112 Chapter 5 with 1000 permutations, and a voxelwise threshold of p < .005 without additional correction for multiple comparisons. Since severity of symptoms is often related to lesion size, we performed lesion volume correction by regressing lesion volume on both behavioral scores and lesion data (Supplementary Table 2), in line with the recommendations by DeMarco and Turkeltaub.40 To assess the effects of this lesion volume correction, all analyses were repeated without correcting for lesion volume (see Repository for maps). We ran the SVR-LSM for each cognitive task for the tumor and stroke data separately. For each LSM analyses the area containing most voxels with peak significance was calculated. Additionally, we combined the data of both groups and performed the SVR-LSM analyses for each cognitive task using etiology (tumor or stroke) as a covariate on both the behavioral scores and lesion data (Supplementary Materials). The AALCAT atlas was superimposed on the results to relate significant voxels to brain regions. This atlas combines the 116 regions from the AAL atlas43 with 34 white matter regions from the tractography atlas.44 Areas with peak significance and/or areas with at least 10% of tested voxels significant are reported in the text. Univariate lesion-symptom mapping In LSM statistical testing is performed to identify voxels in which individuals with a lesion perform significantly worse compared to individuals without a lesion in that voxel. With the univariate method this statistical test is independently applied to each voxel in the brain, whereas multivariate LSM considers the effect of all lesioned voxels simultaneously. To substantiate the results from the multivariate LSM, additional univariate LSM was performed for each cognitive task for the tumor and stroke data separately, using the statistical analyses software NiiStat (https://github.com/neurolabusc/NiiStat). With continuous behavioral data NiiStat computes statistics using a general linear model. Only voxels damaged in at least 3 patients were considered in the analyses. In line with the multivariate LSM, lesion volume correction was performed. Lesion volume control in NiiStat is based on regressing lesion volume with the behavioral data only. To assess the effects of this lesion volume correction, all analyses were repeated without correcting for lesion volume (see Repository for maps). Permutation testing to correct for multiple testing was set to 10,000 permutations and a voxelwise threshold of p < .05 was used. Statistical power maps were generated for each LSM using the “nii_power” function of NiiStat, with a critical one-tailed threshold of p < 0.05 and a power of 0.6. Power maps hereby represent the number of patients that would be needed to replicate the results in 60% of the studies. A maximum number of 200 patients was chosen as adequate power. Next, for each cognitive task the percentage of voxels with adequate power was calculated relative to three different volumes:
RkJQdWJsaXNoZXIy MTk4NDMw