133 Etiology in Lesion-Symptom Mapping: Tumor vs. Stroke each pathology. Secondly, LSM in tumor series may be especially valuable to make assumptions about function localization prior to neurosurgery. Study limitations and further directions Methodological choices LSM tools are used to identify the neural structures critical for a given behavior. However, multiple methodological choices can influence subsequent LSM results. In the current study we performed LSM using two different types of analyses; a univariate and multivariate approach. Previous research has indicated neither method is superior, but rather both have advantages and disadvantages.1,10 For example, the multivariate SVR-LSM used in the current study could have been affected by the hyperparameter selection. As no clear criteria on parameter choice are available yet hyperparameter values were kept in line with the original paper. However, this selection was based on a specific stroke sample, with larger lesion sizes (median 76.79 cc) than our current sample.39 Therefore, we corroborated our multivariate LSM results using univariate LSM. However, univariate voxelwise LSM can be conservative due to strict multiple testing corrections. This was also observed in our results with only a small number of voxels surviving significance testing. Nonetheless, most regions related to task performance in the univariate LSM overlapped with those from the multivariate LSM, thereby corroborating our results. Still, we cannot directly compare results from the univariate and multivariate analyses, since the lesion volume correction differed between methods and multiple comparison corrections were only performed for univariate analyses. Additionally, we chose to perform lesion volume correction in order to eliminate the effect of lesion volume on our comparative analyses as much as possible. The importance of incorporating lesion volume into LSM analyses has been recommended for several years for both multivariate and univariate LSM.40,68,69 Nevertheless, for the SVR-LSM this means a double covariate control was performed, thereby possibly further decreasing the power to find overlapping regions between the tumor and stroke group. When comparing uncorrected and corrected LSM results, more areas were related to worse cognitive performance when lesion volume was not taken as a covariate, for both univariate and multivariate analyses. As to be expected, this effect was most pronounced for the tumor population as lesion size was significantly related to task performance (memory and fluency) in the tumor, but not stroke group. However, regardless of the specific LSM method that we used, there was minimal overlap between tumor and stroke LSM results, thereby substantiating our conclusion. 5
RkJQdWJsaXNoZXIy MTk4NDMw