Salience network hub dysfunction is associated with cognitive impairment in self‐limited epilepsy with centrotemporal spikes: An OPM‐MEG study Authors: Chen Yang, Guangsheng Wu, Nan An, Tiejia Jiang, Yilong Wang, Wenbin Zhao, Wangyang Cao, Xiaofang Wang, Haiteng Jiang, Zhefeng Yuan Published: 2026 ABSTRACT AbstractObjectiveSelf‐limited epilepsy with centrotemporal spikes (SeLECTS), although characterized by well‐controlled seizures, is frequently associated with persistent cognitive deficits in affected children. This study aimed to investigate whether hub dysfunction of the salience network (SN) is associated with cognitive impairment in SeLECTS, using optically pumped magnetometer‐ magnetoencephalography (OPM‐MEG) within the framework of the triple network model (TNM), which encompasses the salience, default mode, and central executive networks (CENs).MethodsThis cross‐sectional case–control study enrolled 30 children with SeLECTS and 15 matched healthy controls (HC). Data acquisition comprised resting‐state OPM‐MEG, structural MRI, and the Das–Naglieri Cognitive Assessment System (CAS). Within the framework of the TNM, graph theory was employed to analyze hub function in the SN across multiple frequency bands.ResultsThe SeLECTS group exhibited widespread reductions in functional connectivity across the theta, alpha, and beta bands, with the beta band demonstrating the most pronounced decrease. Graph‐theoretical analysis revealed that betweenness centrality of the SN was significantly lower in the SeLECTS group and was independently associated with cognitive scores (B = 248.98, 95% CI 57.95–440.02; β = 0.28, p < 0.05). Exploratory mediation analysis suggested that beta‐band SN hub function was statistically consistent with an indirect pathway linking disease status and cognitive performance.SignificanceThis OPM‐MEG study suggests that beta‐band hub dysfunction of the SN is associated with cognitive impairment in children with SeLECTS. These findings provide hypothesis‐generating evidence that altered SN hub topology may contribute to cognitive vulnerability in SeLECTS and may serve as a potential MEG‐based biomarker of cognitive risk.Plain Language SummaryChildren with a common childhood epilepsy called self‐limited epilepsy with centrotemporal spikes may have learning and attention difficulties even when seizures are well controlled. We used a child‐friendly magnetic brain‐recording method to study how key brain networks communicate at rest. These children showed weaker network communication, especially in the beta frequency band, and a less central role of the salience network was linked to poorer cognitive performance. These findings may help identify children at greater risk of cognitive difficulties.Keywords: cognitive impairment, magnetoencephalography, salience network, self‐limited epilepsy with centrotemporal spikes, triple network model Key points Children with SeLECTS showed lower planning, attention, successive processing, and full‐scale CAS scores. Children with SeLECTS showed lower planning, attention, successive processing, and full‐scale CAS scores. OPM‐MEG revealed reduced theta‐, alpha‐, and beta‐band connectivity, most prominently in the beta band. OPM‐MEG revealed reduced theta‐, alpha‐, and beta‐band connectivity, most prominently in the beta band. Reduced beta‐band salience network centrality was associated with poorer cognition and may be a MEG marker. Reduced beta‐band salience network centrality was associated with poorer cognition and may be a MEG marker. INTRODUCTION SeLECTS is the most common idiopathic focal epilepsy in childhood. Due to favorable responses to antiseizure medications (ASMs) and spontaneous remission before adolescence, it has traditionally been classified as a “benign” epilepsy syndrome. 1 However, accumulating long‐term follow‐up studies indicate that effective seizure control does not necessarily equate to full neurocognitive recovery. Neuropsychological and meta‐analytic evidence further indicates that cognitive impairment is common in children with SeLECTS and may involve multiple cognitive domains, including working memory, complex attention, executive function, language‐related processing, and learning‐related abilities. 2 , 3 It is generally hypothesized that frequent epileptiform discharges disrupt the dynamic balance and functional integration of developing neural networks, ultimately contributing to cognitive deficits. 4 , 5 Nevertheless, the specific neural network mechanisms underlying these deficits remain unclear. According to epileptic network theory, seizures and interictal epileptiform activity are increasingly conceptualized as manifestations of abnormal activity within distributed cortical–subcortical networks rather than as events arising from an isolated focal region. 6 Among these large‐scale networks, the default mode network (DMN), central executive network (CEN), and salience network (SN) constitute the triple network model (TNM). 7 , 8 , 9 The DMN, a network that supports internally oriented cognition and is highly active at rest, has been repeatedly implicated in epilepsy, with altered DMN connectivity reported in absence epilepsy, 10 temporal lobe epilepsy, 11 and SeLECTS. 12 The CEN supports externally oriented cognitive control, including attention, working memory, and executive regulation, and its disruption has been observed in SeLECTS and temporal lobe epilepsy, where it is associated with impairments in attention and executive function. 13 , 14 The SN, anchored by the anterior insula and anterior cingulate cortex, is central to detecting salient events and coordinating transitions between internally oriented and externally directed network states. Altered SN connectivity has been reported in childhood absence epilepsy, 15 while disrupted SN functional and effective connectivity has been linked to attention deficits in temporal lobe epilepsy. 16 Moreover, altered functional connectivity among the DMN, SN, and attention/control networks has been reported in childhood absence epilepsy, supporting the relevance of interactions among these large‐scale cognitive networks in pediatric epilepsy. 17 Importantly, Fateh et al. recently demonstrated that altered salience network dynamic functional connectivity was related to cognitive performance in pediatric benign epilepsy with centrotemporal spikes with electrical status epilepticus during sleep (BECTS‐ESES), providing a directly relevant precedent for examining SN dysfunction within the pediatric centrotemporal‐spike epilepsy spectrum. 18 Magnetoencephalography (MEG) provides millisecond‐level temporal resolution for characterizing frequency‐specific neural network dynamics in pediatric epilepsy. OPM‐MEG is particularly suitable for this purpose because its wearable, cryogen‐free sensors can be positioned closer to the scalp and flexibly configured to accommodate smaller head sizes. In school‐aged children with epilepsy, on‐scalp OPM‐MEG has been reported to yield interictal epileptiform discharge (IED) peak amplitudes 2.3 to 4.6 times those recorded with conventional SQUID‐MEG, largely owing to the reduced brain‐to‐sensor distance. 19 , 20 These pediatric advantages motivated our use of OPM‐MEG to examine frequency‐resolved network alterations in children with SeLECTS. Despite growing recognition that cognitive impairment may persist in children with SeLECTS, the network‐level basis of these deficits remains incompletely understood, and objective markers for identifying children at higher cognitive risk are still limited. In particular, SN hub organization has rarely been examined in SeLECTS within the triple network framework using frequency‐resolved, source‐level MEG measures. To address these gaps, the present exploratory study integrates resting‐state OPM‐MEG, development‐specific functional parcellation, and graph‐theoretical analyses to examine whether altered SN hub topology is associated with cognitive performance in children with SeLECTS. Given the cross‐sectional design, the findings should be interpreted as associative and hypothesis‐generating rather than causal. METHODS Participants This study included 30 children with SeLECTS, all definitively diagnosed by pediatric neurology specialists at the Children's Hospital affiliated with Zhejiang University School of Medicine (detailed clinical characteristics in Table S1), and 15 age‐ and sex‐matched HC. All SeLECTS diagnoses were made in strict accordance with the 2017 epilepsy classification criteria issued by the International League Against Epilepsy (ILAE). 21 Exclusion criteria were as follows: (i) metallic implants; (ii) severe psychiatric, neurological, or major systemic disorders; and (iii) inability to complete MRI, MEG, or cognitive assessments due to poor cooperation. The study was approved by the Ethics Committee of the Children's Hospital affiliated with Zhejiang University School of Medicine (approval no. 2023‐IRB‐0067‐P‐01) and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants and their guardians after full explanation of the study. Data acquisition MEG recordings This study acquired brain magnetic signals using the LMEG‐64A, a 64‐channel OPM‐MEG system independently developed by the Hangzhou Institute of Extremely‐Weak Magnetic Field. The system integrates 64 high‐sensitivity optically pumped magnetometer sensors, each with a sensitivity of 15 fT, enabling precise detection of the extremely‐weak magnetic fields generated by neural activity. The system combines passive high‐permeability shielding with active feedback compensation to suppress environmental magnetic interference without a conventional shielded room. Resting‐state OPM‐MEG data were acquired for 15 min while participants were awake with eyes closed. Participants were instructed to remain relaxed, keep their eyes closed, stay awake, and minimize head and body movements during recording, with vigilance monitored by the experimenter and built‐in facial camera. MRI scan All participants underwent high‐resolution structural MRI on a 3.0 T scanner (GE SIGNA Premier, GE Healthcare) using a 3D‐navigated T1‐weighted sequence (TR = 7.0 ms, TE = 2.8 ms, FOV = 240 × 240 mm2, matrix = 240 × 240, voxel size = 1.0 mm3, slice thickness = 1.0 mm). To reduce motion artifacts and optimize B0 homogeneity, 3D shimming, positioning corrections, and head fixation were applied, ensuring images met quality standards for structural‐functional co‐registration, cortical reconstruction, and MEG source localization. All MRI scans were reviewed to identify visible structural abnormalities or focal lesions. The Das–Naglieri Cognitive Assessment System After MRI and MEG acquisition, all participants completed the Chinese version of the Das–Naglieri Cognitive Assessment System (CAS) in a quiet, well‐lit laboratory. Based on the PASS theory (Planning, Attention, Simultaneous, Successive), the CAS evaluates four core cognitive domains: Planning (strategy generation and problem‐solving regulation), Attention (maintaining focus and inhibiting distractions), Simultaneous Processing (integrating spatial or holistic information), and Successive Processing (encoding and recalling sequential information). Compared with traditional intelligence tests, the CAS provides a finer assessment of cognitive strategies, processing styles, and potential learning difficulties, and its reliability and validity have been confirmed. 22 Data preprocessing Structural MRI preprocessing Structural MRI data were processed using the CAT12 toolbox within the SPM12 framework. 23 T1‐weighted images underwent bias‐field correction, skull stripping, and tissue segmentation. Cortical surface reconstruction was performed using the CAT12 surface‐based processing pipeline to generate individual cortical meshes for subsequent MEG source‐space construction. Finally, individual images were linearly registered to MNI152 space to enable standardized ROI extraction and source‐level MEG analyses. MEG preprocessing MEG data were preprocessed in the Brainstorm environment. 24 The preprocessing pipeline included downsampling from 1000 to 500 Hz, 2–40 Hz band‐pass filtering, independent component analysis for removal of physiological artifacts, and segmentation of the 15‐min recording into 450 non‐overlapping 2‐s epochs. Potential IED‐containing epochs were first screened automatically using a dual criterion based on signal amplitude and instantaneous temporal gradient. For the amplitude criterion, a threshold‐search procedure was performed from 3 × MAD to higher thresholds in 0.5 × MAD increments. The 6 × MAD threshold was selected because it covered all clinician‐confirmed IEDs in this dataset while reducing excessive over‐flagging and was used only as a first‐pass screening threshold. All automatically flagged candidate epochs were then reviewed by an experienced clinical neurophysiologist. Confirmed IED‐containing epochs, together with 4 s before and after each IED, were excluded. The remaining clean epochs were concatenated in temporal order, and the middle 300 s of the concatenated clean data were selected for each participant. The number and percentage of excluded epochs are reported in Table S2. Source reconstruction and ROI time series extraction Forward and inverse modeling During source reconstruction, MEG signals were mapped to the cortical surface using depth‐weighted Minimum Norm Estimation (MNE) implemented in Brainstorm. 24 This distributed inverse solution was chosen because the present study aimed to characterize large‐scale ROI‐level network activity rather than focal epileptiform source localization. The use of depth weighting helps mitigate the superficial‐source bias inherent to standard MNE and supports more balanced cortical estimates across distributed regions, which is important for subsequent network‐level analyses in pediatric OPM‐MEG data. The reconstructed cortical signals were then used for ROI time series extraction and functional connectivity estimation. Brain parcellation and ROI signal extraction The selection of appropriate brain atlases is crucial for accurately characterizing functional networks, particularly in developmental populations. This study employed the Dev‐Atlas brain functional atlas, 25 developed from a cohort of typically developing adolescents, to extract neural activity signals from 19 ROIs spanning DMN, CEN, and SN. For each ROI, the representative time series was extracted using principal component analysis. Specifically, the first principal component of all source‐level signals within each ROI was retained as the ROI time series. The spatial distribution of ROIs and detailed functional lateralization are provided in Figure S1 and Table S3. Functional connectivity To accurately characterize information exchange between brain networks, this study employed corrected amplitude‐envelope correlation (AEC‐c) to assess functional connectivity (FC) among ROIs. AEC is an amplitude‐based resting‐state FC metric that quantifies functional coupling between brain regions by computing correlations between the power envelopes of neural signals. 26 To reduce spurious connectivity caused by source leakage, field spread, or volume conduction, signals were orthogonalized pairwise before AEC computation. 27 For each pair of source‐level ROI time series, this procedure removes the shared zero‐lag linear projection between the two signals before calculating amplitude‐envelope correlations. Based on the frequency range retained after data‐quality assessment of the present dataset, oscillatory connectivity was analyzed across five predefined bands: delta (2–4 Hz), theta (4–8 Hz), alpha (8–13 Hz), beta (13–30 Hz), and low‐gamma range (30–40 Hz). Graph theory Based on the weighted ROI‐to‐ROI functional connectivity matrices previously constructed using AEC‐c, this study further conducted graph‐theoretical analyses of the weighted network structure. Betweenness centrality (BC) for each ROI node was calculated using the GRETNA toolbox. 28 BC is well suited for evaluating inter‐network brokerage because it quantifies the extent to which a node lies on the shortest paths between other nodes in the reconstructed functional connectivity graph. In contrast, degree or strength primarily captures the number or cumulative weight of direct connections, eigenvector centrality emphasizes connections to other highly central nodes, and participation coefficient summarizes the distribution of connections across predefined modules. Although these alternative metrics provide complementary information about hub topology, they were less directly aligned with our hypothesis regarding SN‐mediated coordination among the DMN, CEN, and SN. Therefore, BC was defined a priori as the primary graph metric in this exploratory study. 29 , 30 In this study, lower BC was interpreted as reduced topological centrality or bridging potential of SN nodes within the TNM, rather than direct evidence of impaired cognitive‐control function. BC has been widely validated in prior neuroimaging studies as a robust metric for identifying hub regions and characterizing network reorganization across a range of neurological conditions, including epilepsy. 31 , 32 Statistical analyses Statistical analyses were performed using SPSS (version 26.0), the GRETNA toolbox within the MATLAB environment, and a connectomics toolbox implementing Network‐Based Statistics (NBS). All statistical tests were two‐tailed, with a significance threshold set at p < 0.05. Normality of continuous variables was evaluated within each group using the Shapiro–Wilk test. Between‐group comparisons were performed using independent‐samples t‐tests for approximately normally distributed variables or nonparametric tests when distributional assumptions were not met. Between‐group comparisons were accompanied by effect sizes and 95% confidence intervals where appropriate. For cognitive comparisons, Cohen's d and 95% confidence intervals for mean differences were reported. For regression models, unstandardized coefficients, standardized coefficients, and 95% confidence intervals were reported. Between‐group differences in ROI‐to‐ROI functional connectivity were evaluated using NBS within each predefined frequency band, with an initial edge‐wise threshold of t = 3.0, 5000 permutations, and component‐level significance set at NBS‐corrected p < 0.05. As these analyses were designed to characterize frequency‐specific network alterations within a priori defined and physiologically meaningful oscillatory bands, no additional correction was applied across frequency bands. To assess the threshold sensitivity of the primary beta‐band finding, adjacent component‐forming thresholds of t = 2.5 and t = 3.5 were also examined. Associations between network metrics and cognitive performance were examined via multiple linear regression, with CAS total score as the dependent variable, age and sex as covariates, and group and SN BC as primary predictors. Additional exploratory linear regression analyses were conducted using each CAS domain score as a separate dependent variable. Finally, an exploratory indirect‐effect analysis was conducted using PROCESS Model 4, with group status as the independent variable, beta‐band SN BC as the mediator, and CAS total score as the dependent variable, controlling for age and sex. The indirect‐effect was assessed using 5000 bootstrap resamples with 95% percentile bootstrap confidence intervals. To assess the potential influence of illness duration and medication burden, an additional patient‐only sensitivity analysis was conducted within the SeLECTS group. In this model, CAS total score was entered as the dependent variable, beta‐band SN BC was entered as the primary predictor, and age, sex, disease duration, and the number of concurrent ASMs were included as covariates. RESULTS Participants This study ultimately included 30 children with SeLECTS and 15 age‐ and sex‐matched HC for comparative analysis. All participants successfully completed MEG and MRI scanning, as well as cognitive assessment using the CAS. As shown in Table 1, there were no significant differences between the SeLECTS and HC groups in age (t = 0.319, p = 0.752) or sex distribution (χ 2 = 0.06, p = 0.806). No visible structural abnormalities or focal lesions were identified in the SeLECTS cohort on structural MRI review. TABLE 1Demographic and clinical characteristics of participants.CharacteristicsSeLECTS (n = 30)HC (n = 15)Statistics p‐ValueGender (male/female)15/158/7 χ 2 = 0.060.806Age (years, mean ± SD)11.2 ± 1.311.9 ± 1.4 t = 0.3190.752Disease duration (years, Median [Q1, Q3])2.2 [0.7, 4.2]———ASM regimen n (%)Number of ASMs (median [Q1, Q3])1 [1, 2]———Sodium Channel Blockers14 (46.7%)———AMPA Receptor Antagonists7 (23.3%)———SV2A Ligands1 (3.3%)———Polytherapy8 (26.7%)——— Note: Polytherapy refers to regimens involving ASMs from more than one mechanistic category. Drug mechanism classifications are as follows: Sodium Channel Blockers include Lacosamide (LCM) and Oxcarbazepine (OXC); AMPA Receptor Antagonists include Perampanel (PER); SV2A Ligands include Levetiracetam (LEV).Abbreviation: ASM, antiseizure medications; HC, healthy controls; Q1, first quartile; Q3, third quartile. Demographic and clinical characteristics of participants. Note: Polytherapy refers to regimens involving ASMs from more than one mechanistic category. Drug mechanism classifications are as follows: Sodium Channel Blockers include Lacosamide (LCM) and Oxcarbazepine (OXC); AMPA Receptor Antagonists include Perampanel (PER); SV2A Ligands include Levetiracetam (LEV). Abbreviation: ASM, antiseizure medications; HC, healthy controls; Q1, first quartile; Q3, third quartile. Cognitive profiles As shown in Table 2 and Figure 1, children with SeLECTS showed significantly lower scores in Planning, Attention, Successive Processing, and Full‐Scale score than HCs, with moderate‐to‐large effect sizes. In contrast, no significant between‐group difference was observed in Simultaneous Processing. TABLE 2Comparison of CAS scores between the SeLECTS and HC groups.GroupsPLANSIMAITSUCFSSeLECTS (n = 30)83.00 ± 15.75110 ± 13.8594.57 ± 12.7886.90 ± 11.9891.13 ± 12.32Controls (n = 15)98.87 ± 8.35117.67 ± 10.76112.67 ± 9.6397.47 ± 10.71106.93 ± 7.47Mean difference15.877.6718.1010.5715.8095% CI7.08, 24.66−0.57, 15.9110.54, 25.663.18, 17.968.80, 22.80Cohen's d 1.150.591.530.911.44 t 3.6411.8774.8312.8864.551 p‐Value0.0030.269<0.0010.024<0.001 Note: Data are presented as mean ± SD. Mean difference was calculated as HC minus SeLECTS. p‐values for the four CAS domain scores were adjusted using Bonferroni correction. FS was reported as an overall cognitive index.Abbreviations: AIT, Attention index; CI, confidence interval; FS, Full‐ Scale; HC, healthy controls; PLAN, Planning index; SIM, Simultaneous index; SUC, Successive index. Comparison of CAS scores between the SeLECTS and HC groups. Note: Data are presented as mean ± SD. Mean difference was calculated as HC minus SeLECTS. p‐values for the four CAS domain scores were adjusted using Bonferroni correction. FS was reported as an overall cognitive index. Abbreviations: AIT, Attention index; CI, confidence interval; FS, Full‐ Scale; HC, healthy controls; PLAN, Planning index; SIM, Simultaneous index; SUC, Successive index. (A) Scaled scores for the four core subtests of the Das—Naglieri Cognitive Assessment System (CAS). (B) CAS Full‐Scale scores. AIT, attention index; FS, Full‐Scale; HC, Healthy controls; PLAN, planning index; SeLECTS, Self‐limited epilepsy with centrotemporal spikes; SIM, Simultaneous index; SUC, Successive index. *p < 0.05, **p < 0.01, ***p < 0.001. Functional connectivity Using NBS with an initial edge‐wise threshold of t = 3.0 and 5000 permutations, significant reduced‐connectivity components were identified in the theta, alpha, and beta bands in the SeLECTS group compared with HC. The theta‐band component comprised three reduced edges (NBS‐corrected p = 0.032), and the alpha‐band component comprised four reduced edges (NBS‐corrected p = 0.026). The beta‐band component was the most extensive, comprising 13 reduced edges (NBS‐corrected p = 0.012), as illustrated in Figure 2. Threshold‐sensitivity analyses using adjacent primary thresholds showed that the beta‐band reduction pattern remained qualitatively stable. Specifically, the beta‐band component remained significant at t = 2.5 (18 edges, NBS‐corrected p = 0.008) and t = 3.5 (8 edges, NBS‐corrected p = 0.029). NBS‐identified subnetwork with significantly reduced beta‐band functional connectivity in children with SeLECTS. The figure illustrates edges with significantly decreased connectivity (p < 0.05, NBS‐corrected) derived from AEC‐c–based connectivity matrices. The subnetwork comprises 13 reduced connections predominantly located in hub regions of the triple network system (SN–DMN–CEN), including reduced intra‐SN integration, weakened SN–DMN/CEN coupling, and decreased direct connectivity between DMN and CEN nodes. Hub dysfunction We computed BC values for all nodes within the DMN, CEN, and SN, and then averaged the BC values within each network to obtain a global BC measure, as illustrated in Figure 3. The results showed that, in both the SeLECTS and HC groups, the average BC of the SN was higher than that of the DMN and CEN. This pattern is consistent with the proposed bridging role of the SN within the TNM. (A) Comparison of average betweenness centrality (BC) across the three core networks (DMN, CEN, SN), with SN showing significantly higher BC than DMN and CEN. (B) SN BC across theta, alpha, and beta bands, showing significantly higher BC in HC than SeLECTS in the beta band. **p < 0.01, ***p < 0.001. Given the significant functional connectivity abnormalities observed in the theta, alpha, and beta bands, subsequent hub analyses focused on these three frequency bands. Among them, the most prominent group difference was observed in the beta‐band SN, where BC was significantly lower in children with SeLECTS than in HCs, with a large effect size (Cohen's d = 1.28, 95% CI: 0.60–1.95, p < 0.001). This finding indicates reduced graph‐topological centrality or bridging potential of SN nodes in the beta‐band network. By contrast, beta‐band BC differences in the DMN and CEN did not reach statistical significance and showed smaller effect sizes than the SN effect (DMN: Cohen's d = 0.31, 95% CI: −0.31 to 0.93, p = 0.33; CEN: Cohen's d = 0.27, 95% CI: −0.35 to 0.89, p = 0.39). Overall, beta‐band hub alteration was most evident in the SN, whereas DMN and CEN differences were smaller and did not reach statistical significance. Correlation and mediation To explore the clinical association between SN beta‐band BC and cognitive function, we first constructed a multiple linear regression model with CAS total score as the dependent variable. The regression results are summarized in Table S4. Group was associated with lower CAS total score (B = −11.49, SE = 2.76, β = −0.44, 95% CI: −17.01 to −5.97, t = −4.16, p < 0.001). After controlling for age and sex, beta‐band SN BC remained positively associated with CAS total score (B = 248.98, SE = 95.57, β = 0.28, 95% CI: 57.95–440.02, t = 2.61, p < 0.05). To evaluate the robustness of this association to illness duration and medication burden, we performed a patient‐only sensitivity analysis within the SeLECTS group by additionally adjusting for disease duration and the number of concurrent ASMs. As shown in Table S5, beta‐band SN BC remained significantly associated with CAS total score after adjustment for age, sex, disease duration, and number of ASMs (B = 406.79, SE = 169.20, β = 0.42, 95% CI: 57.59–755.99, t = 2.40, p = 0.024). In contrast, neither disease duration (B = 0.74, β = 0.15, 95% CI: −1.54 to 3.01, p = 0.510) nor the number of ASMs (B = −2.80, β = −0.15, 95% CI: −11.03 to 5.43, p = 0.490) was significantly associated with CAS total score. As shown in Figure 4A, regressions showed positive associations between beta‐band SN BC and age‐ and sex‐adjusted CAS total score in both the HC group (R 2 = 0.43, approximate 95% CI: 0.05–0.76, p < 0.01) and the SeLECTS group (R 2 = 0.26, approximate 95% CI: 0.03–0.54, p < 0.01). We further tested the group × beta‐band SN BC interaction, which was not significant (β = 0.12, p = 0.314), indicating no statistical evidence that the association between beta‐band SN BC and CAS total score differed between the HC and SeLECTS groups. Additional domain‐specific regression analyses were performed using each CAS subscale score as a separate dependent variable. After adjustment for group status, age, and sex, beta‐band SN BC was positively associated with Successive Processing, whereas the associations with Planning, Attention, and Simultaneous Processing did not survive FDR correction. Detailed results are provided in Table S6. (A) Linear regression between beta‐band SN network hub centrality (BC) and predicted CAS total score, adjusted for age and gender, in the healthy control (HC) and SeLECTS groups. (B) Exploratory indirect‐effect model illustrating the statistical associations among group status, beta‐band SN betweenness centrality, and CAS total score. Standardized path coefficients (β) and their significance values are shown for each statistical path. Indirect‐effect analysis using PROCESS Model 4 was conducted, with age and sex included as covariates. The indirect‐effect linking group status, beta‐band SN BC, and CAS total score was −0.42, with a 95% percentile bootstrap CI of −0.77 to −0.13. DISCUSSION Dissociation in cognitive profiles: impaired successive but intact simultaneous processing All participants completed the Das–Naglieri CAS, which revealed significantly lower full‐scale cognitive scores in children with SeLECTS, indicating global impairments across multiple cognitive domains consistent with prior evidence of widespread network dysfunction. 33 Compared with HC, SeLECTS children showed deficits in planning, attention, and successive processing, while simultaneous processing remained intact. Successive and simultaneous processing represent two complementary modes of information integration within the PASS framework. Successive processing depends on the ability to encode, maintain, and retrieve discrete elements according to their temporal order and is therefore particularly important for phonological sequencing, ordered verbal recall, and stepwise cognitive operations. 34 These processes place high demands on anterior/fronto‐temporal systems involved in temporal organization, sequential encoding, and language‐related working memory. By contrast, simultaneous processing requires the integration of multiple elements into a coherent relational whole, allowing spatial, visual, and conceptual information to be processed in parallel. 34 Lesion‐based neuropsychological evidence indicates that simultaneous processing is more closely associated with posterior parieto‐occipital regions, whereas successive processing shows stronger associations with anterior brain regions. 35 In children with SeLECTS, centrotemporal epileptiform activity and related network disturbances may therefore be more likely to disrupt temporally ordered and language‐related cognitive operations than posteriorly mediated integrative processing. This provides a plausible explanation for the observed profile of impaired successive processing with relatively preserved simultaneous processing. Widespread functional connectivity deficits and beta‐band specificity Resting‐state functional connectivity analyses revealed that brain network dysfunction in children with SeLECTS is both widespread and frequency‐specific. Compared with HC, children with SeLECTS exhibited significant reductions in functional connectivity across the theta, alpha, and beta bands, which are traditionally implicated in cognitive processing. Notably, beta‐band connectivity demonstrated the most pronounced and spatially extensive decreases, suggesting that disrupted oscillatory coordination in this band may be closely associated with cognitive impairment in SeLECTS. Once regarded primarily as sensorimotor rhythms, beta oscillations are now recognized as supporting higher‐order cognitive control and large‐scale network coordination, and abnormal beta activity has been linked to cognitive deficits in depression 36 and schizophrenia. 37 In SeLECTS, converging evidence links beta‐band dynamics to both disease state and cognitive performance: EEG studies have shown abnormal beta power in seizure onset zones that normalizes with clinical improvement, 38 while MEG studies reveal suppressed beta activity in cognitive hubs such as the posterior cingulate cortex, correlating with domain‐specific deficits. 5 Although localized beta hyperconnectivity has been reported in left temporoparietal regions, 39 this may reflect localized short‐range synchronization alongside broader beta‐network disruption; although this interpretation may vary across modalities and connectivity metrics. Alpha and theta oscillations play complementary but more modest roles. Alpha supports sensory gating and attentional inhibition via suppression of irrelevant information, 40 and disrupted alpha activity in medial frontal regions has been associated with deficits in task switching and inhibitory control in SeLECTS. 5 Theta oscillations coordinate distributed neural ensembles during cognitive control and predict behavioral performance, 41 yet their detectability in resting‐state group analyses is limited by their dynamic, state‐dependent nature. 42 , 43 These characteristics may explain why theta‐band differences, although observed, were less prominent than those in the beta band. The pivotal role of salience network hub dysfunction This study identified a pronounced reduction in beta‐band SN betweenness centrality in children with SeLECTS, whereas beta‐band centrality differences in the DMN and CEN were smaller and did not reach statistical significance. Because BC is a graph‐topological metric, this finding is best interpreted as altered topological centrality or bridging potential of SN nodes within the reconstructed TNM graph, rather than as direct proof of impaired cognitive‐control function. Converging evidence from epilepsy research consistently implicates SN dysfunction. Most directly, Fateh et al. reported that altered SN dynamic functional connectivity was associated with cognitive performance in pediatric BECTS‐ESES. 18 Additional evidence has shown that altered SN connectivity is linked to attentional deficits in right temporal lobe epilepsy, 16 reduced SN gray matter volume and connectivity has been documented in MRI‐negative focal epilepsy, 44 and diminished integration of key SN hubs has been observed in childhood absence epilepsy. 15 In SeLECTS, this vulnerability may be compounded by developmental factors: although discharges originate in the centrotemporal region, their propagation via cortico‐cortical and cortico‐thalamic pathways can affect distal SN nodes such as the anterior insula and anterior cingulate cortex. Given the cross‐sectional design, repeated pathological activity during this developmental window may affect SN maturation and network‐switching capacity, but this interpretation remains hypothesis‐generating. In summary, this study identified selective alteration of beta‐band SN hub topology in children with SeLECTS and provided exploratory evidence for an indirect association between disease status, graph‐derived SN topological centrality, and domain‐specific cognitive deficits. Future research combining directional connectivity analyses with longitudinal study designs will be needed to clarify the temporal and potentially causal relationships between SN hub dysfunction and cognitive performance. From a clinical perspective, beta‐band SN betweenness centrality may represent a non‐invasive network‐level marker of cognitive vulnerability in children with SeLECTS. Given the pediatric tolerability of OPM‐MEG, this measure may help identify children who require closer cognitive monitoring or early neuropsychological intervention. However, these findings remain preliminary and require validation in larger longitudinal cohorts before clinical application. LIMITATIONS This study has several limitations. First, the sample size was modest and imbalanced, limiting statistical power and generalizability. Second, most patients were receiving ASMs. Although sensitivity analyses adjusted for disease duration and number of ASMs, residual effects of illness chronicity and medication class, dose, or exposure duration cannot be excluded. Continuous spike frequency or spike burden was not quantified, so dose–response relationships between interictal activity, network topology, and cognition could not be assessed. Third, the 300‐s clean segment was selected to minimize motion artifacts but may limit the repeatability of connectivity estimates. Relatedly, betweenness centrality may have more variable test–retest reliability than degree or efficiency metrics; therefore, beta‐band SN BC should be regarded as an exploratory candidate marker rather than a validated individual‐level biomarker. Fourth, depth‐weighted MNE and the Overlapping Spheres Model may have affected spatial specificity, residual source leakage, and forward‐model accuracy. Fifth, NBS findings depend on the primary component‐forming threshold. Although the beta‐band pattern remained significant at adjacent thresholds, edge extent varied; moreover, analyses were performed separately across frequency bands without cross‐band correction. Finally, the cross‐sectional resting‐state design precludes causal inference, temporal ordering, and definitive mediation. The absence of task‐based OPM‐MEG validation means that whether resting‐state SN topology reflects task‐related network dynamics remains uncertain. Future studies should combine larger longitudinal, drug‐naïve cohorts with continuous spike‐burden quantification and task‐based validation. CONCLUSION This study identified brain network abnormalities in children with SeLECTS using OPM‐MEG, a pediatric Dev‐Atlas, and the TNM. Patients showed reduced FC across theta, alpha, and beta bands, most prominently in the beta band, affecting connections within the SN and between the SN, DMN, and CEN. Graph analysis revealed decreased beta‐band BC of the SN, suggesting reduced topological centrality or bridging potential within the reconstructed TNM graph. Our findings suggest that altered beta‐band SN hub topology is associated with executive dysfunction in SeLECTS and may represent a candidate network‐level marker of cognitive vulnerability. Future studies should integrate task‐based OPM‐MEG targeting verbal sequencing with longitudinal designs to track network reorganization and cognitive development, thereby further evaluating the potential utility of beta‐band SN betweenness centrality as a MEG‐based biomarker for cognitive risk stratification. AUTHOR CONTRIBUTIONS Chen Yang: writing – original draft, methodology. Zhefeng Yuan: writing – review and editing, project administration. Haiteng Jiang: review and editing. Wangyang Cao: data collection. Guangsheng Wu: formal analysis. Tiejia Jiang: visualization. Yilong Wang: funding acquisition. Nan An: supervision, resources. Wenbin Zhao: data curation. Xiaofang Wang: data curation. FUNDING INFORMATION This work was supported by STI 2030‐Major Projects [Grant No. 2021ZD0201700]; the National Natural Science Foundation of China [Grant No. 82371453]; the Hangzhou Biomedical and Health Industry Development Support Technology Special Project [Grant Nos. 2021WJCY112 and 2024WJCY011]; and the Medical and Health Science Program of Zhejiang Province [Grant No. 2025HY0691]. CONFLICT OF INTEREST STATEMENT None of the authors has any conflict of interest to disclose. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines. Supporting information Appendix S1. Yang C, Wu G, An N, Jiang T, Wang Y, Zhao W, et al. Salience network hub dysfunction is associated with cognitive impairment in self‐limited epilepsy with centrotemporal spikes: An OPM‐MEG study. Epilepsia Open. 2026;11:1706–1717. 10.1002/epi4.70326 Chen Yang and Guangsheng Wu are co‐first authors of the article. Contributor Information Haiteng Jiang, Email: h.jiang@zju.edu.cn. Zhefeng Yuan, Email: chyzf@zju.edu.cn. DATA AVAILABILITY STATEMENT The data that support the findings of this study are available from the corresponding author upon reasonable request. REFERENCES Associated Data Supplementary Materials Appendix S1. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request.