Using AI for early psychosis detection: ‘You can check whether you’re at risk of a relapse’

Recognising schizophrenia on time is still fairly difficult, even for psychiatrists. But physician and linguist Janna de Boer hopes to spot schizophrenia in patients by analysing their speech patterns. For that, she’ll need the help of AI.  

People talk every day. To their colleagues, their friends, their family. They tell their stories, speaking word after word. Without so much as a thought, they speed up, wait for a beat, or lower their voice. This not only makes their story more interesting, but also conveys how they’re feeling. People do all this without realising it.

Wouldn’t it be possible to use language to discern psychological disorders? Perhaps the way patients talk might help psychiatrists recognise difficult to diagnose disorders such as depression, dementia, or schizophrenia. Janna de Boer, PhD student of psychiatry and linguistics, thinks this could work. ‘Psychiatric disorders have an immense impact on your voice, influencing how you talk or what you say.’ 

Difficult to diagnose

According to the Trimbos Institute, nearly 1 percent of people in the Netherlands suffer from schizophrenia. It’s an incurable psychiatric disorder that tends to develop, quite inconspicuously, in puberty. As the disease progresses, people start to experience strange and illogical thoughts. Patients become unable to separate their delusions from reality, which can eventually lead to psychosis. 

Patients can be helped tremendously by a proper diagnosis, but those are not easy to make. That’s because psychoses are comorbid with many different types of mental disorders. They might be a sign of schizophrenia, but they could also be caused by depression or drug use. 

The symptoms also vary greatly. ‘Some people have symptoms of psychosis that don’t impact their lives at all’, De Boer explains. ‘But there are also people suffering from psychoses non-stop, where it gets in the way of living a normal life.’ 

Speech anomalies

Currently, diagnoses are made by cataloguing some of the symptoms that patients have and placing them on a spectrum. These symptoms serve as a guideline for psychiatrists, but they’re not an objective benchmark. ‘We can’t test people’s blood to find it, we can’t put them in an MRI machine, and we can’t do a PET scan.’ 

But De Boer hopes that paying attention to people’s speech patterns might change the diagnostic game. People with schizophrenia tend to talk more slowly. They’re not as animated and use short sentences. ‘But during a psychosis, patients might talk all about their hallucinations, or become distracted and miss part of the conversation’, she explains. 

According to De Boer, it’s difficult for psychiatrists to recognise these speech pattern anomalies, since they’re focused on learning how to understand their patients during a meeting. ‘I think psychiatrists do tend to base their diagnosis on speech and language, but that they’re not always aware they’re doing it’, she says. ‘They might feel like there’s something different about the way someone speaks, but it can be difficult to say exactly what.’ 

AI programmes

De Boer uses speech analysis and AI programmes that are similar to the currently popular ChatGPT. She analyses patients’ speech and language in an effort to learn to recognise the symptoms of a psychosis. ‘We studied how people speak, which includes intonation, tonal variations, and the speed at which they speak, as well what they say and what kind of constructions they use.’

It turned out to work surprisingly well. ‘Not only were we able to see that someone had symptoms, but we could even identify which exact ones.’ 

This means AI is able to distinguish people without psychoses from people with schizophrenia, who suffer from chronic or returning psychoses. ‘I think this can be a real asset, because it can do things that psychiatrists simply aren’t very good at.’

De Boer also hopes to use the same AI programme to predict whether someone is likely to experience another psychosis. ‘People who’ve had a psychosis once have a tendency to relapse, but you can never know when that’ll happen’, she explains. ‘But if you can use speech analysis to see a relapse coming, you might be able to start treatment before someone becomes truly psychotic.’


She also thinks her method will benefit the patients. ‘If this method truly works, we could turn it into an app for their phone. They could then monitor their speech at home and check whether they’re at risk of a serious psychotic relapse. It would make the whole monitoring process much more accessible’, she says, hopeful.

For now, the AI program will need more training before it can be used in practice by psychiatrists. ‘But AI is growing incredibly quickly and is getting increasingly better at understanding human language’, De Boer says. ‘That’s how we’re suddenly able to have all these different applications for it.’


Notify of

De spelregels voor reageren: blijf on topic, geen herhalingen, geen URLs, geen haatspraak en beledigingen. / The rules for commenting: stay on topic, don't repeat yourself, no URLs, no hate speech or insults.


0 Reacties
Inline Feedbacks
View all comments