128 Chapter 5 performance when lesion volume was not taken into account. Largest differences between corrected and uncorrected LSM results were seen for the tumor group, especially for the RAVLT Delayed recognition and Semantic fluency task. Table 3. Percentage of voxels with adequate power for univariate LSM (maximum of 200 patients needed to replicate results in 60% of studies) relative to the etiology-specific coverage for the cognitive task, the overlap in coverage between the tumor and stroke group for the cognitive task and the complete MNI volume. Cognitive task Coverage Overlap Whole-brain Tumor Stroke Tumor Stroke Tumor Stroke Direct recall 31.8 18.2 28.9 19.4 23.7 2.0 Delayed recall 32.2 13.0 26.8 13.7 24.0 1.4 Delayed recognition 20.8 10.5 20.4 11.2 15.5 1.1 Letter fluency 42.6 34.4 30.9 27.7 26.6 2.1 Semantic fluency 42.9 16.0 30.0 17.0 30.2 2.4 DISCUSSION In this study, we investigated whether etiology-specific biases impact brain-behavior associations in LSM. This study is unique in comparing results from state-of-the-art machine learning-based, multivariate LSM between two large study-populations, a tumor versus a stroke population. We expected overlapping lesion-symptom associations in brain areas affected in both populations and population-specific lesion-symptom associations in brain areas covered per pathology. Despite our large sample sizes, substantial differences in lesion distribution conditioned the degree of lesion overlap between stroke and tumor and hindered direct comparison for most brain areas. Still, both our LSM results as well as the post-hoc analyses suggested that the cognitive effects of damage in certain brain areas depend on the underlying pathology. Lesion topography Both study samples showed a wide distribution of lesions covering both hemispheres. Tumors were most prevalent in the left hemisphere, with the highest frequency in the insula, operculum and superior temporal gyrus, while strokes were predominantly located within the territory of the right middle cerebral artery. As patients selected for awake surgery are characterized by tumor localization in eloquent areas, such as language and motor areas, this left-sided predominance was to be expected. Nevertheless, lesion coverage was adequate in both hemispheres. Overall, lesion overlap was both more widespread and higher per brain structure
RkJQdWJsaXNoZXIy MTk4NDMw