Arjen Lindenholz

115 Intracranial Atherosclerosis Assessed with 7T MRI 5 assessed in multiple planes because of the isotropic voxel size. When the vessel wall regions were bilaterally affected, contralateral assessment was performed with extra caution or was omitted. For assessment of contrast enhancement, postcontrast vessel wall MRI scans were assessed for higher signal intensity over at least two slices using the corresponding precontrast MRI scan as a reference. In addition, pre- and postcontrast vessel wall MRI scans were coregistered and subsequently matched for the whole 3D volume using advanced algorithms of the elastix toolbox in MeVisLab (version 2.7; MeVis Medical Solutions, Bremen, Germany). 29 The precontrast images were subtracted from the coregistered postcontrast images, and the resulting subtraction images were then used to visually confirm contrast enhancement. 18 Vessel wall enhancement at the location where the ICA or the vertebral artery pierces the dura mater, which is suspicious for vasa vasorum extension in the intracranial portions of the vessel wall, was not counted as an enhancing vessel wall lesion. 18,30 For participants who underwent imaging during the first 1.5 years of the study with the smaller FOV MPIR-TSE sequence, the posterior circulation could not be completely assessed for vessel wall lesions because the vertebral arteries and basilar artery were often outside the FOV. Therefore, we differentiated between anterior and posterior circulation lesion burden, with lesions in the anterior circulation as the primary outcome. In participants in whom both the anterior and posterior circulation could be assessed, an additional analysis was performed with the total burden of vessel wall lesions as outcome. Statistical analysis The intraclass correlation coefficient (ICC) (with a two-way mixed, single measurements, consistency model for the number of detected vessel wall lesions) and the Dice similarity coefficient (DSC) (for the number of vessel wall lesions detected at the same location) were calculated to measure interrater reproducibility. 31 A Bland-Altman plot was used to assess the agreement between the two readers. Descriptive statistics are given in percentages or are summarized as means with a range or standard deviation. Generalized Estimating Equations (GEE) for Poisson regression were performed to assess the relation between vascular risk factors and the number of (enhancing) vessel wall lesions in the anterior circulation. We selected a maximum of 10 predictors, apart from age and sex, to avoid overfitting (one-in-10 rule) which were previously described as well- established risk factors for ICAS to include in the GEE Poisson regression models. 8 All included variables were tested for multicollinearity using both Pearson and Spearman correlation methods and assessed for the Goodness-of-Fit by using R and R 2 . Where possible, we used continuous variables over dichotomous variables in the regression analysis for improving statistical reliability. 32 The vascular risk