ADVERTISING
International study shows ways for privacy-conscious AI models in the diagnosis and treatment of mental illnesses
Researchers from TU Darmstadt and the Indian Institute of Technology (IIT) Delhi have published a groundbreaking study on data security in artificial intelligence for mental health . Led by computer science professor Iryna Gurevych at the Ubiquitous Knowledge Processing (UKP) Lab, the study, published in the journal Nature Computational Science, a roadmap for AI systems that can support the diagnosis and treatment of mental illnesses – without disclosing sensitive patient data.
AI to support diagnosis and therapy
Mental illnesses are among the most common causes of health problems worldwide. Often, there is a lack of qualified professionals to ensure early treatment. AI systems could therapists by subtle signals in speech, facial expressions, or word choice – an important step towards early detection and improved care.
However, the confidentiality of patient data remains the biggest challenge: voice and video data from therapy sessions are highly sensitive and can reveal the identity of those affected.
Data protection-oriented development approaches
The research team demonstrates how AI systems designed with data protection in mind . This includes combining different methods:
- Removal of personal information
- Anonymization of voices and faces
- Use of synthetic data
- data protection-oriented model training
These steps are intended to ensure that AI applications in healthcare can be used effectively, responsibly and ethically in the future
International cooperation and promotion
First author Aishik Mandal is part of the NLPsych group researches
the intersection of natural language processing (NLP) Co-authors include Professor Tanmoy Chakraborty from IIT Delhi, who was supported by a Humboldt Research Fellowship , and Professor Iryna Gurevych , head of the UKP Lab.
The work was funded by the Federal Ministry of Education and Research (BMBF) , the Hessian Ministry of Science and Art (HMWK) within the framework of the National Research Center for Applied Cybersecurity ATHENE , as well as by the LOEWE Center DYNAMIC and the LOEWE top professorship “Ubiquitous Knowledge Processing” .
The full study can be accessed via the following link:
👉 Nature Computational Science – “Towards privacy-aware mental health AI models”
(DARMSTADT – RED/TU)
