Introduction
Currently, psychotic disorders such as schizophrenia are diagnosed solely through clinical assessments of patients’ symptoms, and no automated, quantitative metrics are available to clinicians. Given that altered speech is often a core symptom of psychotic disorders, this project aims to assess whether computer-generated markers of patients’ speech can be used to detect their disease status. Ultimately, quantitative speech markers of psychosis could revolutionise healthcare of psychotic disorders, since it is non-invasive and cheap to collect a patient's speech, and could be recorded repeatedly over time to track individual patients’ disease trajectories.
Explaining the science
This work will characterise patients’ speech using two main data science approaches: state-of-the-art network science methods and natural language processing techniques.
Network science is a relatively young field, which aims to study systems as networks. This project will study speech as a network where the nodes are words, and directed edges are placed between words in the order in which they appear in the speech. Recent work has suggested that speech graphs constructed in this way from schizophrenia patients’ speech are less connected than those from healthy volunteers. The research will assess whether new, advanced network measures can shed light on more subtle alterations in patients’ speech connectivity patterns.
Natural language processing (NLP) also provides a wide range of computational techniques to analyse natural speech. For example, NLP can be used to study how patients’ speech is altered at the syntactic level (in terms of the grammatical arrangements of words) or at the semantic level (in terms of the meaning of speech). Prior work has suggested that either (or both) of these features of speech may be altered in some patients.
Project aims
This project aims to assess which of the existing methods for characterising individuals’ speech is most likely to be predictive of psychosis in subjects at clinical high risk of psychotic disorders. The project will also develop new data science approaches to measure patients’ speech, and investigate whether these new methods provide extra power to detect psychotic disorders.
Ultimately, the project will build a more complete, quantitative understanding of how patients’ speech is altered in psychotic disorders, and help inform how speech data should be collected in future research studies. The project will also build new links between natural language processing researchers working at The Alan Turing Institute and researchers based in the Psychiatry departments at Cambridge University and King’s College London, promoting inter-disciplinary approaches to the challenges of mental health research.
Applications
Psychotic disorders affect approximately 1-3% of the population and are among the most disabling health conditions. However, reliable methods to detect psychosis and predict psychosis onset have proved elusive. Recent work has suggested that speech markers might provide a powerful solution to these problems, in line with the fact that altered speech has long been established as a frequent symptom of psychotic disorders. If so, speech markers could lead to new diagnostic methods for schizophrenia and other psychotic disorders, which could open the gateway to early identification and preventative intervention.
Organisers
Professor Maria Liakata
Turing AI Fellow
Professor Ed Bullmore
Professor of Psychiatry, University of CambridgeResearchers and collaborators
Dr Thomas Spencer
Clinical Senior Lecturer in Early Detection & Intervention