Could artificial intelligence (AI) be a useful tool in the diagnosis of schizophrenia?

A new application of cognitive neuroscience has emerged. AI chatbots can be trained to detect subtle changes to speech in those diagnosed with schizophrenia.

Approximately 1% of us will develop schizophrenia during our lifetime. In the UK this equates to approximately 685,000 people. Schizophrenia is one of several mental illnesses called ‘psychoses’ which involve a person losing touch with reality. For example, a person may experience visual or auditory hallucinations.

Early diagnosis of schizophrenia by psychiatrists can be very difficult and often takes many weeks or months. This, coupled with the fact that many people take years to seek medical help means that patients often go for long periods of time without the help that a diagnosis of schizophrenia may bring.

How do psychiatrists diagnose schizophrenia?

Typically, diagnosis of schizophrenia is based on a patients’ own account, for example, what symptoms they are experiencing and for how long. Patients can often be unaware of the extent of their difficulties (referred to as lacking ‘insight’) so friends and family may also have an input into proving the psychiatrist with as much detail as possible.

How can AI help?

Many people with schizophrenia experience something called conceptual disorganisation, a sign of disorganised thought processes. This disorganisation may manifest itself in symptoms such as speech derailment which is where, in conversation, people with schizophrenia may jump from one idea to another unrelated idea. Whilst the cause of this is not understood, it is this conceptual disorganisation which gives us an insight into how the mind is structured and it is here where we can utilise the power of AI language models.

In research by Nour et al (2023) 26 patients with schizophrenia and 26 control participants completed verbal fluency tasks. For example, people were asked to name as many animals as they could in five minutes.

Participants’ speech was recorded and analysed using an AI language model (a predecessor of the well-known AI chatbot Chat GPT) which had been trained to represent the meaning of words in a similar way to humans.

The AI language model allowed for analysis of the semantic link between words e.g. cat and dog are closely related, cat and giraffe less so. The model was able to detect subtle differences in the speech of people who had a diagnosis of schizophrenia and those who did not, for example, the order in which the participants with schizophrenia presented their words was much more difficult for the language model to predict, in comparison to the control participants. Furthermore, the model could also predict the severity of symptoms based on the degree of conceptual disorganisation.

Researchers believe that AI language models could be used to help early diagnosis of schizophrenia when subtle differences first become evident in a patient’s speech, enabling thousands of people in the future to obtain valuable help, earlier. If further trials prove successful, the use of AI language models could be commonplace in clinical practice in the next decade.

Questions you might like to consider following this blog:

• How would you implement the findings of this study in the real world?
• What benefit could such findings have on the economy of the UK?
• Which inferential test would be appropriate to analyse the significance of Naur et al’s (2023) study?

Read: Royal College of Psychiatrists (2015). Schizophrenia. https://www.rcpsych.ac.uk/ment...

Read: AI language models could help diagnose schizophrenia https://www.ucl.ac.uk/news/202...