Language Sensing Study: Dementia Diagnosis and Monitoring
Research type
Research Study
Full title
Language Sensing Study: Dementia Diagnosis and Monitoring using Natural Language Processing
IRAS ID
205999
Contact name
Maria Liakata
Contact email
Sponsor organisation
University of Warwick
Clinicaltrials.gov Identifier
N/A, N/A
Duration of Study in the UK
3 years, 5 months, 30 days
Research summary
The rate and accuracy of dementia diagnosis varies greatly. According to recent ARUK figures only 59% of people with dementia receive a formal diagnosis. Current methods for diagnosis are expensive and intrusive, including brain scans and expensive spinal fluid tests.
We aim to use automated methods for processing language to create a cost effective, non-invasive method for monitoring changes in linguistic ability for dementia diagnosis and monitoring. More specifically our objectives are to:
(a) Create a platform for collecting conversation data between people with dementia and their carers over time, as well as written logs, triggered by meaningful images from the past.
(b) Use this longitudinal data collected to develop computational methods for studying patterns of language change.
The project will use a purpose built computer tablet application and secure server to obtain recordings of conversations as well as written text data from study participants living with dementia. The data collection framework is being developed using funding we have obtained through the EPSRC Impact Acceleration Account. Our pilot study will comprise 30 study participants from whom we will collect conversations and written text for a year, in three intermittent 4-week periods. The participants will engage in dialogue with their carers on the basis of images from the past provided by the tablet application, 15 mins of which will be recorded daily. On alternate days, participants will also be asked to write their thoughts on the tablet.
Once data is collected, we will work on a prototype diagnostic software tool based on language processing and statistical methods. Our algorithms will be designed to track use of vocabulary and syntax, emotional content, fluency, topic relevance and how these features interact and change over time. The tool and data analysis will allow us to build predictive models of dementia presence and severity.REC name
West of Scotland REC 5
REC reference
16/WS/0226
Date of REC Opinion
17 Jan 2017
REC opinion
Further Information Favourable Opinion