这是用户在 2024-4-24 11:59 为 https://app.immersivetranslate.com/pdf-pro/uploading 保存的双语快照页面，由 沉浸式翻译 提供双语支持。了解如何保存？
2024_04_24_aead4e84073b7d647cb8g

Network connectivity predicts effectiveness of responsive neurostimulation in focal epilepsy
网络连通性可预测反应性神经刺激对局灶性癫痫的疗效

Joline M. Fan, Anthony T. Lee, Kiwamu Kudo, Kamalini G. Ranasinghe, Hirofumi Morise, Anne M. Findlay, Heidi E. Kirsch, Edward F. Chang, Srikantan S. Nagarajan, Vikram R. Rao
Joline M. Fan, Anthony T. Lee, Kiwamu Kudo, Kamalini G.Ranasinghe, Hirofumi Morise, Anne M. Findlay, Heidi E. Kirsch, Edward F. Chang, Srikantan S. Nagarajan, Vikram R. Rao

Network connectivity predicts effectiveness of responsive neurostimulation in focal epilepsy
网络连通性可预测反应性神经刺激对局灶性癫痫的疗效

OJoline M. Fan,' Anthony T. Lee, Kiwamu Kudo,, 4 - Kamalini G. Ranasinghe, '
OJoline M. Fan、' Anthony T. Lee、 Kiwamu Kudo、, 4 - Kamalini G.Ranasinghe, ' Hirofumi Morise, Anne M. Findlay, Heidi E. Kirsch, Edward F. Chang,
Hirofumi Morise, Anne M. Findlay, Heidi E. Kirsch, Edward F. Chang、 Srikantan S. Nagarajan and Vikram R. Rao"*
Srikantan S. Nagarajan 和 Vikram R. Rao "*

\author{ \作者{
  • Co-senior authors.  共同第一作者
    See Hitten Zaveri (https://doi.org/10.1093/braincomms/fcac114) for a scientific commentary on this article.
    关于这篇文章的科学评论，请参阅 Hitten Zaveri ( https://doi.org/10.1093/braincomms/fcac114)。
    }
Responsive neurostimulation is a promising treatment for drug-resistant focal epilepsy; however, clinical outcomes are highly variable across individuals. The therapeutic mechanism of responsive neurostimulation likely involves modulatory effects on brain networks; however, with no known biomarkers that predict clinical response, patient selection remains empiric. This study aimed to determine whether functional brain connectivity measured non-invasively prior to device implantation predicts clinical response to responsive neurostimulation therapy. Resting-state magnetoencephalography was obtained in 31 participants with subsequent responsive neurostimulation device implantation between 15 August 2014 and 1 October 2020. Functional connectivity was computed across multiple spatial scales (global, hemispheric, and lobar) using pre-implantation magnetoencephalography and normalized to maps of healthy controls. Normalized functional connectivity was investigated as a predictor of clinical response, defined as percent change in self-reported seizure frequency in the most recent year of clinic visits relative to pre-responsive neurostimulation baseline. Area under the receiver operating characteristic curve quantified the performance of functional connectivity in predicting responders ( reduction in seizure frequency) and non-responders ( . Leave-one-out cross-validation was furthermore performed to characterize model performance. The relationship between seizure frequency reduction and frequency-specific functional connectivity was further assessed as a continuous measure. Across participants, stimulation was enabled for a median duration of 52.2 (interquartile range, 27.0-62.3) months. Demographics, seizure characteristics, and responsive neurostimulation lead configurations were matched across 22 responders and 9 non-responders. Global functional connectivity in the alpha and beta bands were lower in non-responders as compared with responders (alpha, ; beta, ). The classification of responsive neurostimulation outcome was improved by combining feature inputs; the best model incorporated four features (i.e. mean and dispersion of alpha and beta bands) and yielded an area under the receiver operating characteristic curve of . The leave-one-out cross-validation analysis of this four-feature model yielded a sensitivity of , specificity of , positive predictive value of , and negative predictive value of . Global functional connectivity in alpha band correlated with seizure frequency reduction (alpha, ). Global functional connectivity predicted responder status more strongly, as compared with hemispheric predictors. Lobar functional connectivity was not a predictor. These findings suggest that non-invasive functional connectivity may be a candidate personalized biomarker that has the potential to predict responsive neurostimulation effectiveness and to identify patients most likely to benefit from responsive neurostimulation therapy. Follow-up large-cohort, prospective studies are required to validate this biomarker. These findings furthermore support an emerging view that the therapeutic mechanism of responsive neurostimulation involves network-level effects in the brain.
反应性神经刺激是治疗耐药性局灶性癫痫的一种很有前景的方法；然而，不同个体的临床结果差异很大。反应性神经刺激的治疗机制可能涉及对大脑网络的调节作用；然而，由于没有预测临床反应的已知生物标志物，患者的选择仍然是经验性的。本研究旨在确定在设备植入前进行的非侵入性脑功能连接测量是否能预测反应性神经刺激疗法的临床反应。在2014年8月15日至2020年10月1日期间，对31名随后植入反应性神经刺激设备的参与者进行了静息态脑磁图测量。利用植入前脑磁图计算了多个空间尺度（全局、半球和脑叶）的功能连通性，并与健康对照组的脑磁图进行归一化处理。归一化功能连通性作为临床反应的预测因子进行了研究，临床反应定义为最近一年门诊中自我报告的癫痫发作频率相对于反应前神经刺激基线的百分比变化。接收器操作特征曲线下面积量化了功能连通性在预测应答者（ ，癫痫发作频率减少）和非应答者（ ，癫痫发作频率减少）方面的性能。此外，还进行了留空交叉验证，以确定模型的性能。癫痫发作频率减少与频率特异性功能连通性之间的关系作为连续测量指标得到了进一步评估。所有参与者接受刺激的中位持续时间为 52.2 个月（四分位间范围为 27.0-62.3 个月）。22 名应答者和 9 名无应答者的人口统计学特征、癫痫发作特征和应答神经刺激导联配置相匹配。与应答者相比，非应答者α和β波段的整体功能连接性较低（α， ；β， ）。通过组合特征输入，神经刺激反应性结果的分类得到了改善；最佳模型包含四个特征（即阿尔法和贝塔波段的平均值和离散度），接收者操作特征曲线下的面积为 。对这一四特征模型进行一出交叉验证分析得出的灵敏度为 ，特异性为 ，阳性预测值为 ，阴性预测值为 。阿尔法波段的全局功能连通性与癫痫发作频率降低相关（阿尔法， ）。与半球预测因子相比，全局功能连通性对反应者状态的预测更强。 脑叶功能连通性不是一个预测指标。这些研究结果表明，非侵入性功能连接可能是一种候选的个性化生物标志物，有可能预测反应性神经刺激的有效性，并识别出最有可能从反应性神经刺激治疗中获益的患者。要验证这一生物标志物，还需要进行大队列、前瞻性的随访研究。这些发现进一步支持了一种新的观点，即反应性神经刺激的治疗机制涉及大脑中的网络级效应。
1 Department of Neurology and Weill Institute for Neurosciences, University of California San Francisco, San Francisco, CA, USA
1 美国加利福尼亚州旧金山，加州大学旧金山分校神经病学系和威尔神经科学研究所
2 Department of Neurosurgery, University of California San Francisco, San Francisco, CA, USA
2 美国加利福尼亚州旧金山加利福尼亚大学旧金山分校神经外科系
3 Medical Imaging Center, Ricoh Company, Ltd., Kanazawa, Japan
3 日本金泽理光株式会社医学影像中心
4 Department of Radiology and Biomedical Imaging, University of California San Francisco, San Francisco, USA
4 美国旧金山加州大学放射学和生物医学成像系
Correspondence to: Joline M. Fan, MD, MS
通讯作者：Joline M. Fan, MD, MSJoline M. Fan, MD, MS
University of California, San Francisco
加州大学旧金山分校
Department of Neurology and Weill Institute for Neurosciences
神经病学系和威尔神经科学研究所
505 Parnassus Ave, Box 0114
San Francisco, CA 94158, USA
美国加利福尼亚州旧金山 94158
E-mail: joline.fan@ucsf.edu
电子邮件： joline.fan@ucsf.edu
Keywords: RNS system; neuromodulation; imaginary coherence; functional connectivity; magnetoencephalography
关键词RNS 系统；神经调节；假想相干性；功能连接；脑磁图
Abbreviations: antiseizure medication; area under the receiver operating characteristic curve; functional connectivity; false discovery rate; intracranial electroencephalography; interquartile range; leave-one-out cross-validation; magnetoencephalography; negative predictive value; non-responder; positive predictive value; responder; receiver operating characteristic; region of interest; responsive neurostimulation; standard deviation; seizure onset zone
缩写： 抗癫痫药物； 接收者操作特征曲线下面积； 功能连通性； 假发现率； 颅内脑电图； 四分位数间范围； 一出交叉验证； 脑磁图； 阴性预测值； 无反应者； 阳性预测值； 有反应者； 接收者操作特征； 感兴趣区； 有反应的神经刺激； 标准偏差； 癫痫发作区

Graphical Abstract 图表摘要

31 Patients with presurgical magnetoencephalography, undergo chronic Responsive neurostimulation therapy
31 手术前接受脑磁图检查的患者，接受慢性反应性神经刺激治疗
  1. Compute functional connectivity and normalize to healthy cohort
    计算功能连通性，并与健康队列标准化
  1. Compare responders and non-responders
    比较应答者和非应答者

3) Classification and comparison of functional connectivity with seizure frequency
3) 功能连通性与癫痫发作频率的分类和比较

Depicting truncated methodology and findings. Thirtyne patients underwent presurgical magnetoencephalography (MEG) evaluation prior to responsive neurostimulation (RNS) implantation. First, resting-tate functional connectivity was computed based on presurgical MEG and normalized to healthy individuals. Second, averaged functional connectivity between RNS responders and non-esponders were compared. Third, the classification performance using functional connectivity and the relationship between seizure reduction and functional connectivity were evaluated.
描述截断的方法和研究结果。三十名患者在接受反应性神经刺激（RNS）植入手术前接受了脑磁图（MEG）评估。首先，根据手术前的脑磁图计算静息态功能连接，并与健康人进行归一化。其次，比较 RNS 响应者和非响应者之间的平均功能连接。第三，评估了使用功能连通性进行分类的性能以及癫痫发作减少与功能连通性之间的关系。

Introduction 导言

Of the 46 million people worldwide with active epilepsy, approximately one-third have seizures that are incompletely controlled with medications. For many of these individuals with drug-resistant epilepsy, surgical resection of seizureproducing brain tissue has the potential to yield seizure freedom. However, resection may not be feasible in patients with multiple seizure foci or seizures that arise from critical brain regions. In these cases, implanted neurostimulation devices, such as the responsive neurostimulation (RNS®) system, represent promising treatment alternatives. In a recent prospective study, treatment with the RNS system demonstrated progressive clinical benefit with a median reduction in seizure frequency of after nine years.
在全球 4600 万活动性癫痫患者中， ，约有三分之一的癫痫发作无法完全通过药物控制。对于其中许多耐药癫痫患者来说，手术切除产生癫痫发作的脑组织有可能使他们摆脱癫痫发作。 然而，对于有多个癫痫发作灶或癫痫发作来自重要脑区的患者来说，切除手术可能并不可行。在这种情况下，植入式神经刺激装置（如反应性神经刺激 (RNS®) 系统）是一种很有前景的替代治疗方法。在最近的一项前瞻性研究中，使用 RNS 系统进行治疗显示出了渐进的临床疗效，九年后癫痫发作频率的中位数减少了 。
Although median outcomes in clinical trials are encouraging, response to RNS therapy is highly variable across individuals. While over a third of patients experience dramatic improvement with reductions in seizure frequency exceeding , nearly a quarter of patients are non-responders, exhibiting seizure frequency reduction. Owing to a lack of methods to prognosticate even such extreme outcomes, patient selection for RNS in contemporary practice is largely empiric. Indeed, response does not appear to depend on age at seizure onset, seizure onset zone (SOZ) location, brain imaging abnormalities, or the number of seizure foci. Similar to all invasive therapies, implantation of the RNS system is associated with morbidity for patients and substantial costs to the medical system. The desire to minimize ineffective implants creates a critical need for biomarkers that reveal which patients are most and least likely to benefit from RNS therapy.
尽管临床试验的中位结果令人鼓舞，但不同患者对 RNS 治疗的反应却千差万别。虽然超过三分之一的患者病情得到显著改善，癫痫发作频率减少超过 ，但近四分之一的患者无应答，癫痫发作频率减少 。 由于缺乏方法来预测即使是如此极端的结果，因此在当代的临床实践中，RNS 患者的选择在很大程度上是经验性的。 事实上，反应似乎并不取决于癫痫发作时的年龄、癫痫发作区（SOZ）位置、脑成像异常或癫痫发作灶的数量。 与所有侵入性疗法类似，植入 RNS 系统也会给患者带来发病率，并给医疗系统带来巨大成本。 为了尽量减少无效植入，亟需生物标志物来揭示哪些患者最有可能或最不可能从 RNS 治疗中获益。
Recently, electrographic biomarkers of clinical response have been identified in RNS system electrocorticograms and in pre-implant intracranial electroencephalography (iEEG). These biomarkers derive from invasive recordings and spatially restricted sampling of the epileptogenic network, limiting their clinical utility. An ideal biomarker would be measurable non-invasively, before device implantation, and would not depend on the specific brain regions sampled by intracranial electrodes.
最近，在 RNS 系统皮层电图 和植入前颅内脑电图 (iEEG) 中发现了临床反应的电图生物标志物。 这些生物标记源于侵入性记录和致痫网络的空间受限取样，限制了其临床实用性。理想的生物标志物应该是在设备植入前就能进行非侵入性测量，并且不依赖于颅内电极采样的特定脑区。
Physiological mechanisms underlying the efficacy of RNS are incompletely understood, however, given the protracted time course of clinical response, they likely involve plasticity and gradual restructuring of the epileptogenic network. Growing evidence for network-level effects of chronic neurostimulation suggests that intrinsic network connectivity may mediate the effects of neurostimulation and help determine the potential for long-term evolution of the network. As such, we hypothesized that different patterns of network functional connectivity, readily assayed by resting-state magnetoencephalography (MEG), may confer differential susceptibility to chronic neurostimulation and thereby help predict the effectiveness of RNS therapy. To test this hypothesis, we examined a retrospective cohort of patients treated with the RNS system and evaluated their clinical outcomes in relation to resting-state FC measured by MEG prior to device implantation.
目前对 RNS 疗效的生理机制尚不完全清楚， ，但鉴于临床反应的时间过程较长，这些机制很可能涉及可塑性和致痫网络的逐步重组。 越来越多的证据表明，慢性神经刺激的网络级效应 表明，内在的网络连通性可能会介导神经刺激的效应，并有助于确定网络长期演变的潜力。因此，我们假设，不同的网络功能连通性模式很容易通过静息态脑磁图（MEG）进行评估，它们可能会对慢性神经刺激产生不同的易感性，从而有助于预测 RNS 治疗的效果。为了验证这一假设，我们对接受 RNS 系统治疗的患者进行了回顾性队列研究，并评估了他们的临床疗效与设备植入前 MEG 测量的静息态 FC 的关系。

Methods 方法

Study cohort 研究队列

All patients who had MEG imaging and were subsequently implanted with the RNS system (NeuroPace, Inc., Mountain View, CA) between 15 August 2014 and 1 October 2020 at UCSF Medical Center were considered for inclusion in this study. Two patients were excluded because they were seizure-free after implantation and RNS stimulation was never enabled. One patient was excluded due to stimulation-related side effects, preventing stimulation from being enabled as intended. The remaining patients were analyzed in this study. In addition, a subgroup analysis was performed on a subset of patients who did not undergo concurrent resective surgery . An intention-to-treat analysis was furthermore performed on all patients , including the three patients who were excluded in the primary analysis due to stimulation not being enabled or not active on both leads.
所有在 2014 年 8 月 15 日至 2020 年 10 月 1 日期间在加州大学旧金山分校医疗中心接受 MEG 成像检查并随后植入 RNS 系统（NeuroPace, Inc., Mountain View, CA）的患者 都被考虑纳入本研究。有两名患者因植入后无癫痫发作且从未启用 RNS 刺激而被排除在外。一名患者因与刺激相关的副作用而被排除在外，导致刺激无法如期启用。本研究对其余患者 进行了分析。此外，还对未同时接受切除手术的患者进行了亚组分析 。此外，还对所有患者进行了意向治疗分析 ，包括在主要分析中因刺激未启用或双侧导联均未激活而被排除在外的三名患者。
Age-matched healthy controls were recruited from the community with eligibility criteria including normal cognition, normal MRI, and absence of neurological or psychiatric illness. Data collection and analysis of MEG and RNS data were approved by the UCSF Institutional Review Board Committee, and all patients provided written consent for the analyses performed in this study.
年龄匹配的健康对照组 从社区招募，资格标准包括认知正常、磁共振成像正常、无神经或精神疾病。MEG 和 RNS 数据的收集和分析已获得加州大学旧金山分校机构审查委员会的批准，所有患者都对本研究中的分析提供了书面同意。

Magnetoencephalography data acquisition and preprocessing
脑磁图数据采集和预处理

All participants in the epilepsy cohort underwent a , clinically indicated routine EEG/MEG recording in the UCSF Biomagnetic Imaging Laboratory, using a whole-head MEG system (CTF, Port Coquitlam, British Columbia, Canada) comprising 275 axial gradiometers. Fiducial coils over the nasion and bilateral preauricular points were used to align the head position within the sensory array and to co-register MEG data with an individual's brain MRI. Age-matched controls underwent a shorter MEG recording session lasting 5 . All participants were required to be awake and interactive immediately prior to the recording session. Participants were then instructed to rest quietly in the scanner with eyes closed during the recording. MEG recording sessions were performed while the participants were on their normal antiseizure medications (ASMs); participants who were cognitively altered from their baseline or who had a seizure before or during MEG recording were not included in this study.
癫痫队列中的所有参与者都在加州大学旧金山分校生物磁成像实验室（UCSF Biomagnetic Imaging Laboratory）接受了由 275 个轴向梯度仪组成的全头 MEG 系统（CTF，加拿大不列颠哥伦比亚省高贵林港） 、临床指示的常规 EEG/MEG 记录。鼻翼和双侧耳前点上的靶线圈用于对准感觉阵列中的头部位置，并将 MEG 数据与个人的脑核磁共振成像共同注册。年龄匹配的对照组进行了一次较短的 MEG 记录，持续时间为 5 。记录前，所有参与者都必须保持清醒并进行互动。然后，参与者被要求在记录期间闭眼在扫描仪中安静休息。MEG 记录过程是在参试者正常服用抗癫痫药物 (ASM) 的情况下进行的；在 MEG 记录之前或期间出现认知改变或癫痫发作的参试者不包括在本研究中。
Raw EEG/MEG traces were parsed into 15 s epochs. Each segment was directly visualized, and any segments with movement/electrical artifact or epileptiform discharges were removed from the data set. The first six epochs to represent an awake, artifact-free and epileptiform discharge-
原始的 EEG/MEG 曲线被解析为 15 秒的时间片段。每个片段都被直接可视化，任何有运动/电伪影或痫样放电的片段都会从数据集中删除。前六个时程代表一个清醒的、无 伪影和癫痫样放电的时程。
free resting state were included in the analysis. The epochs were concatenated to achieve a 90 s time series representing the resting state. Prior work has demonstrated that of resting-state data reliably achieves stationarity. In select patients, a dual signal subspace project filter was used to remove metallic artifact from non-cranial implants. Source reconstruction was performed using an adaptive beamforming method to determine the voxel level time series from the sensor time series.
自由静息状态也包括在分析中。将 epochs 连接起来，得到一个代表静息状态的 90 秒时间序列。先前的研究表明， ，静息状态数据能可靠地实现静止性。 在选定的患者中，使用双信号子空间项目滤波器去除非颅内植入物的金属伪影。 使用自适应波束成形方法 进行信号源重建，以确定传感器时间序列的体素水平时间序列。

Magnetoencephalography network functional connectivity
脑磁图网络功能连接

Utilizing the Brainnetome atlas, the voxel level time series was mapped onto 218 cortical, atlas-based parcellations or regions of interest (ROIs). Time series per ROI were band-pass filtered into the following frequency bands: delta , theta , alpha , beta , and low gamma . Using the FieldTrip MATLAB Toolbox and custom built MATLAB tools, FC between ROIs were computed based on imaginary coherence, an established spectral coherence measure that is robust to volume conduction effects. The 218 cortical ROI level spatial maps were reduced in resolution to 44 modular ROI level maps to increase the signal to noise ratio, giving rise to 990 functional connections, which includes the averaged imaginary coherence within each module. To accommodate for the diverse locations of seizure foci and the native neurophysiologic features of each region, the FC maps of participants were normalized to that of healthy controls by z-scoring the FCs of each modular ROI-ROI connection to the corresponding modular ROI-ROI connections from a healthy cohort (Supplementary Fig. 1, ). The normalization processing enabled the identification of relative increases or decreases in network connectivity compared with a common basis.
利用 Brainnetome 图集， ，将体素水平的时间序列映射到 218 个基于图集的皮层扇区或感兴趣区（ROI）上。每个 ROI 的时间序列都经过带通滤波，分为以下频段：delta 、theta 、alpha 、beta 和低伽马 。使用 FieldTrip MATLAB 工具箱 和定制的 MATLAB 工具，根据假想相干性计算 ROI 之间的 FC，假想相干性是一种成熟的频谱相干性测量方法，对容积传导效应具有鲁棒性。 为了提高信噪比，将 218 个皮层 ROI 级空间图的分辨率降至 44 个模块 ROI 级地图，从而得到 990 个功能连接，其中包括每个模块内的平均虚相干性。为了考虑到癫痫发作灶的不同位置和每个区域的原生神经生理学特征，通过对每个模块化ROI-ROI连接的FC与健康组群中相应模块化ROI-ROI连接的FC进行z-评分，将参与者的FC图归一化为健康对照组的FC图（补充图1， ）。通过归一化处理，可以识别与共同基础相比网络连通性的相对增加或减少。
FC was computed for global and regional (hemispheric and lobar) spatial representations. Global FC was computed by averaging -scores across all ROI-ROI interactions, spanning the whole brain for each participant. Hemispheric FC was computed as the mean -score within the relevant hemisphere. The relevant hemisphere is determined by the location of the RNS leads, which are placed at the hypothesized SOZ and are thus a marker of the suspected SOZ. If RNS leads were placed bilaterally, the hemispheric FCs were averaged across both hemispheres. Lobar FC was computed as the mean -scores within the lobe containing the RNS lead. If RNS leads involved two lobes, then the FCs per lobe were averaged together for each patient. The relevant lobe and hemisphere for each patient were determined by evaluating post-implantation CT scans and identifying the locations of active leads connected to the device.
全球和区域（半球和脑叶）空间表征的FC均已计算。全局FC的计算方法是对所有ROI-ROI相互作用的平均 -scores，涵盖每个参与者的整个大脑。半球 FC 是根据相关半球内的平均 - score 计算得出的。相关半球由 RNS 导联的位置决定，RNS 导联放置在假设的 SOZ 上，因此是可疑 SOZ 的标记。如果 RNS 导联放置在双侧，则计算两个半球 FC 的平均值。脑叶 FC 计算为包含 RNS 导联的脑叶的平均 -scores。如果 RNS 导联涉及两个脑叶，则将每个患者每个脑叶的 FC 取平均值。每位患者的相关脑叶和半球是通过评估植入后的 CT 扫描并确定与设备连接的有源导线位置来确定的。

Responsive neurostimulation outcomes
响应式神经刺激成果

Clinical outcomes were based on averaged patient-reported seizure frequency determined at the most recent clinic visit of two time-samplings performed across all patients between October 2020 through October 2021, to reduce the noise from single time-point measures. Only clinic visits that exceeded a minimum of 6 months after stimulation onset were included. Seizure frequency was quantified relative to their pre-implant baseline (average seizure frequency over the 3 months immediately preceding RNS system implantation). Two patients underwent resective surgery in the years following RNS implantation, due to persistent breakthrough seizures despite RNS therapy. For these two patients, seizure frequency was documented prior to the definitive resection, occurring 63.8 and 39.2 months after stimulation onset. Patient-reported seizure frequency is the most widely used metric to determine clinical response to RNS. For example, a participant who previously had seven seizures per week and now has three seizures per week is computed to have seizure frequency reduction, i.e. the difference between the current and baseline frequency divided by the baseline frequency. Categorical testing was performed based on the standard definitions of participants with reduction in seizure frequency as 'responders' and those with reduction in seizure frequency as 'non-responders'.
临床结果基于 2020 年 10 月至 2021 年 10 月期间对所有患者进行的两次时间取样中最近一次就诊时确定的患者报告的癫痫发作频率平均值，以减少单一时间点测量的噪声。仅纳入刺激开始后至少超过 6 个月的门诊。发作频率相对于植入前基线（植入 RNS 系统前 3 个月的平均发作频率）进行量化。有两名患者在植入 RNS 后的数年内接受了切除手术，原因是尽管接受了 RNS 治疗，但仍有持续的突破性癫痫发作。这两名患者的癫痫发作频率在明确的切除手术前均有记录，分别发生在刺激开始后的 63.8 个月和 39.2 个月。患者报告的癫痫发作频率是确定 RNS 临床反应最广泛使用的指标。 例如，一名参与者以前每周有 7 次癫痫发作，现在每周有 3 次，则计算 癫痫发作频率减少，即当前频率与基线频率之差除以基线频率。根据标准定义 进行分类测试，将癫痫发作频率减少 的参与者视为 "有反应者"，将癫痫发作频率减少 的参与者视为 "无反应者"。

Statistical analysis 统计分析

The relationship between RNS outcomes and FC was assessed through both categorical and continuous statistical testing. The distributions of normalized imaginary coherence across all ROI-ROI interactions were obtained for each participant. Whole-brain spatial map visualizations depicted the mean imaginary coherence or -score for each ROI, i.e. the averaged interaction between the represented ROI and all other ROIs. To evaluate group-level statistics of imaginary coherence between responders and non-responders, statistical testing was performed using a linear mixed effects model (RStudio V 1.2.5033). The linear mixed effects model was performed for each frequency band and compared the -scored FCs between responders and non-responders with lobar ROI as a repeated measure. To account for multiple comparisons across frequency bands, we then applied a post hoc multiple comparison adjustment ( false discovery rate, FDR). Individual models were constructed for global and regional FC approaches.
通过分类和连续统计测试评估了 RNS 结果与 FC 之间的关系。每个受试者在所有 ROI-ROI 相互作用中的归一化假想相干性分布情况均已获得。全脑空间图可视化描述了每个ROI的平均假想相干性或 -score，即所代表的ROI与所有其他ROI之间相互作用的平均值。为了评估应答者和非应答者之间假想相干性的组级统计，使用线性混合效应模型（RStudio V 1.2.5033）进行了统计测试。线性混合效应模型针对每个频段进行，并比较 -评分的有反应者和无反应者的 FCs，以叶 ROI 作为重复测量。为了考虑各频段间的多重比较，我们进行了事后多重比较调整（ false discovery rate, FDR）。我们为全局和区域 FC 方法分别建立了模型。
The distributions of ROI-ROI interactions were additionally represented by mean and SD. Receiver operating characteristic (ROC) curves were computed by sweeping through all classification thresholds to elucidate the decision boundary and trade-off between sensitivity and specificity. The area under the ROC curve (AUC) was calculated as a measure of binary classification performance aggregated across all thresholds. Confidence intervals were determined using 1000 bootstrap replicates. Multivariate logistic regression was used to combine multiple features into a classification model. Specifically, two logistic regression models were constructed to predict responder status using multivariate features of the mean FC of the alpha and beta bands (i.e. 2
此外，ROI-ROI 相互作用的分布还用平均值和标度表示。通过扫描所有分类阈值计算接收者操作特征曲线（ROC），以阐明决策边界以及灵敏度和特异性之间的权衡。计算 ROC 曲线下面积（AUC），以此衡量所有阈值的二元分类性能。置信区间是通过 1000 次引导复制确定的。 多变量逻辑回归用于将多个特征组合到分类模型中。具体来说，利用阿尔法和贝塔带平均 FC 的多变量特征（即 2）构建了两个逻辑回归模型来预测应答者状态。
features), as well as the mean and SD of alpha and beta FCs (i.e. 4 features). Using R (Rstudio V 1.2.5033), logistic regression models were constructed using the generalized linear model (glm) function with a binary outcome (i.e. setting family-type to binomial). Multivariate features were inputted as linear features for the binary outcome of responder or non-responder. Prediction scores were then obtained from the fitted logistic regression model via the predict.glm function.
特征），以及 alpha 和 beta FCs（即 4 个特征）的平均值和 SD 值。使用 R (Rstudio V 1.2.5033)，利用广义线性模型 (glm) 函数构建二元结果的逻辑回归模型（即将家庭类型设置为二项式）。多变量特征作为线性特征输入，用于二元结果中的应答者或非应答者。然后通过 predict.glm 函数从拟合的逻辑回归模型中获得预测分数。
To assess further the logistic regression model performance, we performed leave-one-out cross-validation (LOOCV). The LOOCV was performed by splitting the observations into a training and testing set, in which all samples except for a single observation were used in the training set. The logistic regression model was constructed from the training set and tested on the single observation in the testing set. This process was iterated through all samples (e.g. 31 different models built on 30 samples), such that each sample served as the testing set. The accuracy, sensitivity, and specificity were computed. The optimal thresholds for class prediction were determined by maximizing the geometric mean of sensitivity and specificity. In addition, precision-recall curves were constructed based on LOOCV prediction scores and iterating across all classification thresholds to delineate the trade-off between precision and recall.
为了进一步评估逻辑回归模型的性能，我们进行了留一交叉验证（LOOCV）。LOOCV的方法是将观测数据分成训练集和测试集，其中训练集使用除单个观测数据外的所有样本。根据训练集构建逻辑回归模型，并在测试集中的单个观测值上进行测试。这一过程在所有样本中反复进行（例如，在 30 个样本中建立了 31 个不同的模型），因此每个样本都是测试集。计算准确率、灵敏度和特异性。通过最大化灵敏度和特异度的几何平均数，确定了类别预测的最佳阈值。此外，还根据 LOOCV 预测得分构建了精确度-召回率曲线，并在所有分类阈值之间进行迭代，以确定精确度和召回率之间的权衡。
Finally, a post analysis was performed to probe the association between mean/SD FC and RNS outcome as a continuous variable for frequency bands identified in the prior categorical analysis. The association was assessed using the Spearman rank correlation.
最后，针对先前分类分析中确定的频段，进行了 后分析，以探究作为连续变量的平均/标度 FC 与 RNS 结果之间的关联。相关性采用斯皮尔曼等级相关性进行评估。

Data Availability 数据可用性

The data that support the findings of this study are available from the corresponding author upon reasonable request.
支持本研究结果的数据可向相应作者索取。

Results 成果

Participant demographics and seizure characteristics
参与者的人口统计学特征和癫痫发作特征

Table 1 demonstrates demographics, seizure characteristics, and RNS lead configurations for all participants females), stratified by responder status. Mean age, duration of epilepsy, etiology, duration with RNS stimulation enabled, and baseline seizure frequency were not significantly different between responders and non-responders. There were no statistical differences in other characteristics, including seizure classification, lobar localization, RNS lead laterality, and prior or concurrent resection at the time of RNS implantation. In addition, there were no statistically significant differences between cohorts in the number or mechanism-of-action of ASMs used (Supplementary Fig. 2, Supplementary Table 1). One participant had bilateral mesial temporal lobe epilepsy. Indications for RNS in the other participants included involvement of eloquent cortex, multiple seizure foci, or regional neocortical epilepsy. Furthermore, as RNS leads are placed at the hypothesized SOZ, their locations serve as a proxy for the suspected SOZ. Demographics of the healthy controls were age and gender matched to the epilepsy cohort (Supplementary Table 2).
表 1 显示了所有参与者的人口统计学特征、癫痫发作特征和 RNS 导联配置（ ，女性），并按应答者状态进行了分层。应答者和非应答者的平均年龄、癫痫持续时间、病因、启用 RNS 刺激的持续时间和基线发作频率没有显著差异。其他特征，包括癫痫发作分类、脑叶定位、RNS 导联偏侧、RNS 植入时是否已进行切除术等，均无统计学差异。此外，在使用的 ASM 数量或作用机制方面，各组间也无统计学差异（补充图 2，补充表 1）。一名参与者患有双侧颞叶中叶癫痫。其他参试者的 RNS 指征包括受累的强皮质、多个癫痫发作灶或区域性新皮质癫痫。 此外，由于 RNS 导联被放置在假定的 SOZ 上，因此其位置可作为疑似 SOZ 的代表。健康对照组的人口统计学特征 年龄和性别与癫痫队列相匹配（补充表 2）。

Spatial functional connectivity maps for responders and non-responders
应答者和非应答者的空间功能连接图

To quantify intrinsic network connectivity, spatial FC maps were computed for each participant using pre-implant MEG. Fig. 1A-C compare spatial maps of a representative responder and non-responder to spatial FC maps averaged across healthy controls. Relative to controls, elevated FC is observed in the alpha band of the example responder, whereas reduced FC is observed in alpha and beta bands of the example non-responder. Region-to-region FCs of participants were normalized to those of healthy controls to facilitate comparison across participants with diverse epileptogenic networks. Normalized region-to-region FC maps demonstrate elevated connectivity in the alpha band of the responder and reduced connectivity in the alpha and beta bands of the non-responder (Fig. 1D and E). Summary statistics of the distribution of normalized region-to-region FCs in alpha and beta bands demonstrate an increased mean and dispersion (standard deviation, SD) in the responder, as compared with the non-responder (Fig. 1D and E, inset).
为了量化内在网络连通性，我们使用植入前 MEG 计算了每位受试者的空间 FC 图。图 1A-C 将具有代表性的反应者和非反应者的空间图与健康对照组的平均空间 FC 图进行了比较。与对照组相比，在反应者的 alpha 波段观察到 FC 升高，而在非反应者的 alpha 和 beta 波段观察到 FC 降低。参与者的区域间 FC 与健康对照组的 FC 进行了归一化处理，以便于对具有不同致痫网络的参与者进行比较。归一化的区域到区域 FC 图显示，反应者 alpha 波段的连通性升高，而非反应者 alpha 和 beta 波段的连通性降低（图 1D 和 E）。α和β波段归一化区域到区域FC分布的汇总统计显示，与非反应者相比，反应者的平均值和离散度（标准偏差，SD）都有所增加（图1D和E，插图）。

Functional connectivity as a predictor of responder and non-responders
预测应答者和非应答者的功能连通性

Group-averaged spatial maps of normalized FCs demonstrated global network changes with responders exhibiting higher FC in alpha and beta bands as compared to nonresponders (Fig. 2A and B). Averaging the FCs across the global spatial maps, responders were found to have higher mean FC in the alpha, beta, and gamma bands, as compared with non-responders (Fig. 2C, alpha, p p ; beta, 0.001; gamma, . Responders had positive mean normalized FC in the alpha band, implying an increase in mean connectivity relative to healthy controls. In contrast, non-responders had negative mean normalized FCs in the alpha frequencies, suggesting an overall reduction in connectivity as compared to the healthy controls. Both responders and non-responders had negative mean normalized FC in the beta and gamma frequencies, but non-responders exhibited higher magnitudes, implying more severely disrupted connectivity.
归一化 FC 的组均值空间图显示了全球网络的变化，与非应答者相比，应答者在α和β波段表现出更高的 FC（图 2A 和 B）。将全局空间图中的 FC 平均化后发现，与非应答者相比，应答者在α、β 和γ 波段的平均 FC 更高（图 2C，α，p p ；β， 0.001；γ， 。应答者在α波段的平均归一化 FC 值为正，这意味着相对于健康对照组，应答者的平均连通性有所增加。与此相反，非应答者阿尔法频率的平均归一化 FC 值为负值，表明与健康对照组相比，其连接性总体上有所降低。应答者和非应答者在β和γ频率上的平均归一化FC均为负值，但非应答者的幅度更大，这意味着连接性受到了更严重的破坏。
Regional FC demonstrated less robust frequency-specific differentiation between responders and non-responders than global FC. Hemispheric FC did not yield statistical differences between responders and non-responders in the gamma band but continued to demonstrate a statistically significant increase in the alpha and beta bands in responders as compared with non-responders (alpha,
与全局 FC 相比，区域 FC 对有反应者和无反应者的频率特异性区分较弱。半球 FC 在伽马波段 上没有显示出应答者和非应答者之间的统计学差异，但在阿尔法和贝塔波段上，应答者与非应答者相比继续显示出显著的统计学差异（阿尔法波段：α；贝塔波段：β；阿尔法波段：α；贝塔波段：β）、
Table I Participant characteristics, stratified by responder (R) and non-responder (NR) status
表 I 按应答者（R）和非应答者（NR）状态分列的参与者特征
All participants
所有与会者 		-values
Age, 0.349
Gender, F (%) 性别，女性（%） 0.704
Duration of epilepsy,
癫痫持续时间、 		0.198
Duration stimulation enabled, mos
启用刺激的持续时间（月 		0.948
Number of ASMs, no.
ASM 数量，编号 		0.671
Etiology, no. (%) 病因，人数（%） 0.508
Cryptogenic II (50) -
Encephalitis I (4.5) I (II.I) -
AVM I (4.5) I (II.I) -
PVNH I (II.I) -
Genetic/developmental 遗传/发育 -
FCD -
Stroke I (3.2) I (II.I) -
Seizure type , no. (%)
癫痫发作类型 ，人数 (%)		 0.079
FAS I (II.I) -
FIAS -
FBTC -
Baseline seizure frequency, per wk
基线发作频率，每周 		0.070
RNS lead locations', no. (%)
RNS 牵头地点"，人数（%） 		0.233
Frontal -
Neocortical temporal 新皮层颞叶 -
Mesial temporal 颞中部 I (II.I) -
Insular -
Parietal II (35.5) -
Occipital -
Other I (3.2) I (4.5) -
Prior resection, Y (%)
切除术前，Y（%） 		I (II.I) 0.379
Concurrent resection, Y (%)
同期切除术，Y（%） 		0.677
RNS lead types, no. (%)
RNS 导联类型，数量（%） 		0.569
Strips only -
Depths only I (3.2) I (4.5) -
Neocortical + Depth 新皮层 + 深度 I (II.I) -
RNS lead lateralization, no. (%)
RNS 导联侧向，人数（%） 		0.459
Right -
Left -
Both I (II.I) -
Values for age, duration of epilepsy, duration stimulation enabled, and baseline seizure frequency are given in medians with interquartile ranges in parentheses. Values for number of ASMs are given in means with standard deviations in parentheses.
年龄、癫痫持续时间、启用刺激的持续时间和基线发作频率的数值以中位数表示，括号内为四分位数范围。ASM 次数的数值以均值表示，括号内为标准差。
Differences between ( seizure reduction) and NR ( seizure reduction). Statistical testing performed by the Wilcoxon-Mann-Whitney test for two-sample comparisons. Fisher's exact testing was performed for categorical testing; post hoc P-values from multiple comparison testing is provided if Fisher's exact testing met significance, . May include more than one type for individual participants.
（ 癫痫发作减少）与 NR （ 癫痫发作减少）之间的差异。统计检验采用 Wilcoxon-Mann-Whitney 检验进行双样本比较。费舍尔精确检验用于分类检验；如果费舍尔精确检验具有显著性，则提供多重比较检验的事后 P 值， 。 单个参与者可能包含多个类型。
'Counts include each lead per participant.
计数包括每位参与者的每条线索。
antiseizure medications; arteriovenous malformation; periventricular nodular heterotopia; focal cortical dysplasia; FAS = focal aware seizure; focal impaired awareness seizure; focal to bilateral tonic-clonic seizure.
抗癫痫药物； 动静脉畸形； 室周结节性异位； 局灶性皮质发育不良；FAS = 局灶性意识发作； 局灶性意识障碍发作； 局灶性至双侧强直阵挛发作。
beta, . Although revealing similar trends, the lobar FC did not yield significant findings in any frequency band.
beta, 。虽然揭示了类似的趋势，但叶 FC 在任何频段都没有得出显著的结果。
Given their prominent and robust differences between responders and non-responders (Fig. 2C), alpha and beta frequency bands were used in subsequent classification models. ROC curves were constructed to evaluate the feasibility of classifying responders versus non-responders using the mean global FC (Fig. 3), yielding AUCs of 0.808 (95% CI: and 0.798 (95% CI: for alpha and beta frequency bands, respectively. A combined logistic regression model using the alpha and beta frequency bands yielded an AUC of 0.869 (95% CI: 0.729-1.000). Dispersion of FCs (SD) was further assessed as a predictor of responder status, to capture a different dimension of the distribution of connectivity strengths. In a corresponding
鉴于应答者和非应答者之间存在明显而稳健的差异（图 2C），α 和β 频带被用于随后的分类模型中。构建了 ROC 曲线来评估使用平均全局 FC 对有反应者和无反应者进行分类的可行性（图 3），结果显示，α 频带和β 频带的 AUC 分别为 0.808（95% CI： ）和 0.798（95% CI： ）。使用α和β频带的组合逻辑回归模型得出的AUC为0.869（95% CI：0.729-1.000）。为了捕捉连接强度分布的不同维度，我们进一步评估了功能点的分散性（SD），将其作为预测应答状态的指标。在相应的
ROC curve analysis, the SD of FCs yielded AUCs of 0.884 (95% CI: and 0.783 (95% CI, 0.615-0.951) for alpha and beta frequency bands, respectively. A combined logistic regression model using both the mean and SD of FCs within the alpha and beta frequency bands yielded an AUC of 0.970 (95% CI: 0.919-1.000). The addition of gamma features did not further improve the model. To assess model performance, LOOCV was performed on the best logistic regression model (i.e. combined logistic regression model using four features, mean/SD of FCs within the alpha and beta bands), which yielded an accuracy of , sensitivity of , and specificity of . LOOCV was additionally performed to assess the robustness of the AUC metric (see Supplementary Methods). Finally, precisionrecall curves were constructed to further assess model performance (Supplementary Fig. 3). The precision (i.e. positive
通过 ROC 曲线分析，FCs 的 SD 值在α和β频带的 AUC 分别为 0.884（95% CI： ）和 0.783（95% CI，0.615-0.951）。使用α和β频带内FCs的平均值和标度的组合逻辑回归模型得出的AUC为0.970（95% CI：0.919-1.000）。增加伽马特征并不能进一步改善模型。为了评估模型的性能，对最佳逻辑回归模型（即使用四个特征、α 和 β 频带内 FCs 的平均值/标度的组合逻辑回归模型）进行了 LOOCV 检验，结果显示：准确性为 ，灵敏度为 ，特异性为 。此外，还进行了 LOOCV 以评估 AUC 指标的稳健性（见补充方法）。最后，为了进一步评估模型的性能，我们构建了精确度-召回曲线（补充图 3）。精确度（即正
Figure I Representative global and region-to-region FC maps in the alpha and beta band for a responder and non-responder. (A). Global FC spatial maps of healthy controls (averaged across ) for the alpha (left) and beta (right) frequency bands. (B). Global FC spatial maps for an example responder, revealing regions of elevated FC in the alpha band (left). (C). Global FC spatial map for an example non-responder, revealing regions of reduced FC in both the alpha (left) and beta (right) frequency bands. (D). Normalized region-to-region FC map for the example responder in the alpha (left) and beta (right) bands. Normalization involves z-scoring a participant's FC map to the region-to-region FC maps of the healthy controls. Inset demonstrates the distribution of normalized FCs for the representative responder in the alpha and beta bands with global mean (SD) of and , respectively. The red dotted line indicates the mean of the normalized FC distribution. The white dotted line indicates the null hypothesis for all FCs, i.e. the healthy control. (E). Normalized region-to-region FC map for the non-responder, revealing low region-to-region FCs, as compared to healthy individuals. Inset reveals the distribution of normalized FCs for the representative non-responder in the alpha and beta bands with global mean (SD) of and , respectively.
图 I 有代表性的反应者和非反应者阿尔法和贝塔波段的全局和区域间 FC 图。(A).健康对照组α（左）和β（右）频段的全局 FC 空间图（在 上平均）。(B).一个反应者的全局 FC 空间分布图，显示了阿尔法频段（左）的 FC 升高区域。(C).非应答者的全局 FC 空间图，显示阿尔法（左）和贝塔（右）频段的 FC 下降区域。(D).范例反应者在α（左）和β（右）频段的归一化区域间 FC 图。归一化方法是将受试者的 FC 图与健康对照组的区域到区域 FC 图进行 z 评分。插图显示了具有代表性的反应者在α和β波段的归一化 FC 分布，其全局平均值（标度）分别为 和 。红色虚线表示归一化 FC 分布的平均值。白色虚线表示所有 FC 的零假设，即健康对照组。(E).非应答者的归一化区域到区域 FC 图，与健康人相比，显示出较低的区域到区域 FC。插图显示了具有代表性的非应答者在α和β波段的归一化 FC 分布情况，其全球平均值（SD）分别为 和 。
predictive value, ) and negative predictive value (NPV) for the four-feature model were and , respectively, based on the optimal threshold previously determined by the maximal geometric mean of the specificity and sensitivity. LOOCV metrics for all other features models are additionally presented in Supplementary Table 3.
根据特异性和灵敏度的最大几何平均数确定的最佳阈值，四特征模型的预测值（ ）和负预测值（NPV）分别为 和 。所有其他特征模型的 LOOCV 指标见补充表 3。
In a subgroup analysis of patients who did not undergo resection (Supplementary Fig. 4, , comprised and 7 , the mean global FC in the alpha, beta, and gamma bands remained statistically significant between responders and non-responders (Supplementary Fig. 4, alpha, 0.001 ; beta, gamma, ). In addition, an intention-to-treat analysis was performed, which includes the three patients for whom stimulation was never enabled or was active in only a single lead ( , comprised and 9 NR). Mean global FC in the alpha, beta, and gamma bands remained statistically significant between responders and non-responders in the intention-to-treat analysis (Supplementary Figure 5, alpha, ; beta, 0.001 ; gamma, .
在对未接受切除术的患者进行的亚组分析中（补充图 4， ，包括 和 7 ，α、β 和γ 波段的全局 FC 平均值在有反应者和无反应者之间仍具有统计学意义（补充图 4，α， 0.001；β， γ， ）。此外，还进行了意向治疗分析，其中包括从未启用刺激或仅在单导联激活的三名患者（ ，包括 和 9 NR）。在意向治疗分析中，α、β和γ波段的全局平均FC在应答者和非应答者之间仍具有统计学意义（补充图5，α， ；β， 0.001；γ， 。

Association between functional connectivity and seizure frequency reduction
功能连接与癫痫发作频率降低之间的关系

We next investigated whether frequency-specific FC could provide insight on the degree of seizure frequency reduction, beyond the binary outcome classification of responder versus non-responder. We investigated the two frequency bands, alpha and beta, that most robustly stratified responders and non-responders in the previous analyses. The association between mean normalized FC within the alpha frequency band and seizure reduction revealed the presence of a dose-response relationship (Fig. 4A, Spearman's correlation: . The of the distribution in the alpha band was also positively correlated with seizure reduction (Fig. 4B, ). In addition, a positive but reduced correlation was observed for mean hemispheric FC . Mean lobar in the alpha band was not significant (Fig. 4C, . The correlations of global and regional FC to seizure reduction within the beta
接下来，我们研究了频率特异性 FC 是否能提供有关癫痫发作频率减少程度的洞察力，而不局限于应答者与非应答者的二元结果分类。我们研究了阿尔法和贝塔这两个频段，这两个频段在之前的分析中对应答者和非应答者的分层效果最为显著。阿尔法频段内的平均归一化 FC 与癫痫发作减少之间的关联显示存在剂量-反应关系（图 4A，斯皮尔曼相关性： 。 在阿尔法频段的分布 也与癫痫发作减少呈正相关（图 4B， ）。此外，半球 FC 平均值 也呈正相关，但相关性降低。α波段的平均叶 并不显著（图 4C， 。在贝塔波段内，全球和区域 FC 与癫痫发作减少之间的相关性不显著（图 4C，）。
A
B
C
Figure Group analysis revealing frequency-specific patterns of global FC in the responder and non-responder cohorts. (A). Group-averaged spatial maps of responders reveal global and regional increases in FC in the alpha band (left) and reduced FC in the beta band (right), relative to healthy cohorts. -scores for each patient and each are computed relative to the healthy cohort. (B). Group-averaged spatial maps of non-responders demonstrate broadly reduced in both the alpha (left) and beta (right) bands, relative to healthy cohorts and responders. (C). Mean global FCs, averaged across the spatial maps, are increased in responders as compared to non-responders in alpha, beta, and gamma bands (alpha, Pfdr ; beta, Pfdr 0.001 ; gamma, -values adjusted with FDR). Positive and negative values indicate increased and decreased connectivity relative to healthy individuals, respectively. Statistical testing for each frequency band is obtained from a linear mixed effects model comparing the -scored FCs between responders and non-responders with lobar as a repeated measure. -values are corrected via a post hoc multiple comparison correction across frequency bands (FDR level 0.05 ). LS-means and confidence limits from the linear mixed effects model are depicted in
图 群体分析揭示了应答者和非应答者群体中全球 FC 的频率特异性模式。(A).与健康人群相比，应答者的组平均空间图显示，α 频段（左）的全局和区域 FC 增加，β 频段（右）的 FC 减少。 每个患者和每个 的分数都是相对于健康组群计算的。(B).与健康组群和有反应者相比，无反应者的组均空间图显示，α（左）和β（右）波段的 均广泛降低。(C).与非应答者相比，应答者在 alpha、beta 和 gamma 波段的全局平均 FCs（α，Pfdr ；beta，Pfdr 0.001；gamma， -数值经 FDR 调整）有所增加。正值和负值分别表示相对于健康个体的连接性增加和减少。每个频段的统计检验是通过线性混合效应模型比较 -应答者和非应答者的 FCs 得分，并将叶 作为重复测量。 -值通过各频段的事后多重比较校正（FDR 水平为 0.05 ）进行校正。线性混合效应模型的 LS 均值和 置信限见下表。
C. Number of asterisks indicates significance values of , , and , respectively.
C.星号分别表示 、 和 的显著性值。
band were not significant. In the subgroup analysis of patients who did not undergo resection, the association between seizure reduction and mean global in the alpha band remained statistically significant (Supplementary Fig. . In addition, the association between seizure reduction and mean global FC in the alpha band remained statistically significant in the intention-to-treat cohort (Supplementary Fig. .
带之间的关系并不显著。在对未接受切除术的患者进行的亚组分析中，癫痫发作减少与α波段平均全 ，两者之间的关系仍具有统计学意义（补充图 。此外，在意向治疗队列中，癫痫发作减少与α波段平均全局FC之间的关系仍具有统计学意义（补充图： 。
Figure 3 Global FC predicts RNS response. ROC curves for classification of responders and non-responders using frequency-specific global FC. Predictors include the mean global FC within the alpha (blue) and beta (red) frequency bands. In addition, two logistic regression models combining mean FC of the alpha/beta frequency bands (yellow) and the mean/SD of the alpha/ beta frequency bands (purple) are demonstrated. AUC is highest in the logistic regression model that combines both the mean and SD within the alpha and beta frequency bands (AUC: : .
图 3 全局 FC 预测 RNS 反应。使用特定频率的全局 FC 对有反应者和无反应者进行分类的 ROC 曲线。预测因子包括阿尔法（蓝色）和贝塔（红色）频带内的全局 FC 平均值。此外，结合阿尔法/贝塔频段的平均 FC（黄色）和阿尔法/贝塔频段的平均/标度（紫色）的两个逻辑回归模型也得到了展示。结合阿尔法和贝塔频带的平均值和标度的逻辑回归模型的 AUC 最高（AUC： : .

Discussion 讨论

Clinical response to RNS therapy is highly variable across patients, and currently, there are no established methods to prognosticate treatment response prior to device implant. 3,5 Here, using FC maps derived from pre-implant MEG, we demonstrate evidence that elevated global/regional network connectivity is associated with favorable outcomes and reduced global/regional network connectivity is associated with poor outcomes following subsequent treatment with RNS. Although FC in both alpha and beta frequency bands individually demonstrate discriminability between responders and non-responders, the combination of the two bands yields greater predictive value. To our knowledge, frequency-specific FC is the first non-invasive biomarker that has been demonstrated to have potential to predict clinical response to RNS therapy.
不同患者对 RNS 治疗的临床反应差异很大，目前还没有成熟的方法在设备植入前预测治疗反应。3,5 在此，我们利用植入前 MEG 导出的 FC 图，证明在接受 RNS 治疗后，全局/区域网络连通性的提高与良好的治疗效果相关，而全局/区域网络连通性的降低与不良的治疗效果相关。虽然α和β频段的FC单独显示了应答者和非应答者之间的可区分性，但这两个频段的组合产生了更大的预测价值。据我们所知，频率特异性 FC 是第一个被证明有可能预测 RNS 治疗临床反应的非侵入性生物标志物。
Although the therapeutic mechanism of RNS is unknown, current evidence suggests a role for chronic restructuring of the epileptogenic network. For example, neurophysiological features, such as the frequency modulation of ictal activity or interictal spike rates, evolve with chronic stimulation. A network-level mechanism presumably
尽管 RNS 的治疗机制尚不清楚，但目前的证据表明，RNS 在致痫网络的慢性重组中发挥了作用。 例如，发作活动的频率调制或发作间期尖峰率等神经生理学特征会随着慢性刺激而演变。 网络层面的机制可能是
underlies the efficacy of RNS for treating spatiallydistributed ('regional') SOZs, despite the initial conception of RNS as being most well-suited for the treatment of discrete, highly localized SOZs. Importantly, resting-state FC has been identified to be disrupted in epilepsy cohorts as compared to healthy controls, while subnetworks within the SOZ have been reported to have elevated FC. Indeed, in both the responders and non-responders, within the beta band was reduced relative to healthy controls; however, we demonstrate that within the beta band, the nonresponders had significantly lower FC as compared with the responder cohort, which remained more intact. In the alpha band, RNS responders exhibited increased FC relative to healthy cohorts, whereas non-responders exhibited decreased connectivity. The elevated FC in specific frequency bands (i.e. alpha/theta in responders and theta in nonresponders) may relate to spatial characteristics of the epileptogenic network. Specifically, RNS is often indicated for multifocal or regional SOZs, and, given that frequencyspecific FC can be elevated within the SOZ, the increased global FC seen here in certain frequency bands may relate to the overall larger contribution of a spatially extensive SOZ to the mean global FC. Prior studies have revealed that FC captures evoked functional networks, such as those engaged by neurostimulation. purely speculative possibility is that electrical pulses delivered by the RNS device can more readily diffuse through brain networks with relatively higher global FC, which may in some way potentiate the antiseizure effects of this therapy. Conversely, networks with low global FC may have limited spread of the therapeutic effects of neurostimulation. These findings are consistent with iEEG data suggesting that decreased synchronizability at seizure onset is correlated with poor RNS outcome. In contrast, increased local FC within the resection site or decreased network synchronizability have been shown to predict favorable surgical resection outcomes, suggesting that brain network characteristics which portend favorable response to neurostimulation versus resection may be as distinct as the therapies themselves.
尽管最初的概念认为 RNS 最适合治疗离散、高度局部化的 SOZ，但它是 RNS 治疗空间分布（"区域"）SOZ 疗效的基础。 重要的是，与健康对照组相比，癫痫队列中的静息态功能已被发现受到干扰，而据报道，SOZ 内的子网络具有升高的功能。 事实上，在应答者和非应答者中， ，β波段内的FC相对于健康对照组有所降低；但是，我们证明，在β波段内，非应答者的FC明显低于应答者，而应答者的FC则保持得更为完整。在阿尔法频段，RNS 反应者的 FC 相对于健康组群有所增加，而非反应者的连接性则有所下降。特定频段（即应答者的α/θ和非应答者的θ）的FC升高可能与致痫网络的空间特征有关。具体来说，RNS 通常适用于多灶性或区域性 SOZ， ，鉴于特定频率的 FC 可在 SOZ 内升高， ，此处某些频段的全局 FC 增加可能与空间上广泛的 SOZ 对平均全局 FC 的总体贡献较大有关。先前的研究显示，FC 可以捕捉到诱发功能网络，例如神经刺激所涉及的网络。 纯粹推测的可能性是，RNS 设备发出的电脉冲可以更容易地扩散到全局 FC 相对较高的大脑网络中，这可能在某种程度上增强了这种疗法的抗癫痫效果。相反，全局 FC 值较低的网络可能会限制神经刺激治疗效果的扩散。这些发现与 iEEG 数据一致，即癫痫发作时同步性降低与 RNS 效果不佳相关。 与此相反，切除部位内局部 FC 的增加 或网络同步性的降低 已被证明可预测良好的手术切除结果，这表明预示着对神经刺激和切除术产生良好反应的脑网络特征可能与疗法本身一样不同。
A major challenge in outcomes prediction of extratemporal lobe epilepsies involves the diversity of seizure onset locations and the unique neurophysiological features of each location. In this study, we accounted for this heterogeneity by normalizing FC maps of study participants to agematched healthy controls, which facilitates comparison across patients. Thus, brain regions that naturally have higher FC, such as occipital regions in the alpha frequency band during wakefulness, are accounted for through this normalization process. As such, each patient undergoes the same diagnostic testing, in which MEG sensor data are uniform or 'templated' across all patients, and the heterogenous SOZs are accounted for by the normalization process. By contrast, FC biomarkers based on are challenged by the disparate and 'tailored' electrode locations across patients, oversampling of the SOZ, and the long recording durations often needed to capture seizures.
颞叶外癫痫的预后预测面临的一个主要挑战是癫痫发作位置的多样性和每个位置独特的神经生理学特征。在本研究中，我们通过将研究参与者的 FC 图与年龄匹配的健康对照组进行归一化处理来考虑这种异质性，从而便于对不同患者进行比较。因此，自然具有较高 FC 值的脑区，如清醒时阿尔法频段的枕叶区，也通过这种归一化过程得到了考虑。因此，每位患者都要接受相同的诊断测试，其中所有患者的 MEG 传感器数据都是统一或 "模板化 "的，而异质 SOZ 则通过归一化过程加以考虑。相比之下，基于 的脑功能生物标志物则面临以下挑战：患者的电极位置各不相同且 "量身定制"，SOZ 采样过多，以及捕捉癫痫发作通常需要较长的记录时间。
An additional benefit of utilizing a templated rather than tailored diagnostic approach is that consistent, whole-brain coverage is obtained. Despite its superior temporal resolution, iEEG is tailored for each patient and spatially limited to areas of electrode coverage surrounding the hypothesized SOZ. As network connectivity has been observed to be highest within the epileptogenic core, spatially restricting analysis to the presumed epileptogenic core, i.e. the region adjacent to the RNS leads, could conceivably yield a more robust difference between responders and non-responders. However, in this study, global metrics were more robust than regional measures. With both categorical and continuous testing, we found that spatially restricting the FC analysis from global to regional measures resulted in less discriminability of clinical outcomes. The frequency-specific FC trends between responders and non-responders were similar between global and regional FC but less robust in hemispheric measures and absent in lobar measures. In addition, the correlation between FC and seizure reduction decreased from global to regional measures. Because the epileptogenic network is known to be distributed broad these findings suggest that whole-brain coverage
使用模板诊断法而非定制诊断法的另一个好处是可以获得一致的全脑覆盖。尽管 iEEG 具有更高的时间分辨率，但它是为每位患者量身定制的，而且在空间上仅限于假设 SOZ 周围的电极覆盖区域。由于已观察到致痫核心内的网络连通性最高， ，因此将空间分析限制在假定的致痫核心，即 RNS 导联的邻近区域，可以想象会在应答者和非应答者之间产生更强大的差异。然而，在本研究中，全局指标比区域指标更可靠。通过分类测试和连续测试，我们发现将 FC 分析的空间范围从全局性限制为区域性会降低临床结果的可区分性。有反应者和无反应者之间的频率特异性 FC 趋势在全局和区域 FC 中相似，但在半球测量中不那么稳健，而在叶测量中则不存在。此外，FC 与癫痫发作减少之间的相关性也从全局性降低到了区域性。由于已知致痫网络分布广泛 ，这些发现表明，全脑覆盖
Figure 4 Alpha band FC predicts degree of seizure frequency reduction. (A) Mean and (B) dispersion (SD) of the distribution of region-to-region in the alpha band correlates with degree of seizure frequency reduction (mean, , ). (C) The association between seizure frequency reduction and lobar FC is not statistically significant .
图 4 α 波段 FC 预测癫痫发作频率减少的程度。(A)α波段区域间 分布的平均值和(B)离散度（SD）与癫痫发作频率减少程度相关（平均值， ， ）。(C) 癫痫发作频率减少与脑叶 FC 之间的关联在统计学上并不显著 。
may capture critical network features, including regions that are not directly adjacent to the hypothesized SOZ.
可能会捕捉到关键的网络特征，包括与假设的 SOZ 并不直接相邻的区域。
Outside of the epileptogenic network, large-scale network impairments as measured by reduced FC have also been observed in people with epilepsy. Consistent with prior findings, responders and non-responders were both observed to have regions of reduced FC. We demonstrate that the mean and SD of the FC distribution independently predict RNS response, and that including both summary statistics improves performance of the classifier. One interpretation of this finding is that overall network connectivity and the heterogeneity of regional connectivity are both salient determinants of response to RNS therapy. This is consistent with recent studies showing that network topology influences the effects of neurostimulation. Network biomarkers that probe second-order metrics of the distribution of FCs and intrinsic network topology, such as graph theoretic measures, may also help predict clinical outcome and are of great interest for future work.
在致痫网络之外，在癫痫患者中也观察到了以FC降低来衡量的大规模网络损伤。 与之前的研究结果一致，我们观察到有反应者和无反应者都有 FC 降低的区域。我们证明，FC 分布的平均值和 SD 值可独立预测 RNS 反应，而且包含这两个汇总统计量可提高分类器的性能。对这一发现的一种解释是，整体网络连通性和区域连通性的异质性都是 RNS 治疗反应的显著决定因素。这与最近的研究显示网络拓扑会影响神经刺激效果是一致的。 探测 FC 分布和内在网络拓扑的二阶度量的网络生物标志物（ ，如图论度量）也可能有助于预测临床结果，这也是未来工作的重点。
Our cohort included participants who were treated with RNS and concurrent partial resection of epileptogenic brain tissue. We elected to keep these participants in the original cohort so as to increase the applicability of this study to realworld settings, where combining RNS and resection is an emerging treatment approach. As these participants were distributed equally across both the responder and nonresponder cohorts (Table 1), their inclusion is unlikely to drive our results. Furthermore, subgroup analysis confirmed that our central findings held even when patients treated with resection were excluded (Supplementary Fig 4).
我们的队列包括接受 RNS 治疗并同时部分切除致痫脑组织的参与者。我们选择将这些参与者保留在原始队列中，以提高本研究在现实世界中的适用性，因为在现实世界中，RNS 和切除术相结合是一种新兴的治疗方法。 由于这些参与者在应答者和非应答者队列中平均分布（表 1），因此将他们纳入其中不太可能会影响我们的结果。此外，亚组分析证实，即使将接受切除术治疗的患者排除在外，我们的中心研究结果仍然成立（补充图 4）。
Limitations of this study include the modest sample size and challenges inherent within clinically relevant, patient-reported seizure frequency. While RNS remains an emerging therapy, the sample size in this study is further constrained by the clinical selection of patients who typically undergo diagnostic presurgical MEG studies. In addition, the asymmetric proportion of responders and non-responders in this study reflects the distribution of RNS outcomes observed in retrospective analyses using a standard definition of 'responder' ( seizure frequency reduction). Future studies with larger sample sizes and cross-validation using hold out validation approaches are necessary to establish the generalizability and error rates of this biomarker for predicting response to RNS. Future prospective studies will also enable a more representative distribution of epilepsies, including a higher proportion of bitemporal epilepsy. In addition, RNS outcomes in this study are based on patient-reported seizure frequency, which is well-known to be prone to recall bias and other sources of error yet remains the gold standard endpoint in most epilepsy trials. In this work, we attempted to mitigate the errors of single time-point outcome evaluations by averaging the seizure frequency from two fixed assessments in the most recent year. The heterogeneity of the patient cohort is also a limitation of this work. We attempted to control for potential confounders by demonstrating a similar distribution of patient characteristics (e.g. medications, seizure type, etiology; see Table 1) between the two cohorts; however, larger, multicenter studies will be necessary in the future. Other limitations of the retrospective nature of our analysis, include non-standardized RNS stimulation parameters, ASMs, and behavioral factors.
本研究的局限性包括样本量不大，以及与临床相关的、由患者报告的癫痫发作频率所固有的挑战。虽然 RNS 仍是一种新兴疗法，但由于临床选择的患者通常会接受术前 MEG 诊断性研究，因此本研究的样本量受到了进一步限制。此外，本研究中应答者和非应答者的非对称比例反映了使用 "应答者 "标准定义（ 癫痫发作频率减少）进行回顾性分析时观察到的 RNS 结果的分布情况。 未来有必要进行样本量更大的研究，并采用保持验证方法进行交叉验证，以确定该生物标志物在预测 RNS 反应方面的通用性和误差率。未来的前瞻性研究还将使癫痫的分布更具代表性，包括更高比例的位颞叶癫痫。此外，本研究中的 RNS 结果是基于患者报告的癫痫发作频率，众所周知，患者报告的癫痫发作频率容易出现回忆偏差和其他误差来源 ，但仍是大多数癫痫试验的金标准终点。在这项工作中，我们试图通过平均最近一年两次固定评估的癫痫发作频率来减少单一时间点结果评估的误差。患者队列的异质性也是这项工作的局限性之一。我们试图控制潜在的混杂因素，在两个队列中显示出相似的患者特征分布（如药物、癫痫发作类型、病因；见表 1）；但是，未来有必要进行更大规模的多中心研究。我们的分析具有回顾性，其局限性还包括非标准化的 RNS 刺激参数、ASM 和行为因素。
In contemporary practice, some patients who appear to be good candidates for RNS may ultimately respond poorly to this therapy. Without methods to anticipate this outcome, patients must endure device implantation and potentially years of ineffectual device optimization before moving on to other therapies. Here, we demonstrate evidence that frequency-specific global and regional network connectivity may be associated with RNS outcomes and may therefore potentially serve as a personalized biomarker of treatment response. Such a biomarker obtained non-invasively prior to device implantation opens the door to the rational selection of patients who are most likely to benefit from RNS and, potentially, other neurostimulation devices.
在当代实践中，一些看似适合接受 RNS 治疗的患者最终可能对这种疗法反应不佳。如果没有预测这种结果的方法，患者就必须忍受设备植入和可能长达数年的无效设备优化，然后才能转而接受其他疗法。在这里，我们证明了频率特异性的全球和区域网络连通性可能与 RNS 结果相关，因此有可能作为治疗反应的个性化生物标志物。这种在设备植入前非侵入性获得的生物标志物为合理选择最有可能从 RNS 以及其他神经刺激设备中获益的患者打开了大门。

Acknowledgements 致谢

We would like to thank our patients with epilepsy who contributed their data to this study. We would also like to thank Leighton Hinkley for his helpful feedback on the manuscript.
我们要感谢为本研究提供数据的癫痫患者。我们还要感谢 Leighton Hinkley 对手稿的有益反馈。

Funding 资金筹措

Research reported in this publication was supported by National Institutes of Health grants under the award numbers: TL1TR001871-05, T32EB001631, R01EB022717, R01NS100440, R01AG062196, R01DC013979, R01DC176960, R01DC017091, and K08AG058749. Additional support was provided by a UCOP grant MRP17-454755, a DOD CDMRP grant W81XWH1810741, a grant from the Larry L. Hillblom Foundation 2019-A-013SUP, a grant from the Alzheimer's Association AARG-21849773, and a research contract between Ricoh's MEG Research Group and UCSF. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the funding agencies. J.M.F. is supported under the Doris Duke Physician Scientist Fellowship. V.R.R. is supported by the Ernest Gallo Foundation Distinguished Professorship in Neurology at UCSF.
本出版物所报道的研究得到了美国国立卫生研究院的资助，资助编号为TL1TR001871-05、T32EB001631、R01EB022717、R01NS100440、R01AG062196、R01DC013979、R01DC176960、R01DC017091 和 K08AG058749。此外，美国加州大学旧金山分校（UCOP）资助的MRP17-454755、国防部CDMRP资助的W81XWH1810741、拉里-希尔布洛姆基金会（Larry L. Hillblom Foundation）资助的2019-A-013SUP、阿尔茨海默氏症协会（Alzheimer's Association）资助的AARG-21849773以及理光MEG研究小组与加州大学旧金山分校之间的研究合同也提供了额外支持。本文内容仅由作者本人负责，不代表资助机构的官方观点。J.M.F.得到了多丽丝-杜克医生科学家奖学金的资助。V.R.R. 由加州大学旧金山分校欧内斯特-盖洛基金会神经病学杰出教授职位资助。

Conflict of interest 利益冲突

V.R.R. has received honoraria from NeuroPace, Inc for consulting and speaking engagements. The authors declare no targeted funding or compensation from NeuroPace, Inc for this study. K.K. and H.M. are employed by Ricoh Company Ltd. The remaining authors declare no competing financial interests.
V.R.R.从NeuroPace公司获得了咨询和演讲酬金。作者声明本研究未获得 NeuroPace, Inc 的定向资助或报酬。K.K.和H.M.受雇于理光有限公司。其余作者声明不存在竞争性经济利益。

Supplementary material 补充材料

Supplementary material is available at Brain Communications online.
补充材料可在 "大脑通讯 "网上查阅。

References 参考资料

  1. Beghi E, Giussani G, Abd-Allah F, et al. Global, regional, and national burden of epilepsy 1990-2016: A systematic analysis for the global burden of disease study 2016. Lancet Neurol. 2019;18: 357-375.
    Beghi E、Giussani G、Abd-Allah F 等. 1990-2016 年全球、地区和国家癫痫负担：2016年全球疾病负担研究系统分析》。Lancet Neurol.2019;18: 357-375.
  2. Choi H, Sell RL, Lenert L, et al. Epilepsy surgery for pharmacoresistant temporal lobe epilepsy: A decision analysis. JAMA. 2008;300: 2497-2505.
    Choi H、Sell RL、Lenert L 等人.抗药性颞叶癫痫的癫痫手术治疗：决策分析。美国医学会杂志》。2008;300: 2497-2505.
  3. Razavi B, Rao VR, Lin C, et al. Real-world experience with direct brain-responsive neurostimulation for focal onset seizures. Epilepsia. 2020;61:1749-1757.
    Razavi B、Rao VR、Lin C 等：用直接脑响应神经刺激治疗局灶性癫痫发作的实际经验。Epilepsia.2020;61:1749-1757.
  4. Bergey GK, Morrell MJ, Mizrahi EM, et al. Long-term treatment with responsive brain stimulation in adults with refractory partial seizures. Neurology. 2015;84:810-817.
    Bergey GK、Morrell MJ、Mizrahi EM 等：对难治性部分性癫痫发作成人进行反应性脑部刺激的长期治疗。神经病学》。2015;84:810-817.
  5. Nair DR, Laxer KD, Weber PB, et al. Nine-year prospective efficacy and safety of brain-responsive neurostimulation for focal epilepsy. Neurology. 2020;95:e1244-e1256.
    Nair DR、Laxer KD、Weber PB 等：脑响应神经刺激治疗局灶性癫痫的九年前瞻性疗效和安全性。神经病学》。2020；95:e1244-e1256。
  6. Ma BB, Rao VR. Responsive neurostimulation: Candidates and considerations. Epilepsy Behav. 2018;88:388-395.
    Ma BB, Rao VR.反应性神经刺激：候选者和考虑因素。Epilepsy Behav.2018;88:388-395.
  7. Purser MF, Mladsi DM, Beckman A, Barion F, Forsey J. Expected budget impact and health outcomes of expanded use of vagus nerve stimulation therapy for drug-resistant epilepsy. Adv Ther. 2018;35: 1686-1696.
    Purser MF、Mladsi DM、Beckman A、Barion F、Forsey J.《扩大使用迷走神经刺激疗法治疗耐药癫痫的预期预算影响和健康结果》。Adv Ther.2018;35: 1686-1696.
  8. Kokkinos V, Sisterson ND, Wozny TA, Richardson RM. Association of closed-loop brain stimulation neurophysiological features with seizure control among patients with focal epilepsy. JAMA Neurol. 2019;76:800-808.
    Kokkinos V、Sisterson ND、Wozny TA、Richardson RM。闭环脑刺激神经生理学特征与局灶性癫痫患者发作控制的关系。JAMA Neurol.2019;76:800-808.
  9. Desai S A, Tcheng TK, Morrell MJ. Quantitative electrocorticographic biomarkers of clinical outcomes in mesial temporal lobe epileptic patients treated with the RNS® system. Clin Neurophysiol. 2019;130:1364-1374.
    Desai S A, Tcheng TK, Morrell MJ.接受RNS®系统治疗的颞叶中位癫痫患者临床疗效的定量皮质电图生物标志物。Clin Neurophysiol.
  10. Khambhati AN, Shafi A, Rao VR, Chang EF. Long-term brain network reorganization predicts responsive neurostimulation outcomes for focal epilepsy. Sci Transl Med. 2021;13:6588.
    Khambhati AN、Shafi A、Rao VR、Chang EF。长期脑网络重组可预测局灶性癫痫的神经刺激疗效。Sci Transl Med.2021;13:6588.
  11. Scheid BH, Bernabei JM, Khambhati AN, et al. Intracranial electroencephalographic biomarker predicts effective responsive neurostimulation for epilepsy prior to treatment. Epilepsia. 2022; 63:652-662.
    Scheid BH、Bernabei JM、Khambhati AN 等：颅内脑电图生物标志物可预测癫痫治疗前有效的神经刺激反应。Epilepsia.2022; 63:652-662.
  12. Sohal VS, Sun FT. Responsive Neurostimulation suppresses synchronized cortical rhythms in patients with epilepsy. Neurosurg Clin N Am. 2011;22:481-488.
    Sohal VS, Sun FT.反应性神经刺激抑制癫痫患者的同步皮质节律。美国神经外科临床》。2011;22:481-488.
  13. Chiang S, Khambhati AN, Wang ET, Vannucci M, Chang EF, Rao VR. Evidence of state-dependence in the effectiveness of responsive neurostimulation for seizure modulation. Brain Stimul. 2021;14: 366-375.
    Chiang S, Khambhati AN, Wang ET, Vannucci M, Chang EF, Rao VR.反应性神经刺激对癫痫发作调节效果的状态依赖性证据。2021; 14: 366-375.
  14. Sisterson ND, Wozny TA, Kokkinos V, Bagic A, Urban AP, Richardson RM. A rational approach to understanding and evaluating responsive neurostimulation. Neuroinformatics. 2020;18: 365-375.
    Sisterson ND、Wozny TA、Kokkinos V、Bagic A、Urban AP、Richardson RM。理解和评估响应性神经刺激的合理方法。神经信息学》。2020;18: 365-375.
  15. Yang Y, Qiao S, Sani OG, et al. Modelling and prediction of the dynamic responses of large-scale brain networks during direct electrical stimulation. Nat Biomed Eng. 2021;5:324-345.
    Yang Y, Qiao S, Sani OG, et al. 直接电刺激过程中大规模脑网络动态响应的建模与预测。Nat Biomed Eng.
  16. Papadopoulos L, Lynn CW, Battaglia D, Bassett DS. Relations between large-scale brain connectivity and effects of regional stimulation depend on collective dynamical state. PLoS Comput Biol. 2020;16:e1008144.
    Papadopoulos L, Lynn CW, Battaglia D, Bassett DS.大规模大脑连通性与区域刺激效果之间的关系取决于集体动力学状态。PLoS Comput Biol.
  17. Silva AB, Khambhati AN, Speidel BA, Chang EF, Rao VR. Effects of anterior thalamic nuclei stimulation on hippocampal activity: Chronic recording in a patient with drug-resistant focal epilepsy. Epilepsy Behav Rep. 2021;16:100467.
    Silva AB、Khambhati AN、Speidel BA、Chang EF、Rao VR。丘脑前核刺激对海马活动的影响：耐药性局灶性癫痫患者的慢性记录。Epilepsy Behav Rep. 2021;16:100467.
  18. Tagliazucchi E, Laufs H. Decoding wakefulness levels from typical fMRI resting-state data reveals reliable drifts between wakefulness and sleep. Neuron. 2014;82:695-708.
    Tagliazucchi E, Laufs H. 从典型的 fMRI 静息态数据解码觉醒水平，揭示觉醒与睡眠之间的可靠漂移。神经元。2014;82:695-708.
  19. Guggisberg AG, Honma SM, Findlay AM, et al. Mapping functional connectivity in patients with brain lesions. Ann Neurol. 2008;63: 193-203.
    Guggisberg AG、Honma SM、Findlay AM 等：绘制脑损伤患者的功能连接图。Ann Neurol.2008;63: 193-203.
  20. Cai C, Xu J, Velmurugan J, et al. Evaluation of a dual signal subspace projection algorithm in magnetoencephalographic recordings from patients with intractable epilepsy and vagus nerve stimulators. Neuroimage. 2019;188:161-170.
    Cai C, Xu J, Velmurugan J, et al. 在来自难治性癫痫患者和迷走神经刺激器的脑磁图记录中评估双信号子空间投影算法。神经影像》。2019;188:161-170.
  21. Sekihara K, Hild KE, Nagarajan SS. A novel adaptive beamformer for MEG source reconstruction effective when large background brain activities exist. IEEE Trans Biomed Eng. 2006;53:1755-1764.
    Sekihara K, Hild KE, Nagarajan SS.当存在大量背景脑活动时，用于 MEG 信号源重构的新型自适应波束成形器。IEEE Trans Biomed Eng.
  22. Fan L, Li H, Zhuo J, et al. The human brainnetome atlas: A new brain atlas based on connectional architecture. Cereb Cortex. 2016;26:3508-3526.
    Fan L, Li H, Zhuo J, et al：基于连接结构的新型脑图谱。Cereb Cortex.2016;26:3508-3526.
  23. Oostenveld R, Fries P, Maris E, Schoffelen JM. FieldTrip: Open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput Intell Neurosci. 2011;2011: 156869.
    Oostenveld R, Fries P, Maris E, Schoffelen JM.FieldTrip：用于高级分析 MEG、EEG 和侵入性电生理数据的开源软件。Comput Intell Neurosci.2011;2011:156869.
  24. Bornot JM S, Wong-Lin KF, Ahmad AL, Prasad G. Robust EEG/ MEG based functional connectivity with the envelope of the imaginary coherence: Sensor space analysis. Brain Topogr. 2018;31: 895-916.
    Bornot JM S、Wong-Lin KF、Ahmad AL、Prasad G. 基于假想相干包络的鲁棒 EEG/ MEG 功能连接：传感器空间分析。Brain Topogr. 2018; 31: 895-916.
  25. Ma BB, Fields MC, Knowlton RC, et al. Responsive neurostimulation for regional neocortical epilepsy. Epilepsia. 2020;61: 96-106.
    Ma BB、Fields MC、Knowlton RC 等：针对区域性新皮质癫痫的反应性神经刺激。Epilepsia.2020;61: 96-106.
  26. Heck CN, King-Stephens D, Massey AD, et al. Two-year seizure reduction in adults with medically intractable partial onset epilepsy treated with responsive neurostimulation: Final results of the RNS System Pivotal trial. Epilepsia. 2014;55:432-441.
    Heck CN、King-Stephens D、Massey AD 等.接受反应性神经刺激治疗的药物难治性部分发作癫痫成人患者两年癫痫发作减少情况：RNS 系统关键性试验的最终结果。Epilepsia.2014;55:432-441.
  27. Hirsch LJ, Mirro EA, Salanova V, et al. Mesial temporal resection following long-term ambulatory intracranial EEG monitoring with a direct brain-responsive neurostimulation system. Epilepsia. 2020;61:408-420.
    Hirsch LJ, Mirro EA, Salanova V, et al. 使用直接脑响应神经刺激系统进行长期非卧床颅内脑电图监测后的颞中叶切除术。Epilepsia.2020;61:408-420.
  28. Morrell MJ. Responsive cortical stimulation for the treatment of medically intractable partial epilepsy. Neurology. 2011;77: 1295-1304
    Morrell MJ.用于治疗药物难治性部分性癫痫的反应性皮层刺激。神经病学》。2011;77: 1295-1304
  29. Krucoff MO, Wozny TA, Lee AT, Rao VR, Chang EF. Operative technique and lessons learned from surgical implantation of the neuropace responsive neurostimulation system in 57 consecutive patients. Oper Neurosurg. 2021;20: E98-E109.
    Krucoff MO、Wozny TA、Lee AT、Rao VR、Chang EF。神经空间响应式神经刺激 系统在 57 名连续患者中的手术植入技术和经验教训。2021; 20: E98-E109.
  30. Robin X, Turck N, Hainard A, et al. pROC: An open-source package for and to analyze and compare ROC curves. Bioinformatics. 2011;12:77.
    Robin X, Turck N, Hainard A, et al. pROC: 和 的开源软件包，用于分析和比较 ROC 曲线。 生物信息学。2011;12:77.
  31. Englot DJ, Hinkley LB, Kort NS, et al. Global and regional functional connectivity maps of neural oscillations in focal epilepsy. Brain. 2015;138:2249-2262.
    Englot DJ、Hinkley LB、Kort NS 等：局灶性癫痫中神经振荡的全局和区域功能连接图。脑。2015;138:2249-2262.
  32. Waites AB, Briellmann RS, Saling MM, Abbott DF, Jackson GD. Functional connectivity networks are disrupted in left temporal lobe epilepsy. Ann Neurol. 2006;59:335-343.
    Waites AB, Briellmann RS, Saling MM, Abbott DF, Jackson GD.左侧颞叶癫痫的功能连接网络被破坏。Ann Neurol.2006;59:335-343.
  33. Bernhardt BC, Bernasconi A, Liu M, et al. The spectrum of structural and functional imaging abnormalities in temporal lobe epilepsy. Ann Neurol. 2016;80:142-153.
    Bernhardt BC, Bernasconi A, Liu M, et al.Ann Neurol.2016;80:142-153.
  34. Hebbink J, van Blooijs D, Huiskamp G, Leijten FSS, van Gils SA, Meijer HGE. A comparison of evoked and non-evoked functional networks. Brain Topogr. 2019;32:405-417.
    Hebbink J, van Blooijs D, Huiskamp G, Leijten FSS, van Gils SA, Meijer HGE.诱发和非诱发功能网络的比较。Brain Topogr.
  35. Tomlinson SB, Porter BE, Marsh ED. Interictal network synchrony and local heterogeneity predict epilepsy surgery outcome among pediatric patients. Epilepsia. 2017;58:402-411.
    汤姆林森 SB、波特 BE、马什 ED。发作间期网络同步性和局部异质性可预测儿科患者的癫痫手术结果。Epilepsia.2017;58:402-411.
  36. Kini LG, Bernabei JM, Mikhail F, et al. Virtual resection predicts surgical outcome for drug-resistant epilepsy. Brain. 2019;142: 3892-3905.
    Kini LG、Bernabei JM、Mikhail F 等：虚拟切除预测耐药癫痫的手术效果。脑。2019;142: 3892-3905.
  37. Khambhati AN, Davis KA, Oommen BS, et al. Dynamic network drivers of seizure generation, propagation and termination in human neocortical epilepsy. PLoS Comput Biol. 2015;11: e1004608.
    Khambhati AN, Davis KA, Oommen BS, et al. 人类新皮质癫痫发作产生、传播和终止的动态网络驱动因素。PLoS Comput Biol.
  38. Spencer SS. Neural networks in human epilepsy: Evidence of and implications for treatment. Epilepsia. 2002;43:219-227.
    Spencer SS.人类癫痫的神经网络：治疗的证据和意义。Epilepsia.2002;43:219-227.
  39. Besson P, Bandt SK, Proix T, et al. Anatomic consistencies across epilepsies: A stereotactic-EEG informed high-resolution structural connectivity study. Brain. 2017;140:2639-2652.
    Besson P, Bandt SK, Proix T, et al. Anatomic consistencies across epilepsies：立体定向脑电图高分辨率结构连接研究。脑。2017;140:2639-2652.
  40. Bartolomei F, Lagarde S, Wendling F, et al. Defining epileptogenic networks: Contribution of SEEG and signal analysis. Epilepsia. 2017;58:1131-1147.
    Bartolomei F, Lagarde S, Wendling F, et al：SEEG 和信号分析的贡献。Epilepsia.2017;58:1131-1147.
  41. Zijlmans M, Zweiphenning W, van Klink N. Changing concepts in presurgical assessment for epilepsy surgery. Nat Rev Neurol. 2019; 15:594-606.
    Zijlmans M、Zweiphenning W、van Klink N.癫痫手术术前评估观念的转变。Nat Rev Neurol.2019; 15:594-606.
  42. Rings T, von Wrede R, Lehnertz K. Precursors of seizures due to specific spatial-temporal modifications of evolving large-scale epileptic brain networks. Sci Rep. 2019;9:10623.
    Rings T, von Wrede R, Lehnertz K. 由于演化中的大规模癫痫脑网络的特定时空改变而导致的癫痫发作前兆。Sci Rep. 2019;9:10623.
  43. Morgan VL, Chang C, Englot DJ, Rogers BP. Temporal lobe epilepsy alters spatio-temporal dynamics of the hippocampal functional network. NeuroImage Clin. 2020;26:102254.
    Morgan VL, Chang C, Englot DJ, Rogers BP.颞叶癫痫改变了海马功能网络的时空动态。NeuroImage Clin.2020;26:102254.
  44. Bonilha L, Jensen JH, Baker N, et al. The brain connectome as a personalized biomarker of seizure outcomes after temporal lobectomy. Neurology. 2015;84:1846-1853.
    Bonilha L, Jensen JH, Baker N, et al.大脑连接体作为颞叶切除术后癫痫发作结果的个性化生物标志物。神经病学》。2015;84:1846-1853.
  45. Englot DJ, Konrad PE, Morgan VL. Regional and global connectivity disturbances in focal epilepsy, related neurocognitive sequelae, and potential mechanistic underpinnings. Epilepsia. 2016;57: 1546-1557.
    Englot DJ、Konrad PE、Morgan VL。局灶性癫痫、相关神经认知后遗症和潜在机理基础的区域和整体连接紊乱。Epilepsia.2016;57: 1546-1557.
  46. Hays MA, Coogan C, Crone NE, Kang JY. Graph theoretical analysis of evoked potentials shows network influence of epileptogenic mesial temporal region. Hum Brain Mapp. 2021;42:4173-4186.
    Hays MA, Coogan C, Crone NE, Kang JY.诱发电位的图论分析显示致痫颞中叶区域的网络影响。人类脑图。2021;42:4173-4186.
  47. Keller CJ, Honey CJ, Mégevand P, Entz L, Ulbert I, Mehta AD. Mapping human brain networks with cortico-cortical evoked potentials. Philos Trans R Soc B Biol Sci. 2014;369:20130528.
    Keller CJ, Honey CJ, Mégevand P, Entz L, Ulbert I, Mehta AD.用皮质诱发电位绘制人脑网络图。Philos Trans R Soc B Biol Sci.
  48. Khambhati AN, Kahn AE, Costantini J, et al. Functional control of electrophysiological network architecture using direct neurostimulation in humans. Netw Neurosci. 2019;3:848-877.
    Khambhati AN、Kahn AE、Costantini J 等人利用直接神经刺激对电生理网络结构进行功能控制。Netw Neurosci.2019;3:848-877.
  49. Stacey W, Kramer M, Gunnarsdottir K, et al. Emerging roles of network analysis for epilepsy. Epilepsy Res. 2020;159: 106255.
    Stacey W, Kramer M, Gunnarsdottir K, et al.癫痫研究》，2020; 159: 106255。
  50. Tran DK, Tran DC, Mnatsakayan L, Lin J, Hsu F, Vadera S. Treatment of multi-focal epilepsy with resective surgery plus responsive neurostimulation (RNS): One institution's experience. Front Neurol. 2020;11:545074.
    Tran DK、Tran DC、Mnatsakayan L、Lin J、Hsu F、Vadera S.用切除手术加反应性神经刺激（RNS）治疗多灶性癫痫：一家机构的经验。Front Neurol.2020;11:545074.
  51. Elger CE, Hoppe C. Diagnostic challenges in epilepsy: Seizure under-reporting and seizure detection. Lancet Neurol. 2018;17: 279-288.
    Elger CE, Hoppe C. 癫痫诊断的挑战：癫痫发作漏报与癫痫发作检测。Lancet Neurol.2018;17: 279-288.
  1. Received November 11, 2021. Revised February 23, 2022. Accepted April 21, 2022. Advance access publication April 26, 2022
    2021 年 11 月 11 日收到。2022 年 2 月 23 日修订。2022 年 4 月 21 日接受。2022 年 4 月 21 日接受。
    (C) The Author(s) 2022. Published by Oxford University Press on behalf of the Guarantors of Brain.
    (C) 作者 2022 年。由牛津大学出版社代表大脑担保人出版。
    This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
    这是一篇根据知识共享署名许可协议 ( https://creativecommons.org/licenses/by/4.0/) 条款发布的开放获取文章，该协议允许在适当引用原作的前提下，在任何媒体上不受限制地再利用、传播和复制。