Spoken dialogue analysis for automatic monitoring of cognitive status

  • Research type

    Research Study

  • Full title

    Conversation-based analysis approach for automatic cognitive monitoring of population at risk of dementia. A study within the PREVENT (PREVention of dementia by ENvironmental intervention and Therapy) project to investigate dialogue features that may help predict dementia onset in later life

  • IRAS ID

    245117

  • Contact name

    Saturnino Luz

  • Contact email

    s.luz@ed.ac.uk

  • Sponsor organisation

    University of Edinburgh

  • Clinicaltrials.gov Identifier

    PREVENT Dementia NHS REC (general project), 12/LO/1023

  • Duration of Study in the UK

    0 years, 11 months, 30 days

  • Research summary

    We present a proposal to study a novel way of monitoring cognitive status for healthy people, who are considered to be at risk of developing dementia in the future. Our approach to monitor cognition is based on the a analysis of spoken language, extracted from spontaneous conversations.

    The proposed experimental design for data collection involves presenting the study participants with a map, and ask them to find a route, together with the experimenter. It is designed as a cooperative task in order to generate spontaneous speech in the form of a discussion between the participant, leading the route, and the experimenter, following it. At the same time, this task provides a degree of consistency, as the map will be the same across participants.

    These conversations will be recorded and analysed in order to extract a range of features, which we expect will contribute to predict cognitive status. Despite relying solely on content-free (acoustic) features, our method has obtained an overall 83% accuracy in preliminary analyses of Alzheimer's Disease detection, a result comparable to more sophisticated methods that employ complex lexical and syntactic features. This makes our approach potentially more robust and inclusive, as we aim for it to be generalizable across different settings and languages.

    We will further investigate this method by combining the features extracted from the dialogues, with demographic, neuropsychological and genetic variables from the ongoing PREVENT dementia study. The eventual goal of our project is to automatise this conversational analysis, in order to contribute to the design of non-invasive screening processes, low-cost decision-support tools for clinicians and secondary prevention strategies.

  • REC name

    London - Surrey Research Ethics Committee

  • REC reference

    18/LO/0860

  • Date of REC Opinion

    1 Jun 2018

  • REC opinion

    Further Information Favourable Opinion