Waiting for a dementia diagnosis can take months, and in some parts of the UK more than a year. A major UK programme announced this week aims to change that by testing nine projects directly in the NHS, including AI that could help doctors predict how a patient's dementia is likely to progress.
Up to £80 million to tackle the dementia diagnosis bottleneck
UK Research and Innovation (UKRI) is backing the Dementia Challenge with up to £80 million, bringing together researchers, technology companies, the NHS and regulators.
The ambition is significant. Reporting on the announcement puts the share of patients diagnosed within 18 weeks of referral at around 60% today, with a goal of 92%. UKRI's formal target is for 92% of people referred for dementia assessment to receive their assessment results within 18 weeks by 2029. That is a target, not a result already achieved, and an assessment does not necessarily confirm dementia.
This is not one giant bet on a single piece of technology. The approaches being tested include:
- AI that analyses NHS patient data to help predict how dementia is likely to progress.
- Blood-based dementia biomarkers that could improve diagnostic accuracy.
- A 15-minute remote digital cognitive assessment.
- A home-based digital test examining how the brain processes sound, using a computer and headphones.
- Sensors and digital tools to monitor changes and support earlier intervention.
- Faster MRI approaches that could increase scanning capacity.
The nine projects also include redesigning dementia diagnosis and treatment services in Scotland, and routine sleep-apnoea screening in memory assessment services to reduce the risk of misdiagnosis.
The University of Cambridge leads the AI clinical decision-support project. It will analyse NHS datasets to investigate whether doctors can better predict how an individual's disease will progress. Its clinical benefits still need to be demonstrated.
Sources: Reuters: funding announcement ↗ · Alzheimer Scotland: the nine projects ↗ · UKRI: objectives and assessment target ↗
Could a dementia MRI take just seven minutes?
One of the most immediately understandable ideas is simply making brain scans quicker. Researchers at University College London have already demonstrated an accelerated MRI protocol in 92 outpatient volunteers, reducing imaging time from almost 18 minutes to around six and a half minutes.
Neuroradiologists assessed the standard and faster scans without knowing which was which. UCL reports that the fast scans were as reliable for clinical assessment in that study, including diagnosis, ratings of brain shrinkage and assessment of eligibility for new treatments.
That is potentially important because an MRI scanner is an expensive, limited resource. The approach uses existing scanners rather than requiring new ones.
UCL estimates that cutting the imaging component to six or seven minutes could allow hospitals to scan three or four patients an hour instead of two. But imaging time is not the whole appointment: preparation and helping patients into and out of the scanner still take time. The capacity increase is a possibility, not a throughput gain already demonstrated across the NHS.
The new programme is about taking approaches like these into NHS testing and finding out which ones work in practice. Faster imaging is not itself evidence that an AI tool can diagnose dementia, and the 92-volunteer study does not establish performance in every clinical setting.
Source: UCL: transforming Alzheimer's disease diagnosis ↗
Funding should follow the evidence
There is something I particularly like about how the programme is being funded. UKRI is using what it calls a “fail-fast” approach: projects have to reach milestones and demonstrate results to continue receiving funding.
If something is not delivering, the idea is not to keep pouring money into it simply because it sounded exciting. Further funding goes towards the approaches showing genuine promise. That is a practical way to approach AI and innovation.
This is not £80m of proven AI treatment
There is an important distinction. This announcement does not mean AI has solved dementia diagnosis. Nor is all the funding for AI: the programme includes diagnostic technologies and changes to how services work.
Some approaches already have encouraging research behind them, such as accelerated MRI. The wider programme is about testing, validating and implementing promising approaches in real NHS settings. The AI disease-progression tool, digital assessments and other projects still have to demonstrate that they genuinely improve care.
UKRI sets out plans for evaluations to begin in the first three months of 2027, with evaluations completed by March 2029 and at least two approaches introduced for routine NHS use by the end of 2029. Those are planned milestones, not confirmation that new tools are available everywhere today.
Sources: Alzheimer Scotland: funding and evaluation approach ↗ · UKRI: planned evaluation timetable ↗
Find the bottleneck first
There is a useful business lesson here too. The NHS problem is not simply “We need some AI.” It is “People are waiting too long for a diagnosis. Where are the bottlenecks?”
Scanning capacity might be one. Specialist assessment might be another. Monitoring patients might be another. Only then do you ask whether AI, automation or another technology can help.
That is exactly the approach we take at OTB Optimisations. Do not start with the technology. Start with the problem you are trying to remove.
No AI for the sake of AI.
Less hype. More real-world AI progress.