A national library of brain scans linked to health records in Scotland
We describe how a huge set of Scottish head scans was prepared, linked to health records and made available for AI research.
The question
To build AI that reads brain scans well, researchers need many scans that reflect real hospital care, linked to what later happened to the patients. Such collections are rare, partly because assembling them is hard.
What we did
We built a framework to prepare a very large clinical brain imaging dataset from NHS Scotland, as part of the Brain Health Data initiative. It holds around 417,000 MRI and 846,000 CT head studies from 2010 to 2018, linked to health records and the written scan reports, and is more than 185 terabytes in size. A subset of nearly 42,000 MRI series was used to study dementia prediction.
What we found
The paper describes the dataset’s strengths, including its scale, its coverage of a whole country’s free-at-the-point-of-use care, long follow-up for brain diseases, and real-world variety. It also shares lessons on what was hard, including waiting for permissions, getting secure and affordable computing, messy clinical data, and linking scans and records at scale.
Why it matters
This resource can speed up research on dementia and other brain diseases. Researchers can apply to use it through the Public Health Scotland eDRIS service.