Researchers are using machine learning to recognize patterns in languages spoken by people, enabling doctors to spot early signs of Psychosis.
FREMONT, CA: According to a set of researches at Emory University in Atlanta, and Harvard University in Boston, the researchers were able to predict which individuals will develop psychosis with an accuracy of 93 percent. Psychosis is a condition where it can be challenging to differentiate between what’s real and what’s not. When a person is under such a state of mind, doctors term it as a psychotic episode. During such an event, people experience disturbing thoughts and perceptions.
Philip Wolff, a professor of psychology at Emory University, states that studies in the past had already indicated that subtle signs of future psychosis already exist in people’s language; however, machine learning (ML) is allowing them to unravel hidden aspects of those signs. Along with his colleague, the professor devised two linguistic variables—semantic density and use of words related to sound. They reached to the conclusion that low semantic density signals that a person is highly prone to develop psychosis in the future.
Machine Learning (ML) to Track Psychosis Symptoms
ML is increasingly getting adept at recognizing patterns in people’s language that even experienced doctors in this field may not notice. According to Neguine Rezaii, a fellow in the Department of Neurology at Harvard Medical School noticed that such subtleties are analogous to seeing microscopic germs with the naked eyes.
The warning signs of psychosis are visible usually during the mid to late teenage years with numerous psychosis symptoms popularly termed in the medical world as a prodromal syndrome. Almost 25 to 30 percent of those who reflect prodromal syndrome in their teens develop a psychotic illness such as schizophrenia. Doctors with appropriate training can predict the future development of psychosis with a precision of around 80 percent.
Researchers are trying to improve upon this percentage using several approaches with ML occupying an integral role in the endeavor. For instance, Prof. Wolff and his team used their ML-equipped systems and fed them with online conversations from 30,000 Reddit users. The team incorporated Word2Vec software that maps words in a way that similar meanings are close to each other in “semantic space.” They added another program to analyze the semantics. The team then fed the conversations of 40 participants in the North American Prodrome Longitudinal Study (NAPLS). Then the team compared the ML-analysis with the baseline data, and the follow-up data showed that the suspected participants went on to develop psychosis.