People with psychosis have a certain way of thinking. Indications of a different thought pattern can be recognized in their speech. What if we could capture that way of speaking in a AI model, allowing us to predict that someone will develop psychosis?

Psychiatrist Iris Sommer is going to investigate this together with a European team of brain scientists, linguists and experts in artificial intelligence (AI). For this research, the team received a Synergy grant of almost 10 million euros from the European Research Council.  

Almost 22 million people in Europe have been diagnosed with psychosis. People who have already had a psychosis once, are more likely to suffer another one later in life. 'That is why they are often given long-term antipsychotics, for months, sometimes years,' says Sommer. This reduces the chance of a relapse, but also comes with disadvantages.  ‘Many patients don't want to use long-term medication anymore to prevent a relapse. Therefore, we only want to give people medication when the symptoms seem to return. We call this targeted treatment. But therefore we need good predictors of a relapse.'  

Speech reflects thoughts  

To recognise the symptoms of psychosis early, Sommer is going to study the speech of people with psychosis. 'We think that what and how someone says something can be a good reflection of their thoughts. People with psychosis have a certain way of speaking: they tell a story with many pauses, are easily distracted and often talk about the same subject. We are now going to investigate whether speech does indeed correspond to someone's thoughts and whether we can predict if someone develops psychosis based on the way they speak.'  

Recognising alterations in speech with app   

Using an app that records speech, Sommer's team is going to develop software that recognises alterations in people's speech. They are investigating this in people who already had a psychosis. 'In the app, we show people a picture or a story every week and they record their reaction to it. We will analyse that reaction and see if it has certain characteristics.  With the speech recordings, we train a personal language model with artificial intelligence.'   

Measuring brain activity during speech alterations  

In this study, the researchers also use imaging techniques to measure the brain activity of people with psychosis.  In this way, they are looking at whether there is a link between speech alterations and brain activity in psychosis.   

Alteration in speech as a predictor for psychosis 

How speech alters in people with psychosis depends on the severity of the illness. Psychotic disorders are classified into different phases: acute, remission and relapse. Therefore, Sommer and her team are investigating speech alterations in different phases of psychosis. 'We look at one person's speech alterations during a certain time. If the research is successful, it could become possible to identify a relapse risk at an earlier stage. This would give physicians more time to adjust a patient’s treatment and in the best case, to prevent them from relapsing.'  

Method to use speech as a predictor to be adopted more extensively  

If it turns out that speech alterations are a good predictor of psychosis, the researchers want to apply this method more widely. 'Speech provides a wealth of data to identify the mental well-being of the speaker,' says Sommer. 'Also within other medical disciplines and even outside medicine, think of driving trains, fly planes or operating large devices, it is of great value to be able to reliably predict someone's state of mind.'  

The UMC Groningen coordinates the project, and Iris Sommer, professor of psychiatry, is the principal investigator. The other principal investigators are Prof Wolfram Hinzen, linguist from Barcelona University, Prof Philipp Homan, Zurich-based brain researcher and Prof Brita Elvevåg, expert in computer technology at Tromsö University. 

Prof. dr. Iris Sommer