Ten pediatric patients at the Department of Pediatric Neurosurgery, Beijing Neurosurgical Institute and Beijing Tiantan Hospital were enrolled in the study from January 2020 to June 2021. The inclusion criteria for the study were: (1) Patients between 7 and 14 years old, (2) suspected of having low-grade brain tumors (localized tumors) according to preoperative MRI and CT, and (3) with no history of previous cerebral or systemic diseases. The exclusion criteria were: (1) Patients whose pathological results indicated a high-grade tumor (WHO classification III–IV) (9), and for whom (2) postoperative adjuvant treatments were demanded.

The surgery was performed by Dr. Jian Gong (Director of the Department of Pediatric Neurosurgery). Given the benign nature of the tumors, the aim of the surgery was gross total resection. Demographic information was recorded, and cognitive assessments (CNS Vital Signs, CNS VS), diffusion tensor imaging (DTI), and resting-state functional MRI (rs-fMRI) were conducted. All procedures were carried out in accordance with the relevant guidelines and regulations (the Declaration of Helsinki). The study was reviewed and approved by the Beijing Tiantan Hospital Institutional Review Board (KY2021-100-02). The patients’ guardians provided written informed consent for their participation in the study.

Patients’ cognitive function was measured before the surgery (S0), in the first month after the surgery (S1), and 3–6 months after the surgery (S2). All tests were conducted by one of the authors (Xueyi Guan, a clinician in the Department of Pediatric Neurosurgery), who is responsible for conducting cognitive assessments of all patients in the department and has experience using CNS VS in over 100 pediatric patients. The CNS VS is a computer-administered neuropsychological assessment tool. It provides age-adjusted standard scores for 15 domains (10). These 15 domains include Composite Memory, Verbal Memory, Visual Memory, Psychomotor Speed, Complex Attention, Cognitive Flexibility, Executive Function, Social Acuity, Reasoning, Working Memory, Sustained Attention, Simple Attention, Reaction Time (RT), Processing Speed (PS), and Motor Speed (MotSp). Among them, MotSp was measured with the finger-tapping test. PS was assessed with the symbol digit coding test (SDCT). RT was measured with the Stroop test. The standard score for each domain had a mean of 100 and a standard deviation of 15. The test-retest correlation was reliable across the tested range of 3–156 days, which was suitable for a rapid perioperative assessment.

The data were analyzed with SPSS 17.0 (Chicago, IL, USA). Statistical comparisons of the means between the follow-up periods were carried out with paired Student’s t-test. A p-value<0.05 was considered significant. The Pearson product-moment correlation coefficient (r) was calculated as the strength of the linear association between domains and imaging metrics. An r value in the range 0.7–1.0, 0.5–0.7, 0.3–0.5, or <0.3 in the correlation analysis was regarded as “high correlation,” “moderate correlation,” “low correlation,” or “negligible/no correlation,” respectively (11). A p-value<0.05 was considered statistically significant. The results were displayed using “ggplot” in R.

MRI was acquired on a 3T scanner (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany) with a 20-channel head/neck coil. DTI was acquired using a single-shot spin-echo echo-planar image (SE-EPI) sequence. Diffusion gradients were applied along 64 directions using b-values of 0 and 3000 s/mm2. DWI data were obtained using the following parameters: (TR=2500 ms, TE=70 ms, flipangle=90∘, slices=68, % field of view (FOV)=90.6%, voxel size=2 mm isotropic, and no intersection gap). The imaging protocol included T1 weighted structure imaging with a magnetization prepared rapid acquisition gradient echo (MPRAGE) sequence (TR=1560 ms, TE=1.65 ms, flipangle=8∘, slices=176, % FOV=100%, voxel size=1 mm isotropic, and no intersection gap). The sections were approximately parallel to the anterior commissure-posterior commissure line.

The DTI data were analyzed using the standard procedure of the PANDA software (version 1.3.0) developed by Cui et al. (12), running in MATLAB R2016b version 9.1.0 (MathWorks, Inc., Natick, MA, USA). The processing procedures included skull stripping, a correction of the eddy current distortions, and building diffusion tensor models. The individual fractional anisotropy (FA) images of the native space were first registered to the FA template (FMRIB58_FA template) in the Montreal Neurological Institute (MNI) space, and then the resultant warping transformations were applied to resample the diffusion metrics into the MNI space of 2×2×2 mm (13). The diffusion metrics, including FA, mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD), were extracted for 20 white matter (WM) tracts identified from the WM probtract atlas provided by Hua et al. (14). The 20 white matter tracts include splenium, body and genu of corpus callosum, anterior, superior and posterior corona radiata, cingulum cortex, external capsule, anterior limb, posterior limb and retrolenticular part of internal capsule, cerebral peduncle, posterior thalamic radiation, tapetum, and cerebellar white matter, uncinate fasciculus, superior longitudinal fasciculus, sagittal stratum, and fronto-occipital fasciculus, and cerebellar peduncle. The normalized images were smoothed using an isotropic Gaussian kernel (6-mm full width at the half maximum).

The Tract-based spatial statistics (TBSS) (15) of the DTI metrics was performed between S1 versus S0 and S2 versus S1 using a general linear model through FSL and PANDA. A 5,000 repetition permutation test was conducted. Significant clusters were corrected using the threshold-free cluster enhancement method (TFCE, p<0.05). The results were visualized using FSLeyes (16). Correlations between the regional average values of the WM tracts and the cognitive data were calculated.

Patients’ rs-fMRI was performed at the same timepoints. Resting-state sequences were acquired with an echo-planar imaging (EPI) sequence. The scan parameters for the EPI sequence with a simultaneous multiscale acceleration technique were TR=2000 ms, TE=35 ms, slices=69, SMS=3, % FOV=100%, voxel size=2.2 mm isotropic, volumes=240, and no intersection gap. The patients were instructed to remain seated with their eyes closed. No sedation was applied during the examination. The data were analyzed using the CONN toolbox 20b (17) running on MATLAB R2016b version 9.1.0 (MathWorks, Inc., Natick, MA, USA).

In the CONN toolbox, the “default preprocessing pipeline for volume-based analyses” was used for rs-fMRI preprocessing. First, a slice-timing correction was performed. The slices were acquired in an ascending, interleaved order. The head motion, global brain signal, white matter signal, and cerebrospinal fluid signal were regressed out from the time course of the rs-fMRI.A tumor mask for each structural image was drawn by a neurosurgen. The damaged portion of the brain is masked during calculation of normalization parameters. The images were then normalized to a MNI template, and the normalized images were resliced with a target resolution of 2 mm. The normalized fMRI images were then smoothed (an 8 mm, full-width, half-maximum Gaussian kernel). The motion outlier threshold of the artifact detection tool was set at the 95^th percentiles in a normative sample with a motion threshold of 0.9 mm. An ART-based identification was applied, and acquisitions with displacement above the threshold were removed (18). A brain masking process was applied for a voxel-level analysis. A band-pass filter from 0.01 to 0.1 Hz was used to eliminate the physiological high-frequency cardiac and respiratory noise low-frequency drifts. Quality assurance plots were visually inspected to ensure the images were properly co-registered and transformed into the MNI space. Images, in which the movement in any direction exceeded either 3 mm translation or 3∘ rotation, were excluded from the study. Images of two patients (the S1 images of subjects 8 and 9) were excluded due to massive head movements.

Following the preprocessing steps of rs-fMRI, seed-based functional connectivity (FC) maps were generated for each participant. Regions of interests (ROIs) were defined using the default atlas in the CONN toolbox. A total of 164 ROIs were used as the seeds. Among them, 132 ROIs were atlases of cortical and subcortical areas from the library Harvard-Oxford atlas and the automated anatomical labeling (AAL) atlas, whereas 32 ROIs were atlases of networks (19). FC pairs were obtained with 164 ROIs. Because the tumor mass effect and supplanted surgical residual cavity after the surgery could result in flawed co-registrations, we chose ROIs in the basal ganglia (the average value of the bilateral caudates, putamina, and pallidi) as seed points for whole-brain FC analysis. We investigated the FC difference among the S0, S1, and S2 timepoints using an unpaired t-test on the Fisher z-transformed score FC. Cluster-level inference was based on the random field theory, and a correction for multiple comparisons across the brain was conducted by controlling the false discovery rate (FDR) at both the voxel (p-unc<0.001) and cluster (p-FDR<0.05) levels using CONN’s implementation of the Benjamini-Hochberg algorithm (20). Correlations between the z-transformed score of FC in the basal ganglia and frontal lobes (18 ROIs including bilateral caudate, bilateral putamen, bilateral pallidum, and bilateral frontal pole, bilateral insular cortex, bilateral superior frontal gyrus, bilateral middle frontal gyrus, and bilateral inferior frontal gyrus, pars triangularis, and bilateral inferior frontal gyrus, pars opercularis) and cognitive domains in the CNS VS were calculated. FDR was applied in the correction for multiple comparisons.

In the S0 baseline assessment, the domains most prominently decreased compared to the normative data were SocAcu (77.0±19.3, t=−3.55, p<0.01) and RT (84.3±16.7, t=−2.75, p=0.02). MotSp was within the normal range (96.6±19.3, t=−3.55, p=0.64). In the S1 assessment, MotSp (82.4±18.7, t=4.49, p<0.01) and RT (70.3±19.4, t=2.58, p=0.03) were significantly decreased. In the S2 assessment, MotSp (91.2±18.3, t=2.33, p=0.05) was significantly recovered, while RT (75.2±20.6, t=2.33, p=0.495) was slightly improved. The changes in MotSp and RT are elucidated in Figure 1.

DTI metrics comparisons are displayed in Figure 2. There were no significant changes in the diffusion metrics of the WM tracts between S0 and S1. Compared to S1, the FA was significantly reduced in S2 in nine out of 20 WM tracts, including the bilateral corticospinal tracts, bilateral cingula (cingulate gyri), bilateral internal capsules, bilateral external capsules, and left superior longitudinal fasciculus (TFCE-p<0.05). Further, the AD was significantly reduced in the left superior-frontal blade, left parieto-temporal blade, left internal capsule, left external capsule, and left corticospinal tract (TFCE-p<0.05). There were no significant changes in the RD and MD.

Compared to S0, S1 showed decreased FC in the bilateral frontal lobes between the bilateral caudates (Figure 3), putamina (Figure 4), and pallidi (Figure 5) (T>3.25, voxel threshold p-unc<0.01, cluster threshold p-FDR<0.05). Compared to S1, S2 demonstrated increased FC in the bilateral frontal lobes between the bilateral caudates (Figure 3), putamina (Figure 4), and pallidi (Figure 5). There were no significant differences between S0 and S2.

The analysis of the DTI metrics in 20 WM tracts (Table 2) revealed a significant positive correlation between the FA values in the right cingulum (hippocampus) and the RT scores (r=0.36, p-FDR<0.05). There was a significant positive correlation between the AD values in the left cingulum (cingulate gyrus) and the MotSp scores (r=0.38, p-FDR<0.05) as well as a significant positive correlation between the AD values in the right cingulum (hippocampus) and the RT scores (r=0.50, p-FDR<0.05).

The analysis of the FC z-transformed score between 18 ROIs of the basal ganglia-frontal lobes (Table 3) and 12 ROIs of the basal ganglia-motosensory cortex (Table 4) revealed that the RT was primarily correlated with the FC in the caudates, including the caudate-bilateral insular cortices (IC) (p-FDR<0.05), caudate-bilateral inferior frontal gyri (IFG) (p-FDR<0.05), and caudate-superior frontal gyri (SFG) (p-FDR<0.05). The RT was also positively correlated with the FC in the putamen-left middle frontal gyrus (MFG) (r=0.34, p-FDR<0.05) and putamen-right IFG (pars triangularis) (r=0.26, p-FDR<0.05).

Compared to the RT, the MotSp showed a widespread negative association with the connectivity in the basal ganglia. In the FC of the basal ganglia-motosensory cortex (Table 4), MotSp was negatively correlated with the FC in the caudate-left SMA (r=−0.31, p-FDR<0.05) and putamen-right SMA (r=−0.26, p-FDR<0.05). In the basal ganglia-frontal lobe (Table 3), it was also negatively correlated with the FC in the basal ganglia-IFG (p-FDR<0.05), putamen-frontal pole (p-FDR<0.05), putamen-right SFG (p-FDR<0.05), and pallidum-right SFG (p-FDR<0.05).

The information processing speed is an important predictor of academic achievement (22). An assessment of the information processing speed should include motor and non-motor domains. In the CNS VS battery, MotSp represents a simple motor reaction, while PS and RT involve both motor and cognitive elements of the reaction (23). In S0, RT was significantly impaired. Previous research demonstrated that RT is sensitive to frontal lobe damage (24) and showed an association of structural MRI with the myelin integrity of the frontal lobes (25). A resting-state EEG study also indicated that prefrontal electrical activity predicted the RT (26). Our findings in pediatric patients with LGFLT were consistent with previous reports. However, we found that the MotSp remained at normative levels in S0. This may have been because the simple voluntary motor speed of the hands is related primarily to the SMA (27). The SMA is located on the dorsomedial aspect of the superior frontal gyrus, with the prefrontal sulcus and primary motor cortex as the posterior border. It is involved in the “Go and Stop” process of motor movement (28). Since in our study the tumors of only two patients affected the SMA, the MotSp was not significantly decreased in S0.

Surprisingly, the PS was not impaired at the baseline and remained stable after the surgery (S0=103.7±18.1 versus S1=100.0±9.7, p=0.470). The PS was measured using the SDCT, which is associated with the prefrontal cortex, hippocampus, and superior temporal gyrus (29). The results indicated that the SDCT might not have been sensitive to the insult from the frontal lobe tumor.

Because brain tumors can lead to functional retention within the tumor and functional reorganization in the adjacent cortex (30), an LGFLT resection can further damage the RT. This theory corresponds with the significant RT decrease, which was observed in S1>S0. Notably, brain surgery per se can cause edema in the peri-tumoral tissue and may affect larger cortex areas than the tumor invasion (31). In addition, biochemical (e.g., intracranial pressure and neurotransmitter release) and genomic alternations (e.g., protein synthesis) after brain surgery can lead to the inhibition of distant brain regions (32). In the rs-fMRI analysis, we observed decreased FC between the bilateral SMA (superior frontal gyrus) and bilateral caudate/putamen/pallidum (Figures 2–4). This can explain the corresponding MotSp decrease in S1.

We found a dramatic and rapid recovery of the MotSp in S2, which was approaching the baseline level. The RT also slightly recovered. This phenomenon has been reported in adult patients with brain tumors, in whom the immediate postoperative motor or language function would worsen transiently and recover within 1–3 months (33). The mechanism of the rapid recovery of processing speed is yet to be established. Below, we discuss the basal ganglia-frontal lobes connectivity changes during S0–S1–S2.

Basal ganglia interact closely with the frontal cortex. The frontal-basal ganglia circuit plays a critical role in attention and working memory (34). The basal ganglia surround the diencephalon and are made up of five subcortical nuclei: pallidum, caudate, putamen, substantia nigra, and the subthalamic nucleus of Luys. The caudate and putamen (i.e., the neo-striatum) are the input nuclei of the basal ganglia, which receive afferent input from the motor cortex of the frontal lobes through the cortico-striatal projections. The pallidi are the major output nuclei of the basal ganglia. They are involved with gating incoming sensory input to higher motor areas to coordinate behavioral responses (35). Parallel loops of the neostriatum connect the frontal lobe and the thalamus. Basal ganglia act as a “brake release” for motor actions on the frontal cortex (36).

Figures 3–5 illustrate the FC changes in the three pairs of core nuclei in the basal ganglia. In S1>S0, the FC decreased in the bilateral frontal lobes. This can be explained by trauma from the tumor resection and postoperative edema. In S2>S1, FC recovery was observed in the frontal lobes, where the recovery “shifted posteriorly” from the injured site—the cortex, which showed decreased FC in S1>S0. This phenomenon, consisting of an initially decreased FC in the peritumoral cortex areas and subsequently increased FC in the neighboring cortex areas, may represent the typical “map expansion” neuroplasticity described by Grafman (37). The term ”map expansion” describes that the pool of neurons that respond to behaviorally relevant stimuli expands, to accomplish the task more efficiently (38).

The associations between these “map expansion” cortex areas and cognitive data were verified in a correlation analysis. Among the ROI-ROI with significantly increased FC in S2>S1 (Figure 3 and Table 3), the FC in the caudates-right IFG was positively correlated to the RT (r=0.26-0.33) in the correlation analysis. We also found that the RT was positively correlated to the majority of ROI-ROI connectivity related to the caudate. This indicated that new neural circuits were emerging, as the original cortical areas had been destroyed by the tumor and surgical trauma. This finding provided evidence that “map expansion” neuroplasticity can happen as early as 3–6 months after frontal tumor surgery in children.

We did not find any positive relationship between the MotSp and connectivity in the frontal lobes in the ROI-ROI analysis. As alluded to earlier, MotSp is associated with the SMA activities in healthy subjects. In our study, only two patients had tumors offending the SMA (subjects No. 2 and 5). We supposed that the SMA was mainly affected by preoperative tumor edema (in the S0) and transient postoperative edema (in the S1). We compared the MotSp change between the patients who had significant postoperative edema in the S1 (larger edema volume in the S1 than that of the S0) and those who had not. The result showed that patients without edema (subject No. 1,2,3,6,7,8 and 9) gained a larger improvement in S2 than those having edema (subject No. 4,5 and 10) (13.0±10.5 vs −1.0±10.4, Mann-Whitney test Z=−1.943, two-tailed p=0.052). This indicated that postoperative edema plays a negative effect on MotSp improvement. As the edema subsided, the MotSp rapidly restored to the preoperative level in the S2. The restoration may primarily rely on the recovery of the original neural circuit (basal ganglia-SMA), which had been interrupted but not destroyed.

In the correlation analysis, however, our results showed that the MotSp negatively correlated with the FC of the caudate-SMA (r=−0.31) and the putamen-SMA (r=−0.26) (Table 4). The underlying mechanism of this counterintuitive finding is less clear. One explanation is that the frontal lobe remapping rewrote the functional architecture of the SMA (39). The SMA became more “crowded,” and the activation in the basal ganglia-SMA circuit now represented new allocated functions (attention and processing speed) rather than the motor speed of the patients.

DTI metrics reportedly reflect the integrity of WM tracts and have been applied as neuroimaging biomarkers in a range of cerebral diseases (40). Dennis et al. have investigated WM tract integrity in over 500 children and adolescents with traumatic brain injury (TBI) during acute/subacute (<2 months), post-acute (2–6 months), and chronic post-injury periods (≥6 months). They found significantly lower FA and higher MD in those with TBI compared to the controls during all three periods (41). In our DTI analysis, LGFLT resection led to extensively decreased FA and AD in WH tracts from 0 to 6 months after surgery. The major affected WM tracts were the corticospinal tracts, cingulum, internal capsule, and external capsule (Figure 2). Our results were similar with those of Dennis et al. (41), indicating that tumor resection is a special form of TBI.

FA has been interpreted as a proxy for myelin integrity. Most TBI studies have agreed that decreases in the FA and AD parallel shearing and disintegrated WM structures (42), while AD is less sensitive to pathologic changes from the acute to the chronic stage (43). In our study, the decreased FA and AD were not associated with the rebound of cognitive measurements in S2 (Figure 2 and Table 2). Although a correlation analysis showed that the RT was positively correlated to the FA and AD values in the right hippocampus cingulum (r=0.36-0.50) (Table 2), the corresponding WM tract did not show significant changes in either S1>S0 or S2>S1 (Figure 2). It has been previously demonstrated that the anterior cingulate cortex is responsible for executive functions and the processing speed, whereas the hippocampus cingulum is related to memory (44). However, in pediatric patients with LGFLT, the RT was significantly associated with the hippocampus cingulum rather than with the cingulate cortex. It is plausible that preoperative reorganization of the WM tracts occurred due to tumor disruption or compression in the frontal lobes. While interrupted by a cerebral injury (e.g. tumor invasion or brain surgery), the neuroanatomical changes may not necessarily translate to functional presentations in young brains (45). The above results suggest that DTI is not sensitive enough to predict cognitive changes after surgery in pediatric patients with LGFLT.

The major limitation of our study was that the sample size was relatively small. The tumor size and affected cerebral cortex areas varied, which precluded the direct assessment of rs-fMRI or DTI metrics in the specific cortex areas of the frontal lobes.