*2.5. Statistical Analysis of Behavioral Data*

All trials with a response time (RT) < 150 ms were eliminated. The RT analyses were limited to proper trials. Both RT and error rate (ER) data for the task behavioral performance were analyzed using ANOVAs, with "task condition" (OG, DI, TS) and "group" (mTBI patients, HCs) factors. The significant main effects and interactions were further explored by *post hoc t* tests using Bonferroni's correction. A *p* < 0.05 was used as statistically significant for all the behavior analyses.

#### *2.6. fMRI Data Preprocessing and Statistical Analysis*

Using FEAT in FSL v5.0, we analyzed whole-brain voxel-wise activation on the fMRI scans of the rule-based task-switching experimental paradigm [30]. We performed preprocessing of all individual fMRI runs using the following procedures: both the first and last 2 volumes were deleted, underwent motion correction (MCFLIRT) [31], brain extraction, spatial smoothing (6-mm full-width at half maximum kernel), and high-pass temporal filtering (0.01-Hz cutoff). In addition, linear registration (FLIRT) was performed in T1, fMRI, and MNI152 standard space. The degree of head motion from all participants was

quantified, and according to the head motion criteria (translational or rotational motion parameters less than 2.0 mm or 2.0◦), none of the participants were excluded (the mean translation was 0.08 ± 0.02 mm, and rotation was 0.04 ± 0.01◦).

First-level (within-run) general linear model analyses in native fMRI space were conducted with FILM prewhitening, with 3 separate regressors (the onset of all stimuli for OG condition, DI condition and TS condition). Each one of them convolved with a doublegamma hemodynamic response function with the application of temporal filtering [32]. First-level contrast was set up to create voxel-wise contrast of parameter estimate maps of activation in different task conditions. The maps were then used for second-level (withinsubject) analysis in the 2 runs. The maps were converted to MNI152 space and fixed effects analyses were performed with 3 contrasts to identify OG, DI or TS conditions. The resultant maps for each contrast were then used for third-level (group-level) FLAME 1 + 2 mixedeffects analyses which proved to have high statistic values at a small proportion of voxels in a relatively small sample size (threshold: cluster-based *p* < 0.05; whole-brain familywise-error corrected Z > 2.3) [33], with 3 contrasts: OG, DI or TS activation. We employed Student's *t* test to analyze the effect of the group on brain activation of different task conditions using mTBI patients and HCs as factors.
