This project aims to demonstrate that a future diagnosis of dementia can be predicted from a brain scan. A key challenge for health systems is to identify people at highest risk of dementia who are eligible for prevention treatments. Solving this problem could empower people to manage their own risk of brain diseases, make trials of new treatments easier to perform, and support healthcare systems.
Brain scans acquired in research contexts using machine learning (ML) have been used in the past to accurately predict increased risk of neurodegeneration. Such techniques are not currently available in the NHS and if they were, it would facilitate automated identification of clinical interventions, for high-risk patients. This study aims to prove the concept that a future diagnosis of dementia can be predicted from a routinely collected brain scan and is a tangible step towards being able to implement such a technique in the NHS in future.
We have developed a predictive model that can estimate a person’s risk of dementia with a high rate of success, trained on Magnetic Resonance Imaging (MRI) brain scans of Tayside and Fife patients who, through the SHARE register, consented to their medical records being used for research purposes. We now aim to test the effectiveness of this predictor (a Support Vector Machine or ‘SVM’) on a much larger, population-level dataset in the National Safe Haven. We will use routinely collected brain scans, that people have had taken as part of their care in the NHS Scotland. MRI scans will be linked to other routinely collected health data from hospital and pharmacies. Using data from people across the whole of Scotland will provide previously unavailable scale, population-wide representativeness, long-term follow-up, and real-world variation. We will then see how well the SVM predictor trained on Tayside and Fife data scales up to perform on population-level data, the aim being to demonstrate that it can be implemented in a national collection of routinely collected brain images to predict dementia using information from the scans and from the other information held in medical records.
An efficient solution to readily identify high-risk people from electronic health data would have broad benefits. It would transform the costs and duration of dementia and brain health trials leading to improved identification of new treatments, and accurate prediction of dementia would mean people at the highest risk could improve their risk factor management – as already occurs for cardiovascular diseases – and be identified by health systems for targeted follow-up assessment and care.