Visual rating versus volumetric segmentation of white matter hyperintensities for cognitive assessment: a memory clinic-based comparison and systematic review

INTRODUCTION Clinical impression of WMH commonly relies on visual rating scales (VRS), yet may improve through volumetric segmentation. Comparative studies, particularly considering clinical outcomes, are scarce. METHODS We conducted a retrospective, cross-sectional study among 246 memory clinic patients. We compared WMH assessed through VRS (Fazekas) to automated volumetric segmentation (Quantib), through correlation and explained variance in neuropsychological assessment across six cognitive domains, above and beyond age, sex, and educational attainment in linear regression models. Results were contextualised through systematic literature review until 30 September 2022. RESULTS Fazekas scores and WMH volumes were strongly correlated in the memory clinic population (ρ=0.83, p<0.001), with little overlap in volumes between Fazekas categories. This was supported by similar correlations observed in 24 studies included in the systematic review and 13 in meta-analysis. Jointly with demographics, VRS and segmentation explained 6-19% of variance in cognitive domains in the memory clinic population. Volumetrics explained at most 2% more variance than visual rating, most substantial in MMSE (2%), in contrast to executive function (-1%). The systematic review identified three other comparative studies on cognitive outcome, each reporting similar correlations with volumetrics and visual rating scales. Previously proposed cut-offs in WMH volume that approximate Fazekas categories (3.4, 9.6 and 17.1 mL) showed classification accuracies of 59%, 70% and 78% in our cohort, with optimal thresholds in our sample at 1.6, 6.0 and 19.8 mL. DISCUSSION VRS and volumetric segmentation of WMH are highly correlated. Small improvements in explained variance in cognition with volumetrics were inconsistent across cognitive domains