BACKGROUND:
The Oxford Brain Health Clinic (OBHC) has scanned over 400 NHS memory clinic patients with a magnetic resonance imaging (MRI) protocol aligned with the UK Biobank (UKB). We also acquired the same data from healthy volunteers (HV) of a similar age range. This work explores the added value of UKB-aligned analysis of multimodal MRI to explain variability in cognition in a real-world memory clinic setting.
METHODS:
Six modalities from 140 OBHC patients and 92 HV (Table 1) were analysed with a UKB-aligned image analysis pipeline optimised for memory clinic use (Gillis <i>et al.</i>, 2025), yielding 3520 imaging-derived phenotypes (IDPs). Four models using different sets of IDPs were compared in terms of their ability to predict ACE-III total cognitive score (Table 2). Within a cross-validation framework, model performance (root mean squared error – RMSE) was compared with a Freidman rank and post-hoc Wilcoxon tests with Holm correction. To also explore the value of quantitative IDPs compared to visual ratings, we used the same method on OBHC patients for which both a neuroradiology report (including MTA and Fazekas scores) and structural IDPs were available (N=284).
RESULTS:
The “All IDPs” model outperformed the 3 other models (Figure 1). Upon inspection of IDPs importance within this PLS model, we observed that: 1) the 3 retained components reflect patterns of global brain health, unique vascular changes, and atrophy changes with preserved white matter integrity; 2) the most important IDPs were temporal and peripheral GM volumes, periventricular WMH volume, and inferior fronto-occipital and superior longitudinal fasciculus mean diffusivity. The model with hippocampal and WMH volumes performed slightly better than the visual ratings model (RMSE 0.90 vs 0.93), although not significantly (p=0.08).
CONCLUSION:
Including IDPs from dMRI, ASL, and rfMRI explains more variability in cognition than simpler models without these modalities. The model with hippocampal and WMH volumes performs as well or slightly better than the one with neuroradiologist visual ratings on these structures.