杏吧原创

Artificial intelligence could spot early signs of schizophrenia

Years before schizophrenia can be diagnosed by doctors, artificial intelligence may be able to detect early signs of the condition in people's speech

A picture of a mouth

People in the early stages of schizophrenia have telltale verbal signs that could be spotted by speech analysis software, long before doctors notice. Those who are developing the condition tend to be vaguer in their speech and to use words related to voices and sounds, a study has found.

People with schizophrenia have delusions and hallucinations, such as hearing voices, as well as confusion and memory problems. The condition usually emerges when people are in their twenties, but for a few years beforehand, people may have warning symptoms, like odd behaviours or seeing or hearing unusual things. But only about a third of people with these mild symptoms go on to develop full-blown schizophrenia or a similar disorder.

One hope is that people who are most at risk could be identified during this preliminary stage and given help, such as talking therapies or even just practical assistance; during this period people are more likely to drop out of college, or lose their job, for instance. 鈥淜nowing what鈥檚 going on by itself would help a lot, and they could get connected to support systems,鈥 says Neguine Rezaii of Harvard Medical School.

Rezaii鈥檚 team took advantage of a that tracked the experiences of people with mild symptoms. They transcribed recorded interviews between medical staff and 40 participants, 12 of whom later transitioned into schizophrenia. They used three-quarters of the individuals to train their artificially intelligent software to predict who would deterioriate, and the rest as a test group; in the test the program was 90 per cent accurate.

One of the indicators for schizophrenia was that people鈥檚 language was less detailed. They might say 鈥渢hey did that to me鈥 rather than stating what happened and who did it, says Rezaii.

Those who transitioned were also more likely to mention voices or sounds 鈥 not describing them as hallucinations, but just ambient noises, such as the noise of a plane engine. This might be the start of auditory hallucinations, says Rezaii.

She hopes that in future, the software could be used in real-time as doctors assess people or even to scan social media posts if people give consent.

But first the software needs to be tested on people who are at risk of other mental illnesses, such as depression, to check it is specifically identifiying schizophrenia risk.

npj Schizophrenia