Project Rhapsody: Investigating the clinical feasibility of using AI-based deep audio and language processing techniques to diagnose neurological and psychiatric diseases

Award Number
AI_AWARD01984
Programme
Artificial Intelligence
Status / Stage
Active
Dates
4 February 2021 -
10 January 2022
Duration (calculated)
00 years 11 months
Funder(s)
NIHR
Funding Amount
£149,787.00
Funder/Grant study page
NIHR
Contracted Centre
Novoic Ltd
Contracted Centre Webpage
Principal Investigator
Mr Emil Fristed
PI ORCID
0000-0002-9590-7275
WHO Catergories
Development of clinical assessment of cognition and function
Disease Type
Alzheimer's Disease (AD)
Frontotemporal Dementia (FTD)
Lewy body dementia (LBD)

CPEC Review Info
Reference ID29
ResearcherReside Team
Published12/06/2023

Data

Award NumberAI_AWARD01984
Status / StageActive
Start Date20210204
End Date20220110
Duration (calculated) 00 years 11 months
Funder/Grant study pageNIHR
Contracted CentreNovoic Ltd
Contracted Centre Webpage
Funding Amount£149,787.00

Abstract

Neurological and psychiatric disorders (NPDs) present a threatening economic burden for the NHS.3-7 Similar clinical presentations and prevalent comorbidities of common NPDs31-46, make diagnoses in primary care difficult. State-of-the-art diagnostics in primary care are inaccurate14-20 and cannot disambiguate common NPDs12-13. High misdiagnosis rates lead to mistreated patients, exacerbating the burden47-52. Changes in people s speech have been found in many NPDs, years before they re formally diagnosed21-28,53-55. Our two most recent papers outline speech changes in 14 different neurological and psychiatric conditions29-30. Clinical applications have been limited by lack of harmonisation29-30 across indications, speech tasks, and recording environments. Our proposed innovation addresses these previous barriers using domain-adaptive representation learning, to build the first clinical application of speech biomarkers across multiple NPDs – a universal test. We are building these next-generation speech biomarkers into an API and mobile application that take a few minutes of speech as input and output a disease risk score for clinical decision support. This project firstly evaluates usability and feasibility of administering a universal speech battery – both in clinic and remotely – across common representative NPDs, having similar presentations or being highly comorbid. The following cohorts are enrolled into this observational, case-control study: Alzheimer s Disease, Dementia with Lewy Bodies, Parkinson s Disease, Motor Neuron Disease, Frontotemporal Dementia, Depression, and Bipolar Disorder. Secondly this project evaluates both (1) using our speech analysis libraries29-30 to analyse markers already reported in individual NPDs29-30, consistently across multiple NPDs, and (2) identifying new universal markers using un/semi-supervised learning approaches. The full duration of the project is 12 months. Results will be published. At successful completion, follow-up projects will be initiated to further validate the discovered biomarkers in each of the diseases, and implement them into NHS workflows. NHS adoption strategy is created as part of the project.

Aims

This project firstly evaluates usability and feasibility of administering a universal speech battery – both in clinic and remotely – across common representative NPDs, having similar presentations or being highly comorbid. The following cohorts are enrolled into this observational, case-control study: Alzheimer s Disease, Dementia with Lewy Bodies, Parkinsons Disease, Motor Neuron Disease, Frontotemporal Dementia, Depression, and Bipolar Disorder. Secondly this project evaluates both (1) using our speech analysis libraries29-30 to analyse markers already reported in individual NPDs29-30, consistently across multiple NPDs, and (2) identifying new universal markers using un/semi-supervised learning approaches.