MePheSTO - using AI to detect psychiatric disorders

Changed on 02/11/2023
How can video, voice, language and speech analysis be used to better evaluate psychiatric disorders, or even to detect them earlier? This is a question that the researchers working within MePheSTO, a research project launched within the framework of a partnership between Inria and the DFKI (the German Research Centre for Artificial Intelligence), are looking to answer. The concept is to draw upon artificial intelligence in order to identify and classify objective - and therefore measurable - digital phenotypes of these conditions.
© Unsplash / Photo Osman Rana


The project had been germinating for a number of years in the minds of the scientists involved in the MePheSTO project. It was while working with patients suffering from Alzheimer’s in Nice that some of the researchers began to notice that the subjective clinical impressions formed by clinicians through conversations with patients were not enough to detect early signs of disorders, to make precise diagnoses or to predict relapses. The researchers from Nancy, who were working at the same time on clinical interviews conducted with patients with schizophrenia, arrived at the same conclusion.

Interpersonal communication is based on language and non-verbal behaviour (expressions, movements of the head, gestures, etc.). But there is currently no multimodal corpus comprised of video and speech recordings. MePheSTO is seeking to address this, and are also planning to add biosensor measurements.  His goal is to use all of this information to decode and analyse social interactions between patients and clinicians.

Improving diagnoses, supporting treatment

These artificial intelligence solutions will lead to the development of tools and techniques that will be of use to the medical profession, including demonstration scenarios summarising patient pathways.

© Inria / Photo H. Raguet

This will improve the analysis made by clinicians by enabling them to detect any changes, however minor, in a patient's mood or how they behave or speak. This diagnostic assistance could eventually be provided in real-time, and even remotely for telemedicine appointments, which could meet new needs in the not too distant future.

The sensitive nature of both the subject and the data itself entails strict respect for patients and the information they share with researchers (GDPR, ELSI).

The project, which is still only in its infancy, is set to begin officially in October 2021, and will run for three years. In the meantime, the researchers will work on finalising recruitment and preparing for the launch. 

We are drafting a protocol with our clinical partners in order to begin gathering data in 2021, both in France and in Germany. It is essential during this stage to work closely with clinicians

                         Alexandra König, researcher and member of the project.

Opening up new possibilities

The individuals involved in the project already have multiple data sets that were compiled during previous studies, which they will use to prepare the analysis methodology. The long-term goal is to understand all types of psychiatric and neurodegenerative conditions and to keep indices that are both up-to-date and as objective as possible on how these disorders develop and how effectively they are treated.