Matlab platform

In preprocessing of fMRI time series volume data, the Resting‐State fMRI Data Analysis Toolkit (REST, V1.8; http://www.restfmri.net) and Statistical Parametric Mapping (SPM12; http://www.fil.ion.ucl.ac.uk/spm) were accessed, using MATLAB platform (MathWorks). The first 10 volumes of each functional time series were discarded to avoid transient signal changes before magnetic field steady states were reached, allowing subjects to acclimate in this scanning environment. We then corrected the other images for timing differences (slice 37 used as reference) and head motion, determining translation (mm) and rotation (degrees) by six parameters (three each, translation and rotation). There were no group‐wise exclusions due to head motion beyond 2 mm of displacement or 2° of rotation. Subsequently, we spatially normalized images to the Montreal Neurological Institute (MNI) space using EPI templates with resampling voxel sizes of 3 × 3 × 3 mm. All images generated were spatially smoothed using a Gaussian kernel of 6 × 6 × 6 mm, full width at half maximum (FWHM).

The preprocessing of fMRI data was performed with SPM12 software (Wellcome Trust Centre for Neuroimaging, London, UK, http://www.fil.ion.ucl.ac.uk) and REST Toolkit [39 (link)] implemented on a MATLAB platform (MathWorks, Natick, MA). The first two images of task-state and ten images of resting-state data were discarded to allow the magnetization to approach dynamic equilibrium. Functional images were corrected for slice-timing differences and realigned to the median image to correct rigid body motion. Patients with head movement exceeding 3 mm or 3 degrees and healthy subjects exceeding 2 mm or 2 degrees were rejected. The high resolution anatomical image was co-registered with the mean image of the EPI series and then spatially normalized to the MNI template. After applying the normalization parameters to the EPI images, all volumes were resampled into 3 × 3 × 3 mm³. Then the normalized task-state images were smoothed with an 8-mm FWHM isotropic Gaussian kernel, whereas resting-state images were spatially smoothed with a 4-mm FWHM isotropic Gaussian kernel for conventional sake. The linear detrending and band-pass filtering (0.01–0.08 Hz) were performed on the resting-state time series, followed by regressing out the mean time series of global, white matter and cerebrospinal fluid signals, to remove artifacts and reduce physiological noise.

Radiomics analysis was performed by the use of the freely available open-source software “Imaging Biomarker Explorer” (IBEX), which was developed by the University of Texas MD Anderson Cancer Center and utilized the MATLAB platform (MathWorks Inc, Natick, VA). The CT images in the format of DICOM and the GTVp contours in the format DICOMRTSTRUCT were imported into IBEX. We extracted features that represent intensity, shape, and texture of a tumor. The categorization of these features was ranked as first, second, and higher texture features based on the applied method from pixel to pixel^{23 (link)}. More than 3800 radiomic features were considered in this analysis.
From these radiomic features, we removed those with zero variance and those with a correlation above 99% using the training dataset. Previous studies have identified tumor volume and intensity as relevant features for local control and other clinical outcomes^{3 ,25 (link),26 (link),27 (link)}. To further reduce redundancy, we also removed any radiomic features that were highly correlated (> 80%) to the features: F25.ShapeVolume and F29.IntensityDirectGlobalMean. Ultimately these resulted in a remaining 301 radiomic features that were used for the proximity computation³ .

Source analysis was implemented with a toolbox named SPM12,² based on the MATLAB platform (MathWorks Inc., United States). The procedures were as follows: (1) EEGLAB data were converted to the SPM12-readable format; (2) the EEG sensor positions were linked to the coordinate system of MRI (MNI coordinates), which is a process known as co-registration; (3) using the Greedy search-based algorithm, an inverse reconstruction of the scalp electrode signals into the three-dimensional (3D) brain source signals was carried out; and (4) a factorial design was used to investigate the main effects and the group × condition interaction. Please refer to our previous work for more details (Gao et al., 2017 (link)).

Data preprocessing was performed using DPABI software (http://rfmri.org/dpabi) based on Statistical Parametric Mapping 12 (SPM12, http://www.fil.ion.ucl.ac.uk/spm/) on the MATLAB platform (MathWorks, Inc., Natick, MA). Due to signal equilibration, the first ten volumes were discarded. This was followed by slice timing and realignment with an appraisal of voxel-specific head motion. The head motion parameters were determined for each participant; those with movements above 3 mm translation and 3° rotation were excluded from further examination. A total of four participants were excluded due to excessive head movements. Next, functional scans were registered using T1 images and normalized to the Montreal Neurological Institute (MNI) template using DARTEL (Ashburner, 2007 (link)) at a resolution of 3 × 3 × 3 mm³. In total, seven participants were excluded due to low-quality image registration. Functional data were spatially smoothed using a 4-mm fullwidth half maximum (FWHM) Gaussian kernel to increase the signal-to-noise ratio. The signal was band-pass filtered (0.01–0.1 Hz). Finally, the nuisance signals (24 motion parameters, cerebrospinal fluid, and white matter signals) were removed from the time course of each voxel (Behzadi et al., 2007 (link)). We did not regress out the global signal to keep any additional information (Liu et al., 2017 (link)).