AI and computational psychiatry: Finding ‘hidden cues’

AI-and-computational-psychiatry-Finding-hidden-cues.jpg
(Image: Getty/metamorworks) (Getty Images/iStockphoto)

The pharmaceutical industry has increasingly been adopting artificial intelligence – which has demonstrated the ability to predict instances of schizophrenia with 74% accuracy.

There are many ways in which researchers are using artificial intelligence (AI), said Guillermo Cecchi, principal research staff member, manager of computational psychiatry at IBM, who spoke at AAPS PharmSci 360.

“We are using AI also to find hidden cues that may escape clinicians due to various factors, including their complexity (e.g. a combination of several disparate features) or the impossibility to even detect them (e.g. high frequency features of the voice),” Cecchi told us.

Cecchi’s research in computational psychiatry – defined by the National Institute of Mental Health as way to identify and validate biomarkers and treatment targets related to psychiatric disorders – has demonstrated AI’s ability to predict the likelihood of a previously-unseen patient having schizophrenia.

According to a study conducted in collaboration by IBM and researchers at the University of Alberta, AI and machine learning algorithms can predict instances of schizophrenia with 74% accuracy.

Schizophrenia is an area in which machine-based research has proved to be useful, as the condition leaves a strong trace in language.

“The value of using AI for language is that we can 'standardize' the diagnosis of different mental health conditions, as we can precise the features that define them – as far as they are reflected in language,” explained Cecchi.

However, schizophrenia is not the only mental illness in which AI and machine learning can be applied. Cecchi told us that other psychiatric conditions, such as bipolar disorder, borderline personality disorder, depression, addiction, anxiety disorders, as well as neurologic conditions like Parkinson’s and Alzheimer’s can use AI to make further assessments.

Cecchi said, “I think that the most significant insight we have gained is that it is possible to predict with high accuracy the onset of psychosis in an at-risk population. While we still need to conduct further and more extensive validation studies, the initial studies make us believe we have found a real and robust signal.”

“This has many implications for early intervention and for monitoring of treatment outcomes, including clinical trials,” he added.

Information technology and is also making it possible to complete remote-monitoring of patients at high frequency, explained Cecchi.