Background: Population-scale resources such as UK Biobank (UKB) provide unparalleled statistical power, multimodal neuroimaging, and genetic data, enabling discovery of brain-disease associations. However, their clinical characterization of dementia is limited, reducing relevance to real-world patients. Memory clinic cohorts capture clinically meaningful, heterogeneous presentations of cognitive impairment but have limited participant numbers and data availability, constraining generalisability and discovery potential. The Oxford Brain Health Clinic (OBHC), a real-world memory clinic cohort with brain MRI data matched to UKB, offers a unique opportunity to link clinically grounded dementia phenotypes to population-scale imaging and genetic data.

Methods: Structural T1-weighted MRI from 85 OBHC patients with a dementia diagnosis and 93 age- and sex-matched cognitively unimpaired controls were analysed using a UKB-aligned analysis pipeline optimised for memory clinic use. Voxel-based morphometry was used to generate a dementia-related spatial map of grey matter differences between patients and controls. This map was applied within PANDORA (https://open.oxcin.ox.ac.uk/pages/pandora/web/; Figure 1), which aggregates voxelwise neuroimaging data from 82K UKB participants. Using PANDORA’s voxelwise regression tool, the OBHC dementia map was projected onto UKB imaging data (10K supervoxel version) to generate an individualised dementia score for 82K UKB participants. This score was used for validation against dementia status in UKB, genome-wide association analyses, and interrogation of imaging modalities not used for the original map.

Results: Projection of the dementia-related map produced a continuous dementia score across the UKB population. UKB participants with recorded dementia diagnoses showed elevated scores (Figure A), supporting the validity of translating clinically derived patterns into population data. Genome-wide association analyses using the dementia score as a quantitative phenotype identified genetic variants associated with dementia-related brain structural changes (Figure B). Application of the dementia score to additional UKB imaging modalities revealed associated effects beyond grey matter, including alterations in white matter tracts involved in cognitive pathways, exemplified by associations with diffusion MRI tensor mode (Figure C).

Conclusion: By bridging population-scale imaging with real-world memory clinic data through harmonised imaging and PANDORA, dementia-related information from brain scans can be validated, genetically interrogated, and extended to imaging modalities not available for all individuals, enhancing translational relevance and discovery potential.

References

1 Miller et al., 2016, Nature Neuroscience;

2 O'Donoghue MC. et al, 2023, BMJ Open (https://www.psych.ox.ac.uk/research/research-groups/translational-neuroimaging-group/team/oxford-brain-health-clinic)

3 Griffanti et al., 2022, Neuroimage Clinical; 4Gillis et al., 2025, Human Brain Mapping

5 Abivardi et al., 2026 - https://doi.org/10.64898/2026.01.05.26343425(https://open.oxcin.ox.ac.uk/pages/pandora/web/)

6 Elliott et al, 2028 Nature

Contact: ludovica.griffanti@psych.ox.ac.uk