**2.4 MRI data acquisition and analyses**

The MRI scans were conducted on a 1.5 T scanner (Stratis II, Premium; Hitachi Medical Corporation, Tokyo, Japan), and a high-resolution T1-weighted 3D image (repetition time:

Fig. 3. Schematic representation of methods for making GM images.

30 ms, acquisition time: 8 ms, flip angle: 60°, field of view: 192 × 192 mm2, resolution: 0.75 × 0.75 × 1 mm3) was acquired for each patient. The location of the glioma was first identified on this MR image, and the glioma boundary was semi-automatically determined using MRIcroN software (http://www.mricro.com/) (Rorden & Brett, 2000). T2-weighted MR images (Department of Neurosurgery of Tokyo Women's Medical University) and positronemission tomography (PET) data (Chubu Medical Center for Prolonged Traumatic Brain Dysfunction, Mino-Kamo-shi, Japan) were also used to assist the precise determination of the boundary. Each individual's structural image was spatially normalized to the standard brain space as defined by the Montreal Neurological Institute (MNI) using the "unified segmentation" algorithm, which is a generative model that combines tissue segmentation, bias correction and spatial normalization in a single unified model (Ashburner & Friston, 2005), which was resampled to 1 × 1 × 1 mm3 voxel size using statistical parametric mapping SPM8 software (Wellcome Department of Cognitive Neurology, London, UK) (Friston et al., 1995) on MATLAB (Math Works, Natick, MA, USA).

These resultant individually normalized images were divided into gray and white matters as follows (Fig. 3). Firstly, a GM image was made by dividing the standard brain into gray and white matters using MRIcroN software. This GM image was used as a mask, which applied to the individually normalized image of each glioma.

Using the resultant GM image of each glioma, we next employed voxel-based lesionsymptom mapping (VLSM) to analyze the relationship between glioma location and the error rates on a voxel-by-voxel basis (Bates et al., 2003). The patients were divided into two groups according to whether they did or did not have a glioma including that voxel. The error rates for each condition or the difference in error rates between two conditions (e.g., PS – AS) were then compared for these two groups by a t-test, in which the statistical threshold was set to p = 0.05 after correction for multiple comparisons using the false discovery rate (FDR). To minimize the effects of outlier observations, the voxels used in the VLSM analysis were within the gliomas of at least two patients. Finally, the result of VLSM was projected onto a standard brain using MRIcroN software.
