Abstract 2361: Comparison of Ageing-related White Matter Lesion Quantification by Volume and Visual Rating Scores

Background: White matter lesions (WMLs) increase with age and are associated with stroke, cognitive decline and dementia. WMLs can be quantified by visual rating scores, or by computational measurement of WML volume, but the concurrence between these two methods has not been widely studied. We compared WML visual ratings and computed volumes to determine agreement and sources of disagreement. Methods: We used subjects from the Lothian Birth Cohort 1936 who had brain MRI. We rated WMLs visually on FLAIR using the Fazekas scale and measured WML volume using a validated multispectral image fusion technique: MCMxxxVI ( sourceforge.net/projects/bric1936 ). We summed the deep and periventricular Fazekas scores and calculated the correlation between the total Fazekas score (0-6) and WML volume (Spearman's ρ). We sought explanations for outliers, such as stroke lesions. Results: Amongst 672 subjects with full brain imaging data, the median Fazekas score was 2 and the median WML volume was 7.7 ml (IQR 13.6ml). The Fazekas score and WML volume were highly correlated (Spearman ρ=0.73, p<0.001). Including stroke lesions, most of which were small but which can inflate WML volume values, gave a similar correlation (Spearman ρ=0.75, p<0.001). In 114 subjects (17% of the total) the z-scores of WML volume and Fazekas rating differed by >1, most of whom had a total Fazekas score of 1 (n=26, WML volume:0-14.8ml) or 2 (n=63, WML volume:0-34.37ml). The main reasons for disagreements were: 1) subtle WMLs visually identified which were omitted from the WML volume; 2) prominent periventricular caps with thin ventricular body lining obtaining a Fazekas periventricular score of 1 or 2 but a large WML volume; and 3) small deep white matter focal lesions which increase disproportionally the score obtained in the rating scale if they are beginning to coalesce but add little to the WML volume which may range from <0.5 to >12ml. Conclusions: WML visual rating and volumes show high correlation, but subtle WMLs and two features that are not well differentiated in the WML score weakened the agreement. Methods to improve computational WML quantification (e.g. adaptive filtering) would improve measurement of subtle WML areas. Consideration should be given to modifying WML scores to differentiate ventricular caps from other periventricular lesions and small but early coalescent from larger deep WMLs.