Researchers (L-R) Russ Greiner, PhD, Sunil Kalmady Vasu, PhD, and Andrew Greenshaw, PhD (Photo: University of Alberta Faculty of Medicine & Dentistry)
A machine learning algorithm that previously predicted people with schizophrenia with 87% accuracy has now identified healthy first-degree relatives of such patients with the highest self-reported schizotypal personality scores. Researchers reported their findings in NPJ Schizophrenia.
“Our evidence-based tool looks at the neural signature in the brain, with the potential to be more accurate than diagnosis by the subjective assessment of symptoms alone,” said lead author Sunil Kalmady Vasu, PhD, senior machine learning specialist in the Faculty of Medicine & Dentistry at the University of Alberta in Edmonton, Alberta, Canada.
In the latest study, the EMPaSchiz (Ensemble algorithm with Multiple Parcellations for Schizophrenia prediction) tool analyzed resting-state functional magnetic resonance images of 57 healthy siblings or children of patients with schizophrenia. The algorithm accurately identified the 14 participants who scored highest on a self-reported schizotypal personality trait scale without meeting the full criteria for a schizophrenia diagnosis, researchers reported.
Elevating the Treatment of Bipolar Disorder and Schizophrenia: Examining the Role of LAIs
EMPaSchiz is designed to work as a decision support tool and not to replace diagnosis by a psychiatrist.
“The goal is for the tool to help with early diagnosis, to study the disease process of schizophrenia and to help identify symptom clusters,” said Dr. Kalmady Vasu, a member of the Alberta Machine Intelligence Institute.
Researchers plan to follow study participants over time to track whether they eventually develop schizophrenia. They also intend to test the artificial intelligence tool on nonrelatives with schizotypal traits.
“This study, for the first time, demonstrates a cross-application of a machine-learned schizophrenia diagnostic model in identifying subjects with high levels of negative schizotypy. However, whether similar prediction performance holds for a larger population without familial association remains to be explored,” researchers wrote. “Further application of this approach holds significant promise for exploring related and comorbid symptom clusters in psychiatry.”
Kalmady SV, Paul AK, Greiner R, et al. Extending schizophrenia diagnostic model to predict schizotypy in first-degree relatives. NPJ Schizophrenia. 2020;6(1):30.
Rutherford G. Machine learning tool used to predict early symptoms of schizophrenia in relatives of patients [press release]. Edmonton, Alberta, Canada: University of Alberta Faculty of Medicine & Dentistry; January 26, 2021.