Altered white matter structural network in self-limited epilepsy with centrotemporal spikes

Introduction

Epilepsy, which affects more than 70 million people worldwide, is one of the most common neurologic diseases. It is characterized by neural network impairments caused by abnormally synchronous or excessive activity of neurons in the brain (1). Self-limited epilepsy with centrotemporal spikes (SeLECTS), the most common type of idiopathic childhood epilepsy (accounting for 15.7% of all cases), is typically characterized by onset during early childhood (2,3). Epileptic seizures have a significant impact on the development of children. Early diagnosis of epilepsy is of great significance. Some children with SeLECTS show cognitive dysfunction that cannot be restored to a normal level even in adulthood (4). The heterogeneity of disease progression often complicates early diagnosis of the cognitive impairment associated with SeLECTS. Language dysfunction, therefore, remains a prominent problem in children with SeLECTS (5). Neuropsychological evaluations can be used to detect cognitive disorders, but assessing the brain changes is of vital importance to explore the pathophysiological mechanisms, monitor disease progression, and improve the patients’ quality of life. At present, epilepsy is regarded as a disorder of neural networks, which participate in the generation, maintenance, and propagation of seizures (6,7). White matter (WM) composes the interactive network of the brain and plays a role of transmission and feedback (8). It is reported that the WM network also plays an important role in the propagation and maintenance of epilepsy (9). The epileptogenic processes of SeLECTS might have an effect on the brain development and may interfere with the normal maturation processes (10,11). Seizure activity in SeLECTS patients may result in various types of WM alterations, which has been related to cognitive development and intelligence (12).

As an advanced magnetic resonance imaging (MRI) modality, diffusion tensor imaging (DTI) enables non-invasive quantification of water molecular diffusion patterns in biological tissues (13). Therefore, DTI images can be used to observe the WM microstructure, which cannot be visualized by conventional MRI sequences, promising earlier and more sensitive detection of lesions, and supporting analysis of the structural network in epilepsy. The nodes correspond to anatomical areas and edges stand for the WM connections between pairs of nodes in the WM network (14). A graph-theoretic approach might supply valuable information between integration and segregation processes that clarify brain changes by a range of metrics (15). In the brain network analysis, small-world properties, global efficiency (E_g), local efficiency (E_loc), clustering coefficient (C_i), shortest path length (L_p), and nodal efficiency (E_nodal) may supply several advantages for assessing the properties of WM structural networks from both technical and conceptual perspectives (16). A balance between global integration and local specification of information transmission is the characteristic of small-worldness, which is important for maintaining efficient information integration and segregation (17). E_g is used to evaluate long range information transmission referring to integration capacity; however, E_loc is considered to assess short range information transfer referring to segregation capacity (18). C_i is a way of functional segregation quantifying the presence of locally connected groups, which suggest segregated neural processing (19). L_p evaluates the effective communication between different brain areas to reflect network integration (19). E_nodal is a measurement for how well a specific region is integrated within the network (20). By testing changes in the balance between integration and segregation of information, these metrics may yield new insights in understanding the neuropathological changes in the brain. This evidence aids in the diagnosis and classification of SeLECTS and offers novel radiological insights into its cognitive manifestations. However, to the best of our knowledge, there no prior study has focused on the WM network in SeLECTS patients. Additionally, the brain’s WM network is involved in highly integrated cognitive processes (21). Owing to the reasons above, we concentrated on network metrics that reveal integration and segregation.

Many studies have demonstrated structural connectivity abnormalities in children and adults with epilepsy. These studies indicate that the brain structural abnormalities in epilepsy may underlie impaired network-level information transfer. Ai et al. (22) and Liu et al. (23) discovered decreased E_g and widespread reductions in E_nodal in frontal, temporal, limbic, and occipital lobe regions, and insula in temporal lobe epilepsy. Increased path length was also found (23). Xue et al. (24) reported the disruption of the WM networks in childhood absence epilepsy. The E_g/E_loc, C_i were decreased, but the path length was increased in childhood absence epilepsy (24). Liu et al. (25) observed changed gray matter volume in SeLECTS associated with cognitive impairments and behavioral problems in SeLECTS patients. Li et al. (26) revealed extensive changes in cortical thickness, sulcal depth, and cortical gyrification in SeLECTS. The active phase of SeLECTS occurs in school age, a period during which the brain experiences the complicated process of WM maturation, wherein disproportionate rates of myelination through the brain alter the structural network connectivity (27,28). Additionally, the epileptogenic processes will affect synaptic pruning and exuberance, finally leading to the axonal demyelination, reduced axonal density and replacement of axons by glial cells in SeLECTS patients (29). The brain structural network can detect the subtle neurodevelopmental anomalies in SeLECTS. Given the role of WM integrity in neural synchronization, we conjectured that SeLECTS pathophysiology involves maladaptive topological reconfiguration of WM networks. Furthermore, interplay likely exists between WM network reorganization and disease progression.

Therefore, in this study, we employed a structural network and quantitatively analyzed network topology parameters to explore local and global network characteristics in SeLECTS in more detail, and investigated their relationship with clinical features, with the aim of identifying valuable imaging characteristics and specific cerebral areas for clinical assessment. In addition, this study also explored the role of the WM structural network in the classification between SeLECTS and typically developing (TD) children, in order to identify neuroimaging biomarkers for the classification of SeLECTS according to WM structural connectome. We present this article in accordance with the STROBE reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1131/rc).

Methods

Participants

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of The First Affiliated Hospital of Xi’an Jiaotong University (No. XJTU1AF2023LSL-011). Written informed consent was provided by the parents or legal guardians of all participants prior to enrollment. Assent was also obtained from the children themselves when applicable. All participants were right-handed Han Chinese individuals. Pediatric patients with SeLECTS and age- and sex-matched TD children were recruited from the outpatient clinics and inpatient wards of the Department of Pediatrics at The First Affiliated Hospital of Xi’an Jiaotong University between February 2017 and December 2020. Patients meeting the following eligibility criteria were enrolled in the SeLECTS group: (I) identified as SeLECTS based on the International League Against Epilepsy criteria (30); (II) completed MRI examination; and (III) showing no other neurological disorders. The exclusion criteria were as follows: (I) abnormalities on MRI; (II) incomplete clinical information including age, sex, handedness, seizure duration (the length of a single seizure), seizure frequency (the number of epileptic seizures within six months), epilepsy duration (the length of time a person has had epilepsy), age of onset, and neuropsychological assessments; and (III) patients with SeLECTS who were not newly diagnosed.

The TD group consisted of children meeting the following inclusion criteria: (I) completion of MRI examination; (II) no abnormalities on MRI; and (III) absence of epilepsy and other types of seizures. Children experiencing neurological disorders, headache, and intracranial infection were excluded. Meanwhile, images with artifacts were also excluded. Comprehensive details regarding the inclusion and exclusion criteria are enumerated in Figure S1.

MRI data acquisition

MRI scans were performed uniformly across participants using a 3.0-T scanner system (Signa HDxt, GE Healthcare, Milwaukee, WI, USA) equipped with an 8-channel head coil at our hospital. Each of the participants wore ear plugs, with their head fixed with comfortable foam pads. Three-dimensional fast-spoiled gradient-recalled echo T1-weighted imaging, fast spin echo T2-weighted imaging, and single-shot echo-planar DTI were performed. The parameters of the MRI sequences were as follows: (I) T1-weighted imaging: repetition time (TR)/echo time (TE) =10/4.7 ms; slice thickness =1 mm without a gap; field of view (FOV) =240×240 mm²; and matrix size =256×256; (II) T2-weighted imaging: TR/TE =4,200/120 ms; slice thickness =4 mm without a gap; FOV =240×240 mm²; and matrix size =320×320; and (III) DTI: 30 gradient directions; b values =0 and 1,000 s/mm²; TR/TE =11,000/69.5 ms; slice thickness =4 mm without a gap; FOV =240×240 mm²; and matrix size =128×128.

Neuropsychological assessments

Cognitive functioning was assessed using the Wechsler Intelligence Scale for Children-Fifth Edition, Chinese version. The test yields age-standardized deviation intelligence quotient (IQ) scores, including verbal IQ, full-scale IQ, and practical IQ, which were the metrics used in the present study. All cognitive assessments were conducted on the same day as the MRI examination in a quiet, dedicated testing room. The tests were administered individually by a trained research assistant. Examiners were not blinded to participants’ age and sex, as this information is necessary for appropriate test administration and rapport building. However, they were unaware of the specific study hypotheses regarding the relationship between brain network measures and cognitive performance, thereby minimizing potential expectancy biases.

Data preprocessing and WM structural network construction

Figure 1 illustrates the data analysis for constructing the WM structural network. DTI data processing was carried out with the Pipeline for Analyzing braiN Diffusion imAges toolbox, which relies on the FMRIB Software Library (FSL; https://fsl.Fmrib.Ox.Ac.uk/fsl/fslwiki) (31). The processing steps included skull-stripping, DTI image format conversion (from DICOM to NIFTI images), realignment, eddy-current and head-motion correction, fractional anisotropy (FA) calculation, and diffusion tensor tractography. Deterministic tractography was calculated using the Fiber Assignment by Continuous Tracking algorithm, with a FA value threshold set at 0.2–1 and a turning angle threshold of 45° (32).

Figure 1 WM structural network analysis processing. (A) Three-dimensional fast spoiled gradient-recalled echo T1WI of an individual example participant. (B) In all, 90 cortical and subcortical regions of interest, covering the whole brain. (C) DTI. (D) Whole brain tractography. (E) WM structural connectomes were constructed. DTI, diffusion tensor imaging; T1WI, T1-weighted imaging; WM, white matter.

WM connectomes were reconstructed for both SeLECTS and TD children using an automated anatomical labeling template. The brain was divided into 90 cortical and subcortical regions (45 for each hemisphere) (Table S1). Graph-theoretic analysis was conducted to detect differences in the WM structure network parameters between the SeLECTS and TD groups. Global and regional topological metrics of WM networks were computed using GRETNA software (http://www.nitrc.org/projects/gretna/) (33). A binary adjacency matrix was first defined at the group level: an edge between two regions was included if at least one fiber tract was detected in ≥60% of all participants across all three waves (34). For each individual and each wave, the weight of each preserved edge was then assigned as the mean FA calculated across all streamlines connecting the two regions. Streamline counts were not used as edge weights. Global efficiency and local efficiency were computed on the weighted undirected graphs using the Brain Connectivity Toolbox (19). For each individual brain network at each density threshold, we generated 100 degree-preserving random networks using the Maslov-Sneppen rewiring algorithm. This algorithm randomizes the network topology while strictly preserving the degree distribution and connectedness of the original network.

Small-world properties

Small-world networks [sigma (σ)] are defined by a balance between global integration and local specification of information transmission (35). The network exhibited small-world property if σ>1. This property was calculated as follows:

σ=γλ=CCrandLLrand[1]

Where Crand and Lrand represent the mean weighted C_iand the mean weighted characteristic path length of random networks.

Global parameters

Global efficiency

E_g indicates the effectiveness of the network transporting information in parallel (23). The formula is as follows:

Eg=1N(N−1)∑i,j,i≠jn1dij[2]

N indicates the quantity of nodes, whereas d_ij corresponds to the distance between nodes i and j.

Local efficiency

E_locrepresents the E_g of each sub-network G_i. E_locindicates the network fault tolerance capacity (23). Fault tolerance capacity is the network’s resistance ability to function efficiently even after the weakening or removal of some nodes. The formula is as follows:

Eloc(i)=Eg(Gi)[3]

Clustering coefficient

The C_i represents the proportion of edges between other nodes directly linked to a given node in the network, relative to the maximum possible number of such edges. It serves to quantify the level of network grouping or clustering. C_ican be represented by the following formula:

Ci=Ei12ki(ki−1)[4]

E_i measures the actual edges between node i’s neighboring nodes, k_i reflects how many connections node I has, and k_i(k_i− 1)/2 computes the potential edge total if all its neighbors were fully interconnected.

Shortest path length

L_p, corresponds to the most efficient route for data transmission from one node to another within the network. The formula is as follows:

Lp=1n∑i∈NLi=1n∑i∈N∑i∈N,j≠idijn−1[5]

Where dij indicates the L_pbetween the nodes of i and j.

Regional parameters

Nodal efficiency

E_nodal is regarded as the reciprocal of the harmonic mean of the L_p between node i and all other nodes in the network (16). It reflects the efficiency of a particular cortical region in the network. The formula is as follows:

Enodal(i)=1(n−1)∑j∈N1Li,j[6]

Where L_i,j represents the L_p between node i and node j.

Statistical analysis

Normality of all continuous variables was assessed using visual inspection of Q-Q plots and the Shapiro-Wilk test. Sex comparisons between the two groups were performed using exact statistical test. Other demographic information in the SeLECTS and TD groups was compared using two-sample t-test or Mann-Whitney U tests. Differences in the topological metrics of the structural networks were analyzed by two-sample t-test. Spearman correlation analyses were performed to investigate the relationships between network topological and clinical features, namely seizure duration, seizure frequency, epilepsy duration, age of onset, and IQ scores (full-scale, practical, and verbal). For each correlation, we report the correlation coefficient and its associated P value. The magnitude of between-group differences was quantified using Cohen’s d. All statistical analyses were conducted utilizing the software SPSS 23.0 (IBM Corp., Armonk, NY, USA). A statistical significance was defined as P<0.05. Multiple-comparison correction was performed using the false discovery rate (FDR) at q<0.05.

Classification model according to WM structural network matrix

To rigorously evaluate model performance and mitigate overfitting, we implemented a train-validation-test framework with an inner cross-validation loop. The dataset was first partitioned into an 80% training/validation set and a 20% held-out test set. Within the training/validation set, a 5-fold inner cross-validation loop was employed for hyperparameter optimization via grid search. The support vector machine (SVM) classifier was configured with a radial basis function (RBF) kernel, and hyperparameters (regularization parameter C, kernel coefficient gamma) were tuned to maximize the area under the receiver operating characteristic (ROC) curve (AUC). The final model was retrained on the entire training set using optimal hyperparameters and evaluated exclusively on the untouched 20% test set. This strict separation between model development and evaluation phases provides an unbiased estimate of generalizability. Model performance was quantified using sensitivity (SEN), specificity (SPE), accuracy (ACC), and AUC. To further assess robustness and overfitting risk, we conducted complementary analyses: (I) comparison with alternative classifiers (logistic regression, random forest) using 5-fold cross-validation; and (II) feature stability analysis via recursive feature elimination with cross-validation (RFECV). For linear SVM models, feature importance was derived from absolute coefficient magnitudes; for non-linear kernels, permutation importance was computed. The overfitting risk was quantified as the difference between final training and validation accuracies from learning curves, with gaps >0.15 indicating potential overfitting.

Results

Demographic and clinical characteristics

We selected 32 SeLECTS and 32 matched TD participants through rigorous adherence to eligibility criteria. The detailed demographic information and clinical characteristics are shown in Table 1, which was similar to our previous study (36).

Differences in network characteristics between groups

Small-world properties of structural networks

All participants in the SeLECTS and TD groups showed small-world properties (σ>1). In comparison with the TD group, σ was lower in the SeLECTS group, but the two groups showed no significant difference (P=0.057).

Global parameters of structural networks

Intergroup differences in the global parameters for the WM structural networks are displayed in Figure 2. In comparison with the TD group, the SeLECTS group showed significantly greater L_pand significantly lower E_g (z=−5.05, P<0.001, q<0.001; z=4.58, P<0.001, q<0.001; FDR-corrected). There were no significant distinctions between SeLECTS group and TD group in terms of E_locand C_i(P=0.051).

Figure 2 Histogram of the significant differences in the global network topological properties. *, P<0.05, FDR-corrected. E_g, global efficiency; FDR, false discovery rate; L_p, shortest path length; SeLECTS, self-limited epilepsy with centrotemporal spikes; TD, typically developing.

Regional parameters of structural networks

Nodal efficiency

In contrast to TD group, the SeLECTS group showed lower E_nodal values in the left precentral gyrus (PreCG.L), opercular part of the left inferior frontal gyrus (IFGoperc.L), left Rolandic operculum (ROL.L), and left superior temporal gyrus (STG.L) (z=3.91, P<0.001, q<0.001; z=4.27, P<0.001, q<0.001; z=4.06, P=0.014, q=0.015; z=2.36, P=0.002, q=0.006; FDR-corrected; Figure 3).

Figure 3 Group differences in E_nodal between patients with SeLECTS and TD children. ***, P<0.001. E_nodal, nodal efficiency; IFGoperc, opercular part of inferior frontal gyrus; L, left; PreCG, precentral gyrus; ROL, Rolandic operculum; STG, superior temporal gyrus; SeLECTS, self-limited epilepsy with centrotemporal spikes; TD, typically developing.

Correlations between structural networks and clinical assessments

No significant correlation was observed between the clinical parameters and global network properties. The longer the seizure duration, the lower the E_nodalin ROL.L and STG.L (r=−0.57, P<0.001; r=−0.52, P<0.001). The higher the E_nodal in STG.L, the higher full-scale and verbal IQs (r=0.53, P<0.001; r=0.64, P<0.001) (Figure 4). No significant correlation was found between other network metrics and the results of neuropsychological tests or disease status.

Figure 4 Correlation analysis of E_nodal with seizure duration and the IQ. The E_nodal in the STG.L and ROL.L were negatively correlated with seizure duration (A,B). The E_nodal in the STG.L was positively correlated with the full-scale and verbal IQs (C,D). E_nodal, nodal efficiency; IQ, intelligence quotient; L, left; ROL, Rolandic operculum; STG, superior temporal gyrus.

Classification results according to WM structural network matrix

Model performance on the held-out test set

Employing the train-validation-test framework with an inner cross-validation loop, the SVM classifier with an RBF kernel achieved robust discriminative performance. When evaluated exclusively on the untouched 20% held-out test set, the model demonstrated strong generalization ability. The model reached 78.13% ACC, 75.00% SEN, 81.25% SPE, and 0.82 AUC.

Robustness and overfitting assessment

To ensure the reliability of the primary SVM model, complementary robustness analyses were conducted. Using an independent 5-fold cross-validation on the training/validation set, multiple classifiers yielded consistent performance: logistic regression achieved an accuracy of 76.43%±6.15%, random forest 74.22%±7.33%, and SVM 78.13%±8.25%. The comparable performance across algorithms indicates that the discriminative pattern is inherent to the feature set rather than specific to the SVM model. The convergence trend between training and validation accuracy, with a final gap of 0.12, suggests the model learned generalizable patterns without severe overfitting.

Feature importance and stability

The stability and relevance of the selected biomarkers were confirmed through two complementary analyses. RFECV identified an optimal subset of five features, among which L_p and E_nodalin PreCG. L were the most stable, retained across all validation folds—a finding consistent with the initial univariate analysis. Regarding feature importance, the largest absolute coefficients in linear SVM configurations were assigned to L_p and nodal efficiency in the left superior temporal gyrus. Permutation importance analysis for the final RBF-kernel model confirmed L_p as the most critical predictor, with its shuffling leading to a mean decrease in accuracy of 15.23%.

Discussion

We evaluated and compared brain structural networks between patients with SeLECTS and TD. We also analyzed the association between clinical characteristics and network topological properties. Compared with the TD, the E_g decreased, but the L_p increased in SeLECTS patients. The E_nodal showed differences in specific regions of cerebral between two groups. In addition, the changes in the E_nodal in the STG.L in children with SeLECTS were associated with neuropsychological score. The E_nodal in the STG.L and ROL.L were negatively correlated with the duration of epilepsy. These findings implied that WM structural networks may supply useful information in children with SeLECTS. These findings provide further characterization of structural network alterations in children with SeLECTS and their association with cognitive performance. The observed differences in WM network efficiency may reflect less efficient information transmission and integration.

The human brain is the most complicated neural network. However, to our knowledge, few studies have focused on WM structural networks in patients with SeLECTS. Ostrowski et al. (37) demonstrated significant differences in language modalities in SeLECTS patients. Our previous research focused on the association of glymphatic function with global network properties and cognition in patients with SeLECTS (36). This current study extends this work in two key ways: (I) it investigates both global and regional network efficiency, thereby providing a more comprehensive characterization of structural network alterations; and (II) it examines the direct associations between network measures and clinical variables, which were not assessed previously. This extension allows us to identify region-specific vulnerabilities and their potential clinical relevance in SeLECTS. SeLECTS presents with topological changes, which were associated with duration of epilepsy and neuropsychological score. Among network properties, the small-world network properties and the E_loc/E_g of the network were compared between the two groups in the present study. A normal brain network is characterized by a small-world network with a short path length and high C_i (38). The small-world topology is characterized by a balance between global integration and local segregation in information processing (39). In this study, the WM structural network of all patients showed small-world attributes, indicating that local and global information could be processed simultaneously (17). Epilepsy may affect information segregation and integration in the brain, which can be reflected by the topological properties of brain networks (40). Xiao et al. (41) have indicated alterations in the E_loc, C_i and small-world attributes of functional networks in patients with SeLECTS. By analyzing high-density EEG data, Adebimpe et al. (42) found that the small-world functional topology was disrupted in SeLECTS patients. However, no significant changes in C_i were observed, which was consistent with our study (42). Ji et al. (43) reported that there was no difference in the small-world and C_i attributes of functional network between children with SeLECTS and TD children, which was similar to our research. We supposed that this might be related to the severity of SeLECTS in children, given that the enrolled participants exhibited no abnormality on MRI and had demonstrated neuropsychological scores predominantly within normal limits. Therefore, the small-world property, E_loc and C_i might lack sufficient sensitivity to capture subtle disruptions in brain network organization during the early stage of epilepsy (22).

In our analyses of the global properties, the SeLECTS group showed significantly lower E_g and longer L_p. These changes reflect an abnormal topology network and coincided with the findings of a previous functional MRI study (41). The decreased E_g in children with SeLECTS is suggestive of reduced effectiveness of the network transporting information in parallel (23). An increased L_p indicates that information transmission needs to pass through more links, implying decreased global integration of the brain (44).

Important nodes play crucial roles in the process of information transmission by determining the stability of the network structure and the integration efficiency of neural information (45). The E_nodal reflects the connectivity among nodes in the network and the importance of nodes for communication in the network (16). Careful analysis of the topological properties of these nodes may elucidate the nature of the network. In our study, the regions with decreased E_nodalwere located in the PreCG.L, IFGoperc.L, ROL.L, and STG.L, indicating reduced communication efficiency in these brain areas. All of these regions were close to the epileptic zone in SeLECTS patients. The STG plays an important role in speech perception, interfacing with both language-processing and auditory systems (46). Meanwhile, the inferior frontal gyrus (IFG) was thought to be a representative region for motor speech properties, among others. Thus, structural changes in the IFG.L and STG.L in children with SeLECTS may lead to impaired information transmission through language pathways (47). Since the fiber bundles form a closed loop between the STG and IFG, WM abnormalities in these two areas may affect language processes (47). A previous study also reported changed specialization in the STG.L in patients with SeLECTS, which supported our results (48). The ROL is adjacent to the cortical somatic tongue representation, which plays a major role in speech production and feeding behavior (49). Jiang et al. and Besseling et al. (50,51) observed that patients with SeLECTS showed abnormal functional connectivity related to ROL on resting functional MRI scans. In the present study, the decreased E_nodal observed in the ROL may provide additional evidence to support an ROL cortex abnormality in SeLECTS. Centrotemporal spikes are known to originate from the somatosensory cortex and transmitted to the motor cortex. Oro-pharyngo-laryngeal disorders and facial tonic spasms are characteristics of SeLECTS (52). Therefore, alterations in the E_nodal in PreCG might explain the clinical symptoms in SeLECTS. Thus, the sub-network containing the PreCG, IFG, STG, and ROL nodes may be considered a key network in SeLECTS.

Consistent with prior studies, we observed no significant correlations between global network topology metrics and clinical variables (53). However, our study implied that disease duration was negatively related to the topological properties of the nodal network in the STG.L and ROL.L, suggesting that a longer disease duration was associated with a greater impact on the brain structural network in patients with epilepsy. In this study, the mean IQ was within the normal range for SeLECTS, which is consistent with the findings of previous studies (54). Recent studies have also pointed out that functional and structural abnormalities in STG related to the lower cognitive function scores in SeLECTS (55,56). Meanwhile, SeLECTS has been associated with language dysfunction (53). Language disorder in children with SeLECTS is associated with reduced functional connectivity (53). In our study, the E_nodal of the STG was correlated with the verbal IQ. Considering that IQ is a broad index, our result might not comprehensively reflect the effect of SeLECTS pathology on functions and specific neurocognitive processes or skill. Since the participants included in this study had a relatively short disease course, our research may reflect the early epilepsy-related changes in the brain structural network, potentially providing useful information for studying epilepsy.

Given the abnormal local and global network topological properties observed in SeLECTS children, we implemented exploratory analyses to evaluate the potential of WM structural network metrics as neuroimaging biomarkers for the classification of SeLECTS. Our study indicated that significant classification accuracy was attained using WM network topology as input features, effectively differentiating SeLECTS from TD individuals. This highlights the utility of this method for accurately diagnosing SeLECTS.

This study had several limitations. First, the sample size was relatively small, and a multicenter study with a larger sample size is required to validate our findings. Second, although participants with SeLECTS underwent neuropsychological evaluations, specific neuropsychological assessments, including those for attention and memory, were not performed. Third, as in previous studies, 30 gradient directions of DTI were used to make the MRI protocols tolerable for children (57), which might affect the construction of crossing fibers. Hence, caution is necessary when interpreting our results. Fourth, our neuropsychological evaluation was restricted to IQ scores. Future studies incorporating domain-specific assessments are needed to better understand the relationship between specific network alterations and the distinct cognitive profiles often seen in SeLECTS. Finally, although our cross-sectional findings delineate WM microstructure abnormalities in SeLECTS, longitudinal investigations are imperative to elucidate dynamic network reorganization patterns across disease stages.

Conclusions

This study suggests that brain structural network analysis can be feasibly applied to investigate the abnormal WM microstructure associated with disease status and cognitive dysfunction in epilepsy. Patients with SeLECTS showed a significant decrease in the E_g and increase in the L_p. E_nodalmight reflect the disease severity to some extent. The STG might be a suitable brain area for monitoring the condition of the disease. These findings provided neuroimaging support for the altered WM structural network in children with SeLECTS, advancing our understanding of potential epileptogenic mechanisms.

Acknowledgments

The authors thank Haipeng Hu, Department of Pediatrics, The First Affiliated Hospital of Xi’an Jiaotong University, and Jie Zheng, Department of Radiology, The First Affiliated Hospital of Xi’an Jiaotong University, for help with this study.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1131/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1131/dss

Funding: This work was supported by the Institutional Foundation of The First Affiliated Hospital of Xi’an Jiaotong University, China (grant No. 2021ZYT S-04); and the Clinical Research Award of The First Affiliated Hospital of Xi’an Jiaotong University, China (grant No. XJTU1AF-CRF-2023-011).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1131/coif). All authors report that this work was supported by the Institutional Foundation of The First Affiliated Hospital of Xi’an Jiaotong University, China (grant No. 2021ZYT S-04); and the Clinical Research Award of The First Affiliated Hospital of Xi’an Jiaotong University, China (grant No. XJTU1AF-CRF-2023-011). The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of The First Affiliated Hospital of Xi’an Jiaotong University (No. XJTU1AF2023LSL-011). Written informed consent was obtained from the parents or legal guardians of all participants prior to enrollment. Assent was also obtained from the children themselves when applicable.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Thijs RD, Surges R, O'Brien TJ, Sander JW. Epilepsy in adults. Lancet 2019;393:689-701. [Crossref] [PubMed]
2. Guerrini R. Epilepsy in children. Lancet 2006;367:499-524. [Crossref] [PubMed]
3. Panayiotopoulos CP, Michael M, Sanders S, Valeta T, Koutroumanidis M. Benign childhood focal epilepsies: assessment of established and newly recognized syndromes. Brain 2008;131:2264-86. [Crossref] [PubMed]
4. Doose H, Brigger-Heuer B, Neubauer B. Children with focal sharp waves: clinical and genetic aspects. Epilepsia 1997;38:788-96. [Crossref] [PubMed]
5. Teixeira JM, Santos ME, Oom P. Oral language in children with benign childhood epilepsy with centrotemporal spikes. Epilepsy Behav 2020;111:107328. [Crossref] [PubMed]
6. Widjaja E, Zamyadi M, Raybaud C, Snead OC, Smith ML. Abnormal functional network connectivity among resting-state networks in children with frontal lobe epilepsy. AJNR Am J Neuroradiol 2013;34:2386-92. [Crossref] [PubMed]
7. Thorn EL, Ostrowski LM, Chinappen DM, Jing J, Westover MB, Stufflebeam SM, Kramer MA, Chu CJ. Persistent abnormalities in Rolandic thalamocortical white matter circuits in childhood epilepsy with centrotemporal spikes. Epilepsia 2020;61:2500-8. [Crossref] [PubMed]
8. Zhang Y, Liu Z, Dou W, Wei J, Lv Y, Hou B, You H, Feng F. Study of the microstructure of brain white matter in medial temporal lobe epilepsy based on diffusion tensor imaging. Brain Behav 2023;13:e2919. [Crossref] [PubMed]
9. Bernhardt BC, Bonilha L, Gross DW. Network analysis for a network disorder: The emerging role of graph theory in the study of epilepsy. Epilepsy Behav 2015;50:162-70. [Crossref] [PubMed]
10. Pardoe HR, Berg AT, Archer JS, Fulbright RK, Jackson GD. A neurodevelopmental basis for BECTS: evidence from structural MRI. Epilepsy Res 2013;105:133-9. [Crossref] [PubMed]
11. Ciumas C, Saignavongs M, Ilski F, Herbillon V, Laurent A, Lothe A, Heckemann RA, de Bellescize J, Panagiotakaki E, Hannoun S, Marinier DS, Montavont A, Ostrowsky-Coste K, Bedoin N, Ryvlin P. White matter development in children with benign childhood epilepsy with centro-temporal spikes. Brain 2014;137:1095-106. [Crossref] [PubMed]
12. Koenis MMG, Brouwer RM, Swagerman SC, van Soelen ILC, Boomsma DI, Hulshoff Pol HE. Association between structural brain network efficiency and intelligence increases during adolescence. Hum Brain Mapp 2018;39:822-36. [Crossref] [PubMed]
13. Konomi T, Fujiyoshi K, Hikishima K, Komaki Y, Tsuji O, Okano HJ, Toyama Y, Okano H, Nakamura M. Conditions for quantitative evaluation of injured spinal cord by in vivo diffusion tensor imaging and tractography: preclinical longitudinal study in common marmosets. Neuroimage 2012;63:1841-53. [Crossref] [PubMed]
14. Bullmore E, Sporns O. Complex brain networks: graph theoretical analysis of structural and functional systems. Nat Rev Neurosci 2009;10:186-98. [Crossref] [PubMed]
15. Sporns O. Network attributes for segregation and integration in the human brain. Curr Opin Neurobiol 2013;23:162-71. [Crossref] [PubMed]
16. Achard S, Bullmore E. Efficiency and cost of economical brain functional networks. PLoS Comput Biol 2007;3:e17. [Crossref] [PubMed]
17. Bassett DS, Bullmore ET. Small-World Brain Networks Revisited. Neuroscientist 2017;23:499-516. [Crossref] [PubMed]
18. Latora V, Marchiori M. Efficient behavior of small-world networks. Phys Rev Lett 2001;87:198701. [Crossref] [PubMed]
19. Rubinov M, Sporns O. Complex network measures of brain connectivity: uses and interpretations. Neuroimage 2010;52:1059-69. [Crossref] [PubMed]
20. Zhao X, Chen J, Chen T, Liu Y, Wang S, Zeng X, Yan J, Liu G. Micro-Expression Recognition Based on Nodal Efficiency in the EEG Functional Networks. IEEE Trans Neural Syst Rehabil Eng 2024;32:887-94. [Crossref] [PubMed]
21. Takeuchi H, Taki Y, Sassa Y, Hashizume H, Sekiguchi A, Nagase T, Nouchi R, Fukushima A, Kawashima R. White matter structures associated with emotional intelligence: evidence from diffusion tensor imaging. Hum Brain Mapp 2013;34:1025-34. [Crossref] [PubMed]
22. Ai H, Yang C, Lu M, Ren J, Li Z, Zhang Y. Abnormal white matter structural network topological property in patients with temporal lobe epilepsy. CNS Neurosci Ther 2024;30:e14414. [Crossref] [PubMed]
23. Liu M, Chen Z, Beaulieu C, Gross DW. Disrupted anatomic white matter network in left mesial temporal lobe epilepsy. Epilepsia 2014;55:674-82. [Crossref] [PubMed]
24. Xue K, Luo C, Zhang D, Yang T, Li J, Gong D, Chen L, Medina YI, Gotman J, Zhou D, Yao D. Diffusion tensor tractography reveals disrupted structural connectivity in childhood absence epilepsy. Epilepsy Res 2014;108:125-38. [Crossref] [PubMed]
25. Liu H, Chen D, Liu C, Liu P, Yang H, Lu H. Brain structural changes and molecular analyses in children with benign epilepsy with centrotemporal spikes. Pediatr Res 2024;96:184-9. [Crossref] [PubMed]
26. Li Z, Zhang J, Wang F, Yang Y, Hu J, Li Q, Tian M, Li T, Huang B, Liu H, Zhang T. Surface-based morphometry study of the brain in benign childhood epilepsy with centrotemporal spikes. Ann Transl Med 2020;8:1150. [Crossref] [PubMed]
27. Ross EE, Stoyell SM, Kramer MA, Berg AT, Chu CJ. The natural history of seizures and neuropsychiatric symptoms in childhood epilepsy with centrotemporal spikes (CECTS). Epilepsy Behav 2020;103:106437. [Crossref] [PubMed]
28. Huang H, Shu N, Mishra V, Jeon T, Chalak L, Wang ZJ, Rollins N, Gong G, Cheng H, Peng Y, Dong Q, He Y. Development of human brain structural networks through infancy and childhood. Cereb Cortex 2015;25:1389-404. [Crossref] [PubMed]
29. Rodríguez-Cruces R, Concha L. White matter in temporal lobe epilepsy: clinico-pathological correlates of water diffusion abnormalities. Quant Imaging Med Surg 2015;5:264-78. [Crossref] [PubMed]
30. Specchio N, Wirrell EC, Scheffer IE, Nabbout R, Riney K, Samia P, Guerreiro M, Gwer S, Zuberi SM, Wilmshurst JM, Yozawitz E, Pressler R, Hirsch E, Wiebe S, Cross HJ, Perucca E, Moshé SL, Tinuper P, Auvin S. International League Against Epilepsy classification and definition of epilepsy syndromes with onset in childhood: Position paper by the ILAE Task Force on Nosology and Definitions. Epilepsia 2022;63:1398-442. [Crossref] [PubMed]
31. Cui Z, Zhong S, Xu P, He Y, Gong G. PANDA: a pipeline toolbox for analyzing brain diffusion images. Front Hum Neurosci 2013;7:42. [Crossref] [PubMed]
32. Descoteaux M, Deriche R, Knösche TR, Anwander A. Deterministic and probabilistic tractography based on complex fibre orientation distributions. IEEE Trans Med Imaging 2009;28:269-86. [Crossref] [PubMed]
33. Wang J, Wang X, Xia M, Liao X, Evans A, He Y. GRETNA: a graph theoretical network analysis toolbox for imaging connectomics. Front Hum Neurosci 2015;9:386. [Crossref] [PubMed]
34. de Reus MA, van den Heuvel MP. Estimating false positives and negatives in brain networks. Neuroimage 2013;70:402-9. [Crossref] [PubMed]
35. Humphries MD, Gurney K, Prescott TJ. The brainstem reticular formation is a small-world, not scale-free, network. Proc Biol Sci 2006;273:503-11. [Crossref] [PubMed]
36. Gao L, Li X, Li H, Zhang H, Liu X, Yang J. Associations of glymphatic function with structural network and cognition in self-limited epilepsy with centrotemporal spikes. Seizure 2024;120:104-9. [Crossref] [PubMed]
37. Ostrowski LM, Chinappen DM, Stoyell SM, Song DY, Ross EE, Kramer MA, Emerton BC, Chu CJ. Children with Rolandic epilepsy have micro- and macrostructural abnormalities in white matter constituting networks necessary for language function. Epilepsy Behav 2023;144:109254. [Crossref] [PubMed]
38. Watts DJ, Strogatz SH. Collective dynamics of 'small-world' networks. Nature 1998;393:440-2. [Crossref] [PubMed]
39. Barbey AK. Network Neuroscience Theory of Human Intelligence. Trends Cogn Sci 2018;22:8-20. [Crossref] [PubMed]
40. Guye M, Bettus G, Bartolomei F, Cozzone PJ. Graph theoretical analysis of structural and functional connectivity MRI in normal and pathological brain networks. MAGMA 2010;23:409-21. [Crossref] [PubMed]
41. Xiao F, Lei D, An D, Li L, Chen S, Chen F, Yang T, Ren J, Huang X, Gong Q, Zhou D. Functional brain connectome and sensorimotor networks in rolandic epilepsy. Epilepsy Res 2015;113:113-25. [Crossref] [PubMed]
42. Adebimpe A, Aarabi A, Bourel-Ponchel E, Mahmoudzadeh M, Wallois F. Functional Brain Dysfunction in Patients with Benign Childhood Epilepsy as Revealed by Graph Theory. PLoS One 2015;10:e0139228. [Crossref] [PubMed]
43. Ji GJ, Yu Y, Miao HH, Wang ZJ, Tang YL, Liao W. Decreased Network Efficiency in Benign Epilepsy with Centrotemporal Spikes. Radiology 2017;283:186-94. [Crossref] [PubMed]
44. Fang S, Li L, Weng S, Guo Y, Zhang Z, Wang L, Fan X, Wang Y, Jiang T. Decreasing Shortest Path Length of the Sensorimotor Network Induces Frontal Glioma-Related Epilepsy. Front Oncol 2022;12:840871. [Crossref] [PubMed]
45. Fransson P, Thompson WH. Temporal flow of hubs and connectivity in the human brain. Neuroimage 2020;223:117348. [Crossref] [PubMed]
46. Bhaya-Grossman I, Chang EF. Speech Computations of the Human Superior Temporal Gyrus. Annu Rev Psychol 2022;73:79-102. [Crossref] [PubMed]
47. Rocca R, Coventry KR, Tylén K, Staib M, Lund TE, Wallentin M. Language beyond the language system: Dorsal visuospatial pathways support processing of demonstratives and spatial language during naturalistic fast fMRI. Neuroimage 2020;216:116128. [Crossref] [PubMed]
48. Yin Y, Wang F, Yang Y, Tian M, Gao L, Liu H. Abnormalities of hemispheric specialization in drug-naïve and drug-receiving self-limited epilepsy with centrotemporal spikes. Epilepsy Behav 2022;136:108940. [Crossref] [PubMed]
49. Triarhou LC. Cytoarchitectonics of the Rolandic operculum: morphofunctional ponderings. Brain Struct Funct 2021;226:941-50. [Crossref] [PubMed]
50. Jiang Y, Song L, Li X, Zhang Y, Chen Y, Jiang S, Hou C, Yao D, Wang X, Luo C. Dysfunctional white-matter networks in medicated and unmedicated benign epilepsy with centrotemporal spikes. Hum Brain Mapp 2019;40:3113-24. [Crossref] [PubMed]
51. Besseling RM, Jansen JF, Overvliet GM, van der Kruijs SJ, Vles JS, Ebus SC, Hofman PA. Louw Ad, Aldenkamp AP, Backes WH. Reduced functional integration of the sensorimotor and language network in rolandic epilepsy. Neuroimage Clin 2013;2:239-46. [Crossref] [PubMed]
52. Jin G, Chen J, Du J, He L, Qi L, Wu D, Wang Y, Lin H, Ren L. Repetitive transcranial magnetic stimulation to treat benign epilepsy with centrotemporal spikes. Brain Stimul 2022;15:601-4. [Crossref] [PubMed]
53. Ofer I, Jacobs J, Jaiser N, Akin B, Hennig J, Schulze-Bonhage A, LeVan P. Cognitive and behavioral comorbidities in Rolandic epilepsy and their relation with default mode network's functional connectivity and organization. Epilepsy Behav 2018;78:179-86. [Crossref] [PubMed]
54. Pinton F, Ducot B, Motte J, Arbuès AS, Barondiot C, Barthez MA, Chaix Y, Cheminal R, Livet MO, Penniello MJ, Peudenier S, de Saint-Martin A, Billard C. Cognitive functions in children with benign childhood epilepsy with centrotemporal spikes (BECTS). Epileptic Disord 2006;8:11-23.
55. Shakeri M, Datta AN, Malfait D, Oser N, Létourneau-Guillon L, Major P, Srour M, Tucholka A, Kadoury S, Lippé S. Sub-cortical brain morphometry and its relationship with cognition in rolandic epilepsy. Epilepsy Res 2017;138:39-45. [Crossref] [PubMed]
56. Besseling RM, Jansen JF, Overvliet GM, van der Kruijs SJ, Ebus SC, de Louw A, Hofman PA, Vles JS, Aldenkamp AP, Backes WH. Reduced structural connectivity between sensorimotor and language areas in rolandic epilepsy. PLoS One 2013;8:e83568. [Crossref] [PubMed]
57. Fang H, Wu Q, Li Y, Ren Y, Li C, Xiao X, Xiao T, Chu K, Ke X. Structural networks in children with autism spectrum disorder with regression: A graph theory study. Behav Brain Res 2020;378:112262. [Crossref] [PubMed]