Automated brain image analysis for timely and equitable dementia diagnosis
Award Number
NIHR203373Programme
Invention for InnovationStatus / Stage
ActiveDates
6 February 2022 -6 January 2025
Duration (calculated)
02 years 11 monthsFunder(s)
NIHRFunding Amount
£977,693.00Funder/Grant study page
NIHRContracted Centre
Queen Mary University of LondonContracted Centre Webpage
Principal Investigator
Dr Charles MarshallPI Contact
charles.marshall@qmul.ac.ukPI ORCID
0000-0002-8227-2354,0000-0002-6582-1163Principal Investigator
Dr Hojjat AzadbakhtWHO Catergories
Development of clinical assessment of cognition and functionMethodologies and approaches for risk reduction research
Disease Type
Dementia (Unspecified)CPEC Review Info
| Reference ID | 6 |
|---|---|
| Researcher | Reside Team |
| Published | 12/06/2023 |
Data
| Award Number | NIHR203373 |
|---|---|
| Status / Stage | Active |
| Start Date | 20220206 |
| End Date | 20250106 |
| Duration (calculated) | 02 years 11 months |
| Funder/Grant study page | NIHR |
| Contracted Centre | Queen Mary University of London |
| Contracted Centre Webpage | |
| Funding Amount | £977,693.00 |
Abstract
There are substantial problems with timeliness and equity of dementia diagnosis. Average time from symptom onset to diagnosis is over 3 years. There is regional variation in diagnosis, and those who are more deprived or from minoritised ethnic groups tend to be diagnosed later and less accurately. Timely and accurate diagnosis is important. It empowers people with cognitive concerns to understand their symptoms, plan for the future and seek appropriate treatments, whether or not they have dementia. Timely diagnosis will be vital to ensure access to disease modifying treatments that might reduce the burden of dementia on individuals and the wider society. The first such treatment has recently been approved in the USA and submitted to NICE for evaluation . A substantial proportion of people attending memory clinics receive a diagnosis of mild cognitive disorder (MCD) when it is not yet clear whether their symptoms are due to early dementia. Brain MRI is routinely acquired for these patients, but human interpretation of it does not reliably predict the underlying diagnosis. Currently the only means of establishing the diagnosis is to follow patients up for at least 1-2 years. This follow-up is resource intensive and in many areas is not commissioned. AINOSTICS has developed a proprietary deep learning algorithm for interpretation of routinely acquired brain MRI that is highly accurate at predicting the future development of dementia from MCD in research datasets. This project will obtain efficacy data for this technology in a real-world , clinical setting, and develop a software prototype for further validation and incorporation into NHS pathways. The technology is designed to be cost-effective in the NHS by using only data that are already routinely acquired in the diagnostic pathway. We will perform clinical validation in memory clinics in East London that serve a particularly diverse and deprived population, where diagnostic rates are low. We will identify a retrospective cohort of 150 patients with an initial diagnosis of MCD and at least 2 years passive follow up, and a prospective cohort of 150 patients receiving a diagnosis of MCD whom we will follow up actively at 2 years. This will allow us to maximise numbers in the study whilst being able to assess the effect of possible incomplete ascertainment of dementia diagnoses in the retrospective cohort. We will establish the diagnostic accuracy of the technology in this real-world setting, before performing cost effectiveness analyses of its potential value in NHS diagnostic pathways both for earlier diagnosis of dementia and for earlier exclusion of dementia. We will estimate potential improvements in diagnostic equity that could result from its use by estimating how much closer diagnostic performance in our region would be to national standards. We will map the future regulatory and commercial strategy for the technology in the NHS. We will co-produce the research with members of the public and clinicians involved in routine care, working with them to understand the potential value of the technology to patients and carers, and how best to communicate the outputs in clinical settings.
Plain English Summary
We aim to provide more timely and accurate diagnosis of dementia by using artificial intelligence to interpret brain scans. People are referred to memory clinics when there is a suspicion that they might have dementia. However, there are other causes of memory difficulty, and often it is impossible to know whether someone has dementia when they are first assessed. Doctors refer to this uncertain situation as “mild cognitive disorder” (MCD). Currently, the only way to establish the diagnosis is to follow people over time to see if things get worse. This follow-up is often not available, and so people are discharged without a clear answer. Timely diagnosis is important. Those without dementia benefit from early reassurance and treatment for other causes of memory difficulty. For those with dementia, diagnosis enables them to plan for the future, understand their symptoms, and access treatments. Early diagnosis will be even more important when there are treatments that could slow or stop dementia from worsening. The first such treatment was recently approved for use in the USA. Current NHS infrastructure would be unable to provide the timely diagnosis required for everyone to benefit from such treatments. Brain scans are a routine part of memory clinic assessment. Dementia causes shrinking of the brain, but when humans interpret scans, this only provides a clear diagnosis when dementia is quite advanced. We have developed a technology for computerised interpretation of brain scans. This can predict whether somebody with MCD will develop dementia with 92% accuracy. There is regional variation in dementia diagnosis, and those who are more deprived or from minoritized ethnic groups tend to be diagnosed later. East London is very diverse and deprived, with low rates of dementia diagnosis. We will use our technology to interpret the brain scans of 300 people with MCD in East London memory clinics to predict who will develop dementia and who will not. Because we are still establishing the accuracy of the technology, we will not feed back individual results during this study. The project will only include people that have been referred to memory clinics with a suspicion of dementia, so we will not be attempting to predict future development of dementia in people who are completely well. We will establish how many people without dementia could be reassured and discharged earlier and how many people with dementia could be diagnosed earlier. We will estimate improvements in diagnostic accuracy, fairness and access to treatment that result, as well as the cost-effectiveness of the technology for use in the NHS. We will co-design the project with patients and the public, and work with them to understand how to communicate the results of the technology in clinics.
Aims
This project will obtain efficacy data for this technology in a real-world , clinical setting, and develop a software prototype for further validation and incorporation into NHS pathways. The technology is designed to be cost-effective in the NHS by using only data that are already routinely acquired in the diagnostic pathway.