News

Machines could be more sensitive than humans in mental health diagnosis


Neelima Choahan


27/09/2018 3:18:44 PM

Artificial intelligence, or AI, could be a useful tool for predicting how people at high risk of psychosis or with recent-onset depression will function socially in the future, new research has found.

Professor Stephen Wood says the study found clinicians were specific, but not necessarily sensitive, in their diagnosis of which patients will have a poor social functioning outcome.
Professor Stephen Wood says the study found clinicians were specific, but not necessarily sensitive, in their diagnosis of which patients will have a poor social functioning outcome.

Machines may be more sensitive than humans when it comes to predicting how well people with mental illness will function socially in the future, according to new research.
 
Published in JAMA Psychiatry, the study saw Australian and international researchers examine the outcomes for those at high risk of psychosis or with recent-onset depression.
 
Professor Stephen Wood, from Orygen, the National Centre of Excellence in Youth Mental Health, said researchers compared a machine algorithm and mathematical modelling with expert clinical opinion on what a patient’s social and occupational functioning would be 12 months after treatment.
 
The machine algorithm analysed clinical data collected at presentation, as well as brain imaging. It also looked at clinical data and brain imaging together to predict the outcome for the patient.
 
‘[The] machine algorithm … tries to separate the group on the basis of the data that we have available to us,’ Professor Wood told newsGP.
 
‘So we were able to see whether our machine learning algorithms could do better than our clinician could.’
 
The researchers found the artificial intelligence – or machine learning – outperformed human experts. It could correctly predict social outcomes one year later in up to 83% of patients in clinical high-risk states for psychosis and 70% of patients with recent-onset depression.
 
‘When [the experts] said someone was going to have a poor outcome they were usually right, but they missed lots of people who actually did have a poor outcome,’ Professor Wood said.
 
‘So, in other words, they were specific, but they weren’t very sensitive.
 
‘Our algorithm actually did much better in both of those measures; it was more sensitive and more specific.’
 
Professor Wood also said the algorithm produced better results when the clinical data was combined with neuroimaging. 
 
For the study, which started in 2012, the research team followed 116 people at high risk of developing psychosis and 120 people experiencing recent-onset depression aged 15–40, as well as 176 control participants.
 
Depending on funding, the next stage of study will involve specific clinical trials where people would be stratified into good and bad outcomes. More intensive therapy could then be directed at those patients with a poor outcome.
 
‘Predicting social outcomes is important, as among young people and emerging adults in OECD [Organisation for Economic Co-operation and Development] countries the top causes of “disability” – and poor social functioning is included in that – are mostly disorders of mental health, including those that typically present with a first episode of psychosis,’ Professor Wood said.
 
‘By being able to better predict what will happen to people at high risk of psychosis or with recent-onset depression over time, we are able to provide individualised treatments to clients when they first present to mental health services and potentially improve their social functioning.’



Artificial Intelligence JAMA Psychiatry machine learning Orygen





Comments



 Security code