Machine learning helps to predict the treatment outcomes of schizophrenia
Could the diagnosis and treatment of mental health disorders one day be aided through the help of machine learning? New research from the University of Alberta is bringing us closer to that future through a study published in Molecular Psychiatry.
The research was led by Bo Cao at the U of A’s Department of Psychiatry, with the collaboration of Xiang Yang Zhang at the University of Texas Health Science Center at Houston. They used a machine-learning algorithm to examine functional magnetic resonance imaging (MRI) images of both newly diagnosed, previously untreated schizophrenia patients and healthy subjects. By measuring the connections of a brain region called the superior temporal cortex to other regions of the brain, the algorithm successfully identified patients with schizophrenia at 78% accuracy. It also predicted with 82% accuracy whether or not a patient would respond positively to a specific antipsychotic treatment named risperidone.
Approximately one in 100 people will be affected by schizophrenia at some point in their lives, a severe and disabling psychiatric disorder that comes with delusions, hallucinations, and cognitive impairments. Most patients with schizophrenia develop the symptoms early in life and will struggle with them for decades.
According to Cao, early diagnosis of schizophrenia and many mental disorders is an ongoing challenge. Coming up with the personalized treatment strategy at the first visit with a patient is also a challenge for clinicians. Current treatment of schizophrenia is still often determined by a trial-and-error style. If a drug is not working properly, the patient may suffer prolonged symptoms and side effects, and miss the best time window to get the disease controlled and treated.
Cao hopes to expand the work to include other mental illness such as major depressive and bipolar disorders. While the initial results of schizophrenia diagnosis and treatment are encouraging, Cao says that further validations on large samples will be necessary and more refinement is needed to increase accuracy before the work can be translated into a useful tool in a clinical environment.
