Poststroke cognitive outcome is better accounted for by white matter abnormalities automated segmentation than visual analysis.

Background and aims: The association between white matter abnormalities (WMA) and cognitive decline previously reported in poststroke patients has been mainly documented using visual scales. However, automated segmentation of WMA provides a precise determination of the volume of WMA. Nonetheless, it is rarely used in the stroke population and its potential advantage over visual scales is still unsettled. The objective of this study was to examine whether automated segmentation of WMA provides a better account than the visual Fazekas and Wahlund scales of the decline in executive functions and processing speed in stroke patients.

Methods: The analyses were conducted on the 358 patients of the GRECogVASC cohort with an MRI performed at six months poststroke in the Amiens center. WMA were visually analyzed using the Fazekas (subcortical abnormalities) and Wahlund scales. Segmentation was performed using LST (3.0.3). Following preliminary studies to determine the optimal segmentation threshold, we examined the relationship between cognitive status and WMA volume computed at each threshold using receiver operating characteristic (ROC) curves. Finally, we assessed the ability of both Fazekas and Wahlund visual scores and WMA volume to account for cognitive scores by using a bivariate Pearson correlation analysis, comparing correlation coefficients with the Fisher transformation and repeating correlation analysis after adjustment for the lesion volume.

Results: Increasing the threshold led to an underestimation of WMA (P=0.0001) (significant for a threshold ≥0.2) and an improvement in correct rejection of signal changes in the stroke cavity (P=0.02) (significant for a threshold ≤0.5), susceptibility artifacts (P=0.002) (significant for a threshold ≤0.6), and corticospinal degeneration (P=0.03) (significant for a threshold ≤0.5). WMA volume decreased with increasing threshold (P=0.0001). Areas under the curve (AUC) did not differ according to the threshold (processing speed: P=0.85, executive cognitive functions: P=0.7). Correlation coefficients between cognitive scores and WMA were higher for WMA volume than the Fazekas (processing speed: Z=-3.442, P=0.001; executive functions: Z=-2.751, P=0.006) and Wahlund scores (processing speed: Z=-3.615, P=0.0001; executive functions: Z=-2.769, P=0.006). Adjustment for lesion volume did not alter the correlations with WMA volume (processing speed: r=-0.327 [95%CI: -0.416; -0.223], P=0.0001; executive functions: r=-0.262 [95%CI: -0.363; -0.150], P=0.0001).

Conclusion: This study shows that WMA volume assessed by automated segmentation provides a better account of cognitive disorders than visual analysis. This should favor its wider use to refine imaging determinants of poststroke cognitive disorders.