The disorder schizophrenia is a grievous mental disorder that is characterised by symptoms like hallucinations, delusional beliefs and abnormalities in perception, thoughts and behaviours. The word schizophrenia evolved from two Greek words, ‘schizo’, which means splitting and ‘phren’, which means mind. Eugen Bleuler, a Swiss psychiatrist in 1908, coined the term.  

According to the World Health Organisation, an estimated 24 million people live with Schizophrenia. It is considered one of the top 15 leading causes of disability worldwide. The average potential life expectancy for schizophrenic patients in America is over 28 years, as per the estimation of the National Institute of Mental Health.  

A recently published study in the “Proceedings of the National Academy of Sciences (PNAS) exhibits the potential of using Artificial Intelligence language models to find speech signatures in Schizophrenic patients. As per the studies, the lead author Matthew Nour, MD, PhD, University College London Queen Square Institute of Neurology, along with his colleagues Daniel C McNamee, Yunzhe Liu, and Raymond J. Dolan, wrote, “these findings shed light on the neural basis of semantic representation in schizophrenia.”  

In their new study, the researchers wrote, “Human cognition is underpinned by structured internal representations that encode relationships between entities in the world. Clinical features of schizophrenia from thought disorder to delusions are proposed to reflect disorganisation in such conceptual representations.”  

The experiment  

The study took 52 schizophrenic patients as participants, and they were given two 5-minute fluency tasks. In the category task, participants were asked to name as many that belonged to the animal category. The other task was to name as many words as possible that start with the alphabet P.   

The researchers took this opportunity to test whether an AI algorithm could predict the words the participants provided and gauge the difference in predictability compared to those with Schizophrenia. The scientists were with the hypothesis that the answers provided by those with Schizophrenia would be less predictable for the algorithm.   

The model fastText  

The research team used an open-source library named fastText, which is a pre-trained Natural Language Processing word-embedding model for creating effective sentence classification and learning word representations. FastText was developed by Facebook Artificial Intelligence Research (FAIR), a part of Meta engineering. Within five minutes, fastText can classify a half-million sentences among 300,000 categories.   

The result  

The answers provided by the control-group participants were more predictable by the AI algorithm than those produced by participants with Schizophrenia. The more critical the disorder, the greater the difference in its predictability. “At a behavioural level, patients with schizophrenia showed reduced semantically guided word sampling during a verbal fluency task,” the scientists reported.  

However, the researchers hypothesised that the mental representation of cognitive maps may elaborate their findings. To test the hypothesis, the researchers ran magnetoencephalography scans on the brain areas that are associated with learning and storing cognitive maps while patients were asked to learn the sequential relationship between eight task pictures and during a post-task, five-minute resting-state scan.   

The scientists reported, “In line with our hypothesis, the influence of semantic similarity on behaviour was reduced in schizophrenia relative to controls, predicted negative psychotic symptoms, and correlated with a MEG signature of hippocampal ripple power”. They added, “The findings bridge a gap between phenomenological and neurocomputational accounts of schizophrenia.”  

• Photo by Camila Quintero Franco on Unsplash
• https://www.psychologytoday.com/intl/blog/the-future-brain/202310/can-ai-detect-schizophrenia-from-speech