IBM previously attempted this study in 2015 with limited results. Back then, they had asked patients to talk about themselves for close to an hour, and then analysed the results of those interviews using the Natural Language Processing (NLP) technique. The model, however, had several limitations as IBM mentioned in their blog post: only one cohort, from a single location, and using a single evaluation protocol.
This time around, the research team used a different approach for creating the model. To get data for the research, the team analysed the speech patterns of 59 people. Patients were given a story to read and were then asked to talk about what they understood from it. Following that, the transcripts were broken down into different parts and a score was assigned based on how coherent the sentences were.
The model was able to predict which patients would develop psychosis with an impressive 83% accuracy. In other words, 19 participants developed a psychotic disorder within 2 years on the study, while 40 did not. The research found that those at a huge risk of developing the illness used far less coherent sentences than healthy people.
IBM is also currently researching and developing models for other mental health problems like depression, Parkinson’s disease, Alzheimer’s and chronic pain. Guillermo Cecchi, the author of the post and this study, has also posted the below video:
Our take on this
This is a huge step forward in neuropsychiatric assessments. It shows how AI can be an extremely effective tool to assist mental health professionals both inside and outside the hospital. If the risk of psychosis can be predicted years in advance, the patient can be given treatment well ahead of time to prevent the onset of the illness.
Also, currently the feeling is that as far as mental health issues go, homeless and poor people struggle to get any sort of medical treatment for it. The hope is that AI will enable these folks to get an equal chance of getting help.You can also read this article on our Mobile APP