya Schutte, and includes work connected to the TRUSTING project.
The TRUSTING project receives funding from the European Union’s Horizon Europe programme. The views expressed are those of the authors and do not necessarily represent those of the European Union or the European Commission.
👉 Read the original article in Dutch here: https://www.nemokennislink.nl/publicaties/psychose-herkennen-via-taal/
For more updates, insights, and news about TRUSTING’s progress, visit the TRUSTING Project website or follow us on social media:
Sign up for the yearly newsletter: shorturl.at/SvxFN
