Assessment of the utility of optically pumped magnetometer magnetoencephalography in preoperative localization of refractory epilepsy: A prospective study

Abstract

Objective

Precise localization of the epileptogenic zone (EZ) is crucial for epilepsy surgery success. Optically pumped magnetometer magnetoencephalography (OPM‐MEG) is a promising noninvasive technique requiring rigorous clinical validation.

Methods

In this prospective diagnostic study, 68 patients with refractory epilepsy underwent 90‐min interictal OPM‐MEG. Dipoles were fitted to interictal epileptiform discharges for localization. The primary objective was to evaluate the spatial concordance between OPM‐MEG and the EZ defined by intracranial electroencephalography (iEEG; stereo‐EEG or electrocorticography), assessed at the sublobar level using Gwet AC1. The secondary objective was to evaluate the diagnostic value of OPM‐MEG for surgical outcome. This analysis included 51 patients who underwent curative intervention (resection or thermocoagulation). The reference standard was a composite of the treated brain region and seizure freedom (International League Against Epilepsy [ILAE] class 1 or Engel class I) at ≥12‐month follow‐up, from which sensitivity, specificity, and diagnostic odds ratio (OR) were calculated.

Results

OPM‐MEG showed almost perfect agreement with iEEG‐based EZ localization overall (AC1 = .885, concordance rate = 90.0%), with substantial agreement in temporal (80.1%, AC1 = .723) and almost perfect agreement in extratemporal regions (92.0%, AC1 = .926). The Euclidean centroid distance between OPM‐MEG and iEEG localizations was significantly shorter in concordant versus discordant cases. In the assessment of diagnostic value, OPM‐MEG demonstrated a sensitivity of 85.7% and specificity of 65.2% (OR = 11.25) under ILAE criteria, and a sensitivity of 73.0% and specificity of 64.3% (OR = 4.86) under Engel criteria.

Significance

OPM‐MEG demonstrates high concordance with iEEG for EZ localization and provides robust diagnostic value for predicting postoperative seizure freedom, supporting its utility in the presurgical evaluation of refractory epilepsy.

Key points.

• OPM‐MEG demonstrates almost perfect agreement (90.0%, Gwet AC1 = .885) with iEEG for epileptogenic zone localization at the sublobar level.

• OPM‐MEG provides robust diagnostic value for predicting postoperative seizure freedom, with a sensitivity of 85.7% and specificity of 65.2% (OR = 11.25) under ILAE criteria.

• This large prospective study validates OPM‐MEG as a promising, noninvasive, and cost‐effective tool in the presurgical evaluation of refractory epilepsy.

1. INTRODUCTION

Epilepsy, one of the most widespread neurological disorders worldwide, affects more than 70 million people, and approximately 30% of these cases are identified as refractory epilepsy (RE). ¹ , ² Surgical resection of the epileptogenic zone (EZ) continues to be the main treatment for RE. Following surgical intervention, between 30% and 85% of epilepsy patients may achieve freedom from seizures. ³ The success of epilepsy surgery hinges on precise preoperative localization of the EZ, which in complex cases necessitates intracranial electrode implantation guided by reliable noninvasive evaluations. Here, magnetoencephalography (MEG) is well established as a supplementary noninvasive functional imaging technique that detects neurogenic magnetic fields to complement existing localization methods. ⁴ , ⁵ , ⁶ , ⁷

Conventional superconducting quantum interference device (SQUID)‐based MEG systems, clinically validated since the 1970s, face operational constraints such as high cryogenic maintenance requirements and signal attenuation. ⁸ , ⁹ Unlike SQUID‐MEG, which relies on superconducting principles requiring cryogenic liquid helium cooling and maintains a fixed sensor‐to‐scalp distance, optically pumped magnetometer (OPM)‐MEG operates based on optical pumping of atomic vapors at room temperature. ¹⁰ This fundamental difference eliminates the need for cryogenic systems, thereby significantly lowering operational costs and enhancing patient comfort. Furthermore, the capacity for direct scalp proximity in OPM‐MEG sensors facilitates stronger signal acquisition compared to conventional systems. Supporting evidence from a small‐sample, single‐center prospective study demonstrates that OPM‐MEG outperforms SQUID‐MEG in detecting interictal epileptiform discharges (IEDs) in pediatric populations, yielding higher amplitude, superior signal‐to‐noise ratio (SNR) and comparable localization precision. ¹¹ , ¹² Vivekananda et al. used OPM‐MEG to perform preoperative EZ localization in an epilepsy patient. Source reconstruction of an average of interictal spikes was performed using dipole fitting within FieldTrip software (http://fieldtriptoolbox.org). After surgical resection of the epileptic area, the frequency of epileptic seizures decreased, and postoperative magnetic resonance imaging (MRI) showed that the resection area was almost identical to the IED area located by OPM‐MEG. The study suggests that OPM‐MEG can improve the quality of EZ localization. ⁹ Although above studies have demonstrated the feasibility of OPM‐MEG in localizing EZ from multiple perspectives, there have not yet been large‐scale data reports on its accuracy in localizing EZ.

To evaluate the localization performance and diagnostic value of OPM‐MEG in preoperative localization of epilepsy, we conducted this prospective study. We used the localization results from intracranial electroencephalography (iEEG), including stereo‐EEG (SEEG) and electrocorticography (ECoG), along with postoperative prognostic outcomes obtained more than 12 months after surgery, as reference standards to assess the consistency, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, and odds ratio (OR) of OPM‐MEG. This study was designed and reported with reference to the STARD (Standards for Reporting Diagnostic Accuracy) guidelines. ¹³

2. MATERIALS AND METHODS

2.1. Approval and registration of the study protocol

The study was approved by the ethics committee of Tongji Hospital, affiliated with Tongji Medical College, Huazhong University of Science and Technology (ethics approval number: TJ‐IRB202404111). It was also prospectively registered with the Chinese National Health Information Assurance Platform Medical Research Registration and Filing System (https://www.medicalresearch.org.cn/). Written informed consent was obtained from all participants. A total of 15 minor patients had their informed consent signed on their behalf by a legal guardian.

2.2. Study design and patients

This study was conducted at Tongji Hospital, affiliated with Tongji Medical College, Huazhong University of Science and Technology. From January 2024 to May 2025, as part of the National Key Research and Development Program of China, we prospectively recruited 68 patients with RE who were scheduled for neurosurgical treatment.

In accordance with the institutional review board‐approved protocol, all participants first completed a comprehensive noninvasive preoperative evaluation phase. This standardized assessment included the following: detailed clinical history review, OPM‐MEG recording, video‐EEG monitoring, MRI, positron emission tomography (PET), and multidisciplinary team (MDT) consultation. Candidates meeting predetermined criteria subsequently progressed to the second‐phase invasive monitoring protocol. Notably, OPM‐MEG data acquisition and interpretation were performed independently by two blinded neurosurgeons who were excluded from clinical decision‐making processes, with findings excluded from therapeutic decision‐making (Figure 1).

FIGURE 1.

Flow diagram of the study. Magnetic source imaging was conducted using the following four steps: (1) review raw magnetoencephalographic data and perform noise reduction; (2) visually identify and mark interictal epileptiform discharges (IEDs) belonging to the same group, with each group of IEDs independently detected by two professionals; (3) create an individual head model for each patient and perform alignment procedure; and (4) localize every spike's source position and orientation by using FieldTrip's ft_dipolefitting function. ECoG, electrocorticography; EEG, electroencephalography; MDT, multidisciplinary team; MEG, magnetoencephalography; MRI, magnetic resonance imaging; OPM, optically pumped magnetometer; PET, positron emission tomography; SEEG, stereo‐EEG.

The inclusion criteria were as follows: (1) admission diagnosis of refractory epilepsy, with planned surgical treatment, and requiring iEEG, including SEEG and ECoG; (2) age between 6 and 75 years; and (3) ability to understand, cooperate with examinations, and sign an informed consent form. Exclusion criteria included the following: (1) MRI fitting failure or missing; (2) clear diagnosis of psychiatric disorders or uncontrolled psychiatric symptoms, making it impossible to understand and cooperate with the examination; and (3) patient had no prior diagnosis of psychiatric disorders or uncontrolled psychiatric symptoms, but during the examination, specific situations such as fear and involuntary movements arose, or the patient reported an inability to continue the examination.

A total of 80 patients were screened, and 12 of them were excluded for the following reasons: three presented with artifacts in the MRI three‐dimensional (3D)‐T1 sequence that compromised OPM‐MEG source reconstruction; four showed suboptimal compliance during scanning, resulting in severe motion artifacts that precluded source localization; two had severe structural brain abnormalities that prevented neuroanatomical atlas registration and ultimately did not undergo intraoperative ECoG monitoring; two subsequently underwent disconnection surgery instead of the planned procedures; and one lacked raw MRI 3D‐T1 sequence data due to personal reasons.

A total of 68 participants were included in the analysis cohort. Among these, 31 individuals had clear preoperative localization and directly underwent resection surgery. During the operations, ECoG monitoring was performed to determine the localization of the lesions. Another 31 participants exhibited unclear preoperative localization and underwent SEEG implantation for more precise localization, followed by ablation surgery. The remaining six participants, who also presented with unclear preoperative localization, received SEEG implantation for diagnostic evaluation. However, instead of undergoing ablation surgery, they ultimately opted for lesion resection (Figure S3).

In clinical efficacy evaluation, we followed 51 patients for more than 12 months postoperatively. Surgical outcomes were systematically evaluated using dual standardized metrics: the International League Against Epilepsy (ILAE) classification system ¹⁴ and Engel Epilepsy Surgery Outcome Scale. ¹⁵

2.3. MEG recordings

All MEG recordings were produced by a 128‐channel (corresponding to 64 OPM sensors working at dual mode) OPM‐MEG system (Marvel MEG 128, Beijing X‐MAGTECH Technologies Limited) with a magnetic field measurement sensitivity better than 15 fT/Hz^1/2. The system employs a magnetic field control solution combining passive shielding with active compensation. Passive shielding is provided by a multilayered permalloy shielding cylinder, whereas the active compensation coil system performs real‐time compensation of residual magnetic fields. Under normal conditions, residual magnetic fields within a 50‐cm radius centered on the helmet will be maintained below ±.5 nT. Before scanning was performed, the relative positional relationship between participants' facial features and helmets were accurately digitized using a 3D structured light scanner for imaging alignment (EinScan H, SHINING 3D). Recordings were performed in supine position and lasted approximately 90 min with a 1000‐Hz sampling rate. An MEG technician monitored the whole process to ensure the safety of participants in the case of sudden seizures or other adverse events. During the data recording, patients were instructed to maintain a closed‐eyes posture, with the duration of the experiment being predetermined at 1.5 h. It should be noted that no supplementary measures (such as sleep deprivation or the administration of sedatives) were implemented to facilitate sleep during the recording process.

2.4. MEG data analysis

First, MEG recordings were denoised by a synthetic gradiometer to significantly reduce low‐frequency environmental magnetic field drift and common mode noise between the sensor arrays. ¹⁶ , ¹⁷ , ¹⁸ Then the data were notched at 50 Hz and its harmonics and filtered into 2~70 Hz by a band‐pass filter. Independent component analysis procedure was performed to eliminate obvious artifacts like heartbeats and eye movements. Two blinded neurosurgeons, who were board‐certified and have received specialized training in clinical neurophysiology and epilepsy with more than 5 years of experience in interpreting presurgical evaluations, browsed all the recordings and identified IEDs by marking the spikes' peak points as epileptic events (Figure 2).

FIGURE 2.

Methodologic flowchart of magnetic source imaging. MEG, magnetoencephalographic.

To perform a dipole fitting procedure to these epileptic events, all the recordings were aligned between sensor arrays and native MRI (Discovery MR 750, GE Medical Systems, 3‐T) through optical scanning. Single‐shell head models were utilized calculated from participants' native space to accurately constrain the potential source points with lead fields computed. In the present study, the ft_dipolefitting function in FieldTrip was employed to perform dipole fitting on the peak times of fast wave components within each identified IED. During the fitting process, the topography map of each dipole was carefully examined, and only those channels exhibiting distinct dipole distributions were included in the calculations. Additionally, dipoles with a goodness of fit value below 80% were excluded from the fitting results. Furthermore, no spike averaging was performed for IEDs exhibiting identical patterns.

2.5. SEEG and ECoG protocols

The indications for SEEG include the following: (1) confirmed RE with a reasonable hypothesis of the epilepsy origin and propagation network formed through noninvasive evaluation; (2) contradictory localization information between seizure symptoms, electrophysiological patterns, and anatomical structures, preventing accurate localization of the EZ; (3) the EZ is located in or near critical functional areas, requiring precise brain functional localization to accurately determine the surgical resection range and avoid or minimize functional impairment; and (4) the EZ is limited in scope or located deep within the brain, with the intention of performing radiofrequency ablation after SEEG electrode placement. Patients must meet condition 1 and at least one of conditions 2, 3, or 4. Patients meeting these criteria underwent SEEG. The placement of the electrodes was based solely on clinical reasons and differed across patients.

The indications for ECoG include the following: (1) confirmed RE with a well‐characterized EZ identified through noninvasive evaluation, requiring intraoperative verification of electrophysiological boundaries for resection; (2) EZ adjacent to or overlapping with eloquent cortical areas (e.g., motor, language, or sensory cortex), necessitating real‐time functional mapping during surgery to optimize resection margins and preserve neurological function; and (3) suspected EZ (e.g., focal cortical dysplasia, tumor‐related epilepsy) with ill‐defined electrophysiological boundaries on preoperative imaging or scalp EEG, necessitating intraoperative guidance for tailored resection. Patients must meet at least one of conditions 1–3. ECoG was performed intraoperatively in eligible patients to refine surgical strategy, delineate resection margins, and balance seizure control with functional preservation.

SEEG signal acquisition was done using NSH0256 EEG amplifier (Neuracle Technology) and the Natus Quantum (PK1174) amplifier (Natus Medical Incorporated) at a typical sampling frequency of 1 or 2 kHz. Each subject had 4–8 electrodes implanted based on the results of the first‐phase noninvasive evaluation, with simultaneous video recording. ECoG signal acquisition was done using the Natus Quantum amplifier (Natus Medical Incorporated) to determine the EZ during the operation.

2.6. EZ localization

Thirty‐one patients underwent SEEG. SEEG signals underwent preprocessing and source localization via Brainstorm, ¹⁹ with electrode reconstruction achieved through postimplantation computed tomography (CT) and preoperative 3D T1‐weighted MRI coregistration to verify spatial positioning. Three 10‐min epochs of sleep‐state electrophysiological data (30 min total per subject), collected ≥1 h from the nearest seizure events, ²⁰ were analyzed. Following artifact rejection of compromised channels and notch filtering at 50‐Hz harmonics to mitigate electrical interference, data were bandpass‐filtered to 1–100 Hz (2000‐Hz sampling rate). Two board‐certified neurosurgeons collaboratively annotated interictal epileptiform discharges prior to source reconstruction. Localization results were registered to the Automated Anatomical Labeling (AAL) ²¹ atlas via Advanced Normalization Tools (ANTs), ²² segmented, and visualized in 3D Slicer. ²³ Based on the results of quantitative analysis, the original surgical team incorporated comprehensive medical record information and electrophysiological records and rectified the results via visual analysis methods.

For SEEG, the seizure onset zone (SOZ) was defined by consensus between two board‐certified epileptologists through detailed visual analysis of the iEEG recordings. The analysis prioritized the electrographic pattern at seizure onset, characterized by the presence of low‐voltage fast activity, repetitive spike/polyspike discharges, or rhythmic sinusoidal activity that clearly evolved in frequency, amplitude, and spatial distribution. ²⁴ , ²⁵ , ²⁶ Cortical regions exhibiting this definitive ictal onset pattern were annotated as the primary SOZ. Supporting evidence from IEDs that were spatially concordant with the ictal onset region and the anatomical location of any structural lesions identified on preoperative MRI were also integrated to refine the SOZ delineation. This visually defined SOZ, representing the cortical area deemed necessary and sufficient for seizure generation, served as the reference standard for EZ localization in the SEEG cohort.

Thirty‐seven patients underwent intraoperative ECoG monitoring with selective regional sampling. The resected regions, anatomically delineated as EZ by two neurosurgeons through postoperative CT acquired within 24 h with itk‐SNAP version 4.2.0 (http://www.itksnap.org), ²⁷ were subsequently coregistered with preoperative 3D T1‐weighted MRI sequences and the AAL neuroanatomical atlas using ANTs registration. ²² These volumetric datasets underwent multiplanar segmentation and 3D volumetric reconstruction within 3D Slicer for spatial validation. The localization results were also corrected via visual analysis by the original surgical team.

For ECoG, the EZ was defined intraoperatively as the cortical region exhibiting persistent or frequent interictal epileptiform discharges and/or electrographic seizure activity that was deemed eloquent by the operating neurosurgeon and clinical neurophysiologist. ²⁸ The boundaries of the resection were planned to encompass this electrophysiologically abnormal cortex while preserving functional areas identified by cortical stimulation mapping. ²⁹

2.7. Brain regions

To evaluate localization concordance, we implemented the AAL atlas template for cerebral parcellation, segmenting the brain into 120 distinct subregions. Following exclusion of cerebellar structures, 94 cortical and subcortical regions were included in our comparative spatial analysis between OPM‐MEG‐identified irritative zone (IZ) and clinically validated EZ (Figure S1).

In multiregional epileptogenic involvement, threshold‐based classification was applied; should a single region exhibit IED counts exceeding twofold the cumulative total of all other implicated regions, it was designated the primary IZ for inclusion while excluding secondary regions. All coinvolved territories were retained when no dominant region met this electrophysiological dominance criterion. ⁵ Definitive IZ received positive annotations, whereas nonepileptogenic areas were systematically labeled negative through this binary classification schema.

2.8. Reference standard

This study implemented dual reference standards to evaluate both agreement and clinical efficacy of OPM‐MEG. Given the absence of an ideal source imaging benchmark during presurgical evaluation, ³⁰ we chose the EZ localized by iEEG as the primary reference standard to assess the agreement between the localization by OPM‐MEG and that by iEEG. ⁵ Based on iEEG, after conducting visual analysis and correction, we determined the EZ localization and designated it as the reference standard for evaluating the regional concordance of OPM‐MEG in EZ delineation. However, it should be noted that the iEEG analysis results encompass two distinct localization methods (SEEG and ECoG), which might exhibit disparities in treatment effects. Consequently, we also carried out separate statistical analyses on these two subgroups. Operational definitions were established; agreement required OPM‐MEG localization within resection volumes at the sublobar anatomical level, whereas discordance indicated nonoverlapping localization across hemispheric or lobar divisions (Figure S2). ³¹ , ³²

To provide a quantitative measure of spatial agreement beyond categorical concordance, the Euclidean centroid distance was calculated between the OPM‐MEG dipole cluster and the reference standard localization. For patients in the ECoG cohort, the reference standard was the centroid of the surgically resected volume. For patients in the SEEG cohort, the reference standard was the centroid of the SEEG‐defined EZ. Validation metrics included comparative analysis of EZ centroid distances between concordant and discordant cohorts, ensuring methodological robustness in agreement classification. A dipole cluster was formally defined as such only when the number of dipoles belonging to the same mode reached a minimum of five, at which point it was incorporated into subsequent analysis. Each individual dipole and the distance between the center of mass of a dipole cluster and any other dipole was considered within the cluster if it did not exceed 15 mm.

To evaluate the diagnostic value of OPM‐MEG, this study prospectively analyzed 51 postoperative patients with ≥12‐month follow‐up, stratifying outcomes according to both ILAE and Engel classification systems. Prognostic categorization designated ILAE class 1 and Engel class I as favorable outcomes (seizure‐free), with all other classifications systematically classified as suboptimal outcomes (not seizure‐free).

2.9. Definition of prognostic analysis and diagnostic classification

To evaluate the prognostic value of OPM‐MEG localization for surgical outcome, an analysis was performed on the subset of patients (n = 51) who underwent a potentially curative intervention (either cortical resection guided by ECoG or thermocoagulation ablation guided by SEEG) and had postoperative follow‐up of ≥12 months. For this analysis, the reference standard for a “positive” localization was defined as a combination of the following: (1) the treated brain region (i.e., the resection zone for ECoG patients or the ablation zone for SEEG patients); and (2) a favorable surgical outcome, defined as seizure freedom (ILAE class 1 or Engel class I) at the last follow‐up. A “negative” localization reference was defined as either discordance between OPM‐MEG localization and the treated zone, or concordance but with an unfavorable surgical outcome (ILAE class ≥2 or Engel class ≥II).

Based on this composite reference standard, the following diagnostic categories were defined for the calculation of sensitivity, specificity, PPV, NPV, accuracy, and OR:

1. True positive (TP): OPM‐MEG localization was concordant with the treated zone, and the patient was seizure‐free (ILAE 1/Engel I).

2. False positive (FP): OPM‐MEG localization was concordant with the treated zone, but the patient was not seizure‐free (ILAE ≥2/Engel ≥II).

3. True negative (TN): OPM‐MEG localization was discordant with the treated zone, and the patient was not seizure‐free (ILAE ≥2/Engel ≥II).

4. False negative (FN): OPM‐MEG localization was discordant with the treated zone, but the patient was seizure‐free (ILAE 1/Engel I).

2.10. Outcome measures

To assess consistency, we calculated the following metrics. Using the EZ localization as the reference standard, we determined the agreement between OPM‐MEG and EZ localization at the sublobar level. We calculated the proportion of sublobar regions that met the reference standard as well as the overall consistency with the reference standard, and carried out subgroup analyses for the SEEG group and the ECoG group.

Sensitivity, specificity, PPV, NPV, accuracy, and OR for predicting surgical outcome were calculated based on the definitions provided in the preceding Definition of Prognostic Analysis and Diagnostic Classification section. We used the following formulas:

$$\mathrm{\text{Sensitivity}}=\mathrm{TP}/\left(\mathrm{TP}+\mathrm{FN}\right)$$

$$\mathrm{\text{Specificity}}=\mathrm{TN}/\left(\mathrm{TN}+\mathrm{FP}\right)$$

$$\mathrm{PPV}=\mathrm{TP}/\left(\mathrm{TP}+\mathrm{FP}\right)$$

$$\mathrm{NPV}=\mathrm{TN}/\left(\mathrm{TN}+\mathrm{FN}\right)$$

$$\mathrm{\text{Accuracy}}=\left(\mathrm{TN}+\mathrm{TP}\right)/\left(\mathrm{TP}+\mathrm{FP}+\mathrm{TN}+\mathrm{FN}\right)$$

$$\mathrm{\text{OR}}=\left(\mathrm{TP}\times \mathrm{TN}\right)/\left(\mathrm{FP}\times \mathrm{FN}\right)$$

2.11. Statistical analysis

To analyze the agreement between OPM‐MEG and EZ localization, we computed Gwet AC1 (first‐order agreement coefficient) and denoted the AC1 as κ. ³³ The advantage of the Gwet AC1 method is that it provides a more stable measure of agreement and does not rely on the PPV, whereas the traditional Cohen kappa method may be less accurate in this context, particularly as the data distribution may be subject to category bias. ³³ The concordance was assessed at the sublobar anatomical level, defined by the AAL atlas. For each patient, the localization results from OPM‐MEG and EZ localization were mapped onto the 94 cortical and subcortical regions. A case was classified as "concordant" if there was any overlap between the set of regions identified by OPM‐MEG and the set identified by iEEG. To handle cases with multiple localizations, we applied a threshold‐based rule detailed in the Brain Regions section; if a single region's cluster count exceeded twice the cumulative total of all other implicated regions, it was designated the primary focus for analysis; otherwise, all coinvolved regions were considered.

The agreement between the two methods was interpreted according to conventional groupings: poor (κ < 0), slight (κ = .01–.2), fair (κ = .21–.4), moderate (κ = .41–.6), substantial (κ = .61–.8), and almost perfect agreement (κ > .8). ³⁴ These analyses were completed using the irrCAC package ³⁵ in R (v4.5.0; R Core Team, 2025. https://www.R‐project.org/). Centroid distances between concordant and discordant cohorts were computed in MATLAB (version 23.2.0 [R2023b], MathWorks, https://www.mathworks.com). The Mann–Whitney U nonparametric test (two‐tailed) was used to conduct the difference test on the centroid distances of different groups. This was calculated using the rstatix packages. ³⁶ Sensitivity, specificity, localization accuracy, PPV, and NPV, along with their 95% confidence intervals (CIs), were calculated using the binom ³⁷ and epiR ³⁸ packages. All reported p‐values underwent false discovery rate correction for multiple comparisons.

3. RESULTS

3.1. Demographic and clinical characteristics of patients

As detailed in Table 1, the cohort's demographic profile included 68 right‐handed patients (26 female; mean age = 28 years, range = 6–60) diagnosed with RE through comprehensive preoperative assessments including clinical history, OPM‐MEG, video‐EEG monitoring, MRI/PET, and MDT evaluation. The MRI reports indicated that, in total, there were 58 cases of single focus, two cases of multiple foci, and eight cases with negative MRI reports. Surgical candidates included 63 patients with quantifiable interictal discharges (five exhibited nonlocalizable OPM‐MEG profiles). With magnetic source imaging (MSI) localization, 57 cases were identified as having a single focus, and six cases were identified as having multiple foci. The cohort exhibited IED counts ranging from 3 to 162 (median = 13, interquartile range [IQR] = 7–30.5, inclusive of MEG‐negative cases). Illustrative presurgical MSI results from three representative cases appear in Figure 3.

TABLE 1.

Summary of patient demographic and clinic characteristic.

Characteristic	n (%)
Sex
Male	42 (62)
Female	26 (68)
Handedness
Right	68 (100)
Left	0
Indeterminate	0
Age, mean, years	28
<10	3 (4)
10–18	12 (18)
>18	53 (78)
MRI findings
Single focus	58 (86)
Multiple foci	2 (3)
Negative	8 (11)
MSI localization
Single focus	57 (84)
Multiple foci	6 (9)
No interictal discharges	5 (7)
MEG IED, median (interquartile range)	13 (7–30.5)
<10	26 (38)
10–20	19 (28)
>20	23 (34)
Type of surgery
SEEG	31 (46)
ECoG	37 (54)
SEEG IEDs, median (interquartile range)	46 (15–101)
<10	3 (10)
10–20	3 (10)
>20	25 (80)
Clinical localization
Temporal	51 (75)
Extratemporal	17 (25)
ILAE, mean follow‐up time, months	17.9
1	28 (55)
≥2	23 (45)
Engel, mean follow‐up time, months	17.9
I	37 (73)
≥II	14 (27)

Note: This table summarizes the baseline characteristics of the 68 prospectively enrolled patients with refractory epilepsy. Data are presented as n (%) or as indicated.

Abbreviations: ECoG, electrocorticography; IED, interictal epileptiform discharge; ILAE, International League Against Epilepsy; MEG, magnetoencephalography; MRI, magnetic resonance imaging; MSI, magnetic source imaging; SEEG, stereo‐electroencephalography.

FIGURE 3.

Illustration of optically pumped magnetometer magnetoencephalography (OPM‐MEG) magnetic source imaging (MSI) localization results. In Case A, MSI results show interictal epileptiform discharges were primarily located in the right temporal lobe. In Case B, the MSI results show that interictal epileptic discharges were mainly centered in the left insula and left temporal lobe. In Case C, MSI results show that interictal epileptic discharges were predominantly found in the right frontal lobe. LH, left hemisphere; R, right; RH, right hemisphere.

The cohort comprised 31 patients undergoing SEEG‐guided ablation (IEDs: 5–219, median = 46, IQR = 15–101, no subsequent resections) and 37 patients receiving intraoperative ECoG‐directed resections, including five who underwent prior SEEG‐guided ablation (excluded from the SEEG cohort). The study population (N = 68) included 51 temporal lobe and 17 extratemporal epilepsy cases, all prospectively enrolled in postoperative surveillance. Among 51 patients completing ≥12‐month follow‐up (mean = 17.9 months), 28 (55%) achieved ILAE class 1 seizure freedom, whereas 23 (45%) experienced persistent seizures. By Engel criteria, 37 patients (73%) achieved Engel class I status (seizure‐free), with 14 (27%) demonstrating ongoing seizure activity.

3.2. Concordance with iEEG and comparative analysis of subgroups

The agreement between OPM‐MEG and iEEG for localization was quantitatively assessed. As summarized in Table 2, whole‐brain analysis demonstrated a high spatial agreement of 90.0% between the two modalities, with an almost perfect agreement coefficient (κ = .885, 95% CI = .875–.894). Stratification by brain region revealed a concordance rate of 80.1% (κ = .723, 95% CI = .688–.758) in temporal lobes, indicating substantial agreement, which was lower than the 92.0% concordance (κ = .926, 95% CI = .911–.928) observed in extratemporal regions, reflecting almost perfect agreement. Subgroup analysis based on the iEEG method showed highly consistent agreement rates for both SEEG (89.9%, κ = .884) and ECoG (90.0%, κ = .885) subgroups.

TABLE 2.

Agreement between OPM‐MEG and EZ localization.

Reference standard	EZ localization
Outcome measure	κ	PA, %
Full brain	.885 (.875–.894)	90.0
Temporal	.723 (.688–.758)	80.1
Extemporal	.926 (.911–.928)	92.0
SEEG	.884 (.871–.898)	89.9
ECoG	.885 (.872–.898)	90.0

Note: This table presents the agreement analysis between OPM‐MEG localization and EZ localization. Agreement was assessed at the sublobar anatomical level (94 cortical and subcortical regions of the Automated Anatomical Labeling atlas). The metric PA represents the proportion of cases where OPM‐MEG localization overlapped with the EZ localization. The agreement coefficient κ represents Gwet AC1 (interpreted as: slight: .01–.2, fair: .21–.4, moderate: .41–.6, substantial: .61–.8, almost perfect: >.8). The analysis is shown for the full brain (all regions), and subgroup analyses are provided for temporal lobe, for extratemporal lobe, and based on the different methods (SEEG or ECoG). Values for κ are presented with 95% confidence intervals in parentheses.

Abbreviations: ECoG, electrocorticography; EZ, epileptogenic zone; MEG, magnetoencephalography; OPM, optically pumped magnetometer; PA, percentage agreement; SEEG, stereoelectroencephalography.

Spatial analysis of centroid distances provided further validation. The mean centroid distance between OPM‐MEG‐defined IZ and iEEG‐defined EZ was significantly shorter in concordant cases (1.98 ± .93 cm) compared to discordant cases (4.12 ± 2.66 cm, p < .05; Figure 4A). However, no significant difference in centroid distances was observed between temporal and extratemporal epilepsy cohorts (p > .05; Figure 4B). Analysis of the SEEG and ECoG subgroups provided additional insights (Figure 5). In both the SEEG and ECoG subgroups, the centroid distance was significantly shorter in concordant cases than in discordant cases (p < .05; Figure 5A,B). When comparing the two iEEG methods directly, no significant difference in centroid distances was found for patients with concordant localization and discordant localization (Figure 5C,D).

FIGURE 4.

Comparative analysis of Euclidean centroid distances. (A) Comparison of Euclidean centroid distances between the concordant and discordant groups, showing significant differences. (B) Comparison of Euclidean centroid distances between patients with temporal lobe epilepsy and those with extratemporal lobe epilepsy, showing no significant difference. FDR, false discovery rate; ns, not significant. *Statistically significant.

FIGURE 5.

Comparative analysis of Euclidean centroid distances in different subgroups. (A) Comparison of Euclidean centroid distances between the concordant and discordant subgroups within the stereoencephalography (SEEG) group, showing a significant difference. (B) Comparison of Euclidean centroid distances between the concordant and discordant subgroups within the electrocorticography (ECoG) group, showing a significant difference. (C) Comparison of Euclidean centroid distances between the ECoG and SEEG groups for patients with concordant localization, showing no significant difference. (D) Comparison of Euclidean centroid distances between the ECoG and SEEG groups for patients with discordant localization, showing no significant difference. FDR, false discovery rate. *Statistically significant.

3.3. Diagnostic value of OPM‐MEG localization

The diagnostic value of OPM‐MEG for predicting postoperative outcomes was evaluated against two reference standards—the ILAE and Engel classification systems—with results detailed in Table 3. Using the ILAE criteria, OPM‐MEG demonstrated a sensitivity of 85.7% (95% CI = 67.3%–96.0%), specificity of 65.2% (95% CI = 42.7%–83.6%), PPV of 75.0% (95% CI = 56.6%–88.5%), NPV of 78.9% (95% CI = 54.4%–93.9%), and localization accuracy of 76.5% (95% CI = 62.5%–87.2%). The diagnostic OR was 11.25 (95% CI = 2.88–43.9). Under the Engel criteria, the performance metrics were sensitivity of 73.0% (95% CI = 55.9%–86.2%), specificity of 64.3% (95% CI = 35.1%–87.2%), PPV of 84.4% (95% CI = 67.2%–94.7%), NPV of 47.4% (95% CI = 24.4%–71.1%), accuracy of 70.6% (95% CI = 56.2%–82.5%), and OR of 4.86 (95% CI = 1.31–18.00).

TABLE 3.

Diagnostic value of OPM‐MEG localization in different standards and groups.

Reference standard	Prognosis
Outcome measure	Sensitivity, %	Specificity, %	PPV, %	NPV, %	Accuracy, %	Odds ratio
ILAE	85.7 (67.3–96.0)	65.2 (42.7–83.6)	75.0 (56.6–88.5)	78.9 (54.4–93.9)	76.5 (62.5–87.2)	11.25 (2.88–43.9)
SEEG_ILAE	77.8 (40.0–97.2)	61.5 (31.6–86.1)	58.3 (27.7–84.8)	80.0 (44.4–97.5)	68.2 (45.1–86.1)	5.60 (.81–38.5)
ECoG_ILAE	89.5 (66.8–98.7)	70.0 (34.8–93.3)	85.0 (62.1–96.8)	77.8 (40.0–97.2)	82.8 (64.2–94.2)	19.80 (2.70–145.68)
Engel	73.0 (55.9–86.2)	64.3 (35.1–87.2)	84.4 (67.2–94.7)	47.4 (24.4–71.1)	70.6 (56.2–82.5)	4.86 (1.31–18.00)
SEEG_Engel	57.1 (28.9–82.3)	50.0 (15.7–84.3)	66.7 (34.9–90.1)	40.0 (12.2–73.8)	54.5 (32.2–75.6)	1.30 (.23–7.63)
ECoG_Engel	82.6 (61.2–95.0)	83.3 (35.9–99.6)	95.0 (75.1–99.9)	55.6 (21.2–86.3)	82.8 (64.2–94.2)	23.75 (2.15–262.48)

Note: The diagnostic value of OPM‐MEG was evaluated using postoperative seizure freedom at ≥12 months as the reference standard. "Prognosis" is defined as a favorable outcome (seizure‐free), corresponding to ILAE class 1 or Engel class I. Metrics include sensitivity, specificity, PPV, NPV, accuracy, and odds ratio, presented with 95% confidence intervals. Performance is stratified by the outcome classification system (ILAE and Engel) and by the intracranial recording method (SEEG and ECoG) to address inherent differences in the patient populations selected for each technique. This table provides detailed metrics, and key results are also visualized in Figure 6 for comparative clarity.

Abbreviations: ECoG, electrocorticography; ILAE, International League Against Epilepsy; MEG, magnetoencephalography; NPV, negative predictive value; OPM, optically pumped magnetometer; PPV, positive predictive value; SEEG, stereoelectroencephalography.

Subgroup analysis based on the iEEG method revealed distinct performance profiles between SEEG and ECoG cohorts (Figure 6). For the SEEG subgroup under ILAE criteria, sensitivity was 77.8% (95% CI = 40.0%–97.2%), specificity was 61.5% (95% CI = 31.6%–86.1%), PPV was 58.3% (95% CI = 27.7%–84.8%), NPV was 80.0% (95% CI = 44.4%–97.5%), accuracy was 68.2% (95% CI = 45.1%–86.1%), and OR was 5.60 (95% CI = .81–38.5). In contrast, the ECoG subgroup under ILAE criteria showed higher metrics: sensitivity of 89.5% (95% CI = 66.8%–98.7%), specificity of 70.0% (95% CI = 34.8%–93.3%), PPV of 85.0% (95% CI = 62.1%–96.8%), NPV of 77.8% (95% CI = 40.0%–97.2%), accuracy of 82.8% (95% CI = 64.2%–94.2%), and OR of 19.80 (95% CI = 2.70–145.68). Comparative analysis of these clinical metrics between SEEG and ECoG subgroups under ILAE classification showed no statistically significant differences (all p > .05; Figure 6A).

FIGURE 6.

Diagnostic value comparative analysis of stereoelectroencephalography (SEEG) and electrocorticography (ECoG). (A) Comparison of clinical performance metrics (sensitivity, specificity, positive predictive value [PPV], negative predictive value [NPV], accuracy) between SEEG and ECoG subgroups under the International League Against Epilepsy (ILAE) outcome classification. The error bars indicate 95% confidence intervals (CIs). All comparisons between subgroups showed no statistically significant difference (all p > .05). (B) Comparison of the same clinical metrics between SEEG and ECoG subgroups under the Engel outcome classification. All comparisons between subgroups showed no statistically significant difference (all p > .05). (C) Forest plot comparing the diagnostic odds ratio (OR) for localization concordance between SEEG and ECoG subgroups, analyzed under the ILAE outcome criteria. The vertical dashed line indicates the null value (OR = 1). The point estimates and 95% CIs for both subgroups are shown, with no statistically significant difference in ORs between the groups (p = .557). (D) Forest plot comparing the diagnostic OR for localization concordance between SEEG and ECoG subgroups, analyzed under the Engel outcome criteria. The pattern and statistical conclusion are consistent with those observed under ILAE criteria (p = .229). FDR, false discovery rate.

A similar pattern was observed under the Engel criteria. The SEEG subgroup had a sensitivity of 57.1% (95% CI = 28.9%–82.3%), specificity of 50.0% (95% CI = 15.7%–84.3%), PPV of 66.7% (95% CI = 34.9%–90.1%), NPV of 40.0% (95% CI = 12.2%–73.8%), accuracy of 54.5% (95% CI = 32.2%–75.6%), and OR of 1.30 (95% CI = .23–7.6). The ECoG subgroup under Engel criteria demonstrated superior performance: sensitivity of 82.6% (95% CI = 61.2%–95.0%), specificity of 83.3% (95% CI = 35.9%–99.6%), PPV of 95.0% (95% CI = 75.1%–99.9%), NPV of 55.6% (95% CI = 21.2%–86.3%), accuracy of 82.8% (95% CI = 64.2%–94.2%), and OR of 23.75 (95% CI = 2.15–262.48). Comparative analysis of these clinical metrics between SEEG and ECoG also showed no statistically significant differences (all p > .05; Figure 6B). Forest plots comparing the diagnostic ORs for both ILAE (Figure 6C) and Engel (Figure 6D) criteria showed no statistically significant differences between the SEEG and ECoG subgroups (ILAE: p = .557; Engel: p = .229).

4. DISCUSSION

Our study validates the capability of OPM‐MEG in localizing EZ and represents the largest prospective study to date utilizing surgical prognosis as a reference standard. Our findings demonstrate almost perfect agreement (κ > .8) between OPM‐MEG‐defined IZ and iEEG‐defined EZ at a sublobar anatomical resolution. The assessment of its diagnostic value, benchmarked against postoperative outcomes with ≥12‐month follow‐up, demonstrated robust diagnostic efficacy under both ILAE (sensitivity 85.7%, specificity 65.2%) and Engel (sensitivity 73.0%, specificity 64.3%) classification standards. These metrics align closely with previous SQUID‐MEG investigations, reinforcing OPM‐MEG's clinical equivalence in presurgical epilepsy evaluation. ³² , ³⁹ , ⁴⁰ , ⁴¹

4.1. Value of MEG in the presurgical workflow

In the preoperative evaluation of epilepsy surgery, MEG—a noninvasive functional imaging technique—provides unique and complementary information for localizing the EZ and guiding clinical decision‐making and has been widely adopted. The prospective study conducted by Mohamed et al. demonstrated that MSI influenced the surgical management of 25% of the 32 patients, leading to either direct surgical intervention or modifications in the SEEG protocol. ⁴² Ultimately, 21 patients who underwent surgery achieved favorable outcomes. In recent years, MEG has seen continuous advancements in both algorithmic development and hardware technologies. Specifically, in the field of algorithms, MEG has demonstrated promising potential in the analysis of high‐frequency oscillations (HFOs) and in the integration of multimodal data. Owen et al. proposed that abnormal mapping based on MEG band power can identify three mechanisms underlying neocortical epilepsy surgery failure: mislocalization, partial resection, and insufficient impact on the overall epileptogenic abnormality. ⁴³ The predictive model they developed demonstrates a reasonable ability to determine whether seizure freedom will be achieved following surgery, with area under the curve (AUC) values ranging from .64 to .80. Li et al. integrated MEG with SEEG and demonstrated a consistent localization of HFOs between the two modalities, with 70% of patients exhibiting either complete or partial concordance. ⁴⁴ Moreover, a significantly higher dipole resection rate was observed in patients who achieved seizure freedom following surgery. The developed regression model exhibited a predictive accuracy with an AUC of .81. Yan et al. further demonstrated that integrating ripples (80–200 Hz) with PET or MEG can enhance the accuracy of prognosis prediction and that a multimodal approach outperforms single‐modality methods. ⁴⁵ In terms of hardware, current research indicates that OPM‐MEG achieves an SNR comparable to or better than that of SQUID‐MEG, ¹¹ , ¹² , ⁴⁶ , ⁴⁷ although its construction and maintenance costs are significantly lower. ⁴⁸ However, it is important to note that the data regarding the service life and long‐term cost of the OPM‐MEG sensors have not been comprehensively determined, and additional longitudinal studies are required.

Whether to adopt OPM‐MEG as a novel noninvasive tool for EZ localization in clinical practice does not hinge on whether it can supplant certain methods, but rather on whether it can ultimately aid in determining the location of the lesion and facilitating surgical decision‐making. Taking into account the cost and universality, long‐term video‐EEG remains the preferred method for EZ localization. Previous retrospective studies have also demonstrated that there is no significant difference in the clinical diagnostic efficacy between MEG and high‐density EEG. ³⁹ , ⁴⁹ However, a unique application advantage of MEG over EEG is in patients who have had previous craniotomies and surgical resections. ⁵⁰ In these patients, electrical potentials recorded at the scalp are distorted by cranial defects and the biophysical disturbances of brain‐volume changes. This situation is particularly applicable in patients with recurrent seizures after epilepsy surgeries who are contemplating repeat surgery. It has been confirmed with SEEG that MEG can be successful in localizing postsurgical epileptiform disturbances. ⁵¹ In our study, there was also a patient (S10) of this type who suffered a skull defect due to a car accident. Because the defect was not repaired in a timely manner, it led to brain tissue herniation and subsequently epilepsy. When scalp EEG was not applicable, OPM‐MEG successfully achieved the localization of the lesion, and the result was consistent with the intraoperative localization of the epileptic focus.

Compared with other major diagnostic methods, MEG holds a unique position in functional localization, particularly for epilepsy patients with negative MRI results. Detecting these lesions represents one of the primary concerns in the clinical application of functional epilepsy imaging, including PET, single photon emission computed tomography, and functional MRI (fMRI). When compared to these diagnostic modalities, MEG occupies a distinct functional niche. fMRI is proficient in identifying blood flow changes, and PET excels in revealing hypometabolic regions, whereas MEG directly measures neuroelectrophysiological activities. MEG can directly establish a correlation between the epileptogenic significance and focal functional abnormalities in imaging, thus complementing anatomical and metabolic imaging.

4.2. Considerations on reference standards and subgroup analysis

A critical methodological consideration involves the use of iEEG reference standards for EZ localization. It is important to note that SEEG and ECoG represent distinct approaches for defining the EZ, with SEEG often used for deep‐seated foci and ECoG for cortical surface mapping. This inherent difference could potentially introduce variability when assessing concordance with a noninvasive modality like OPM‐MEG. To address this, we conducted separate subgroup analyses for SEEG and ECoG cohorts. Reassuringly, the agreement between OPM‐MEG and iEEG was highly consistent between the two subgroups (SEEG: 89.9%, κ = .884; ECoG: 90.0%, κ = .885; Table 2), indicating that the OPM‐MEG localization performance was not substantially influenced by the choice of iEEG technique. However, in the assessment of diagnostic value (Table 3, Figure 6), although there was no significant difference, metrics such as accuracy and diagnostic OR tended to be slightly higher in the ECoG subgroup compared to the SEEG subgroup, particularly under the Engel classification. This discrepancy may be attributed to fundamental differences in the patient populations selected for each technique. ECoG is typically employed when a well‐defined hypothesis exists from noninvasive comprehensive assessment, potentially leading to a cohort with more straightforward EZs that are inherently easier to localize correctly with any modality, including OPM‐MEG. In contrast, SEEG is used in more complex cases with ambiguous noninvasive findings, which may present a greater localization challenge. This patient selection bias, inherent to the clinical application of these techniques, likely explains the observed trend toward better performance metrics in the ECoG subgroup.

4.3. Spatial heterogeneity and temporal lobe concordance

In assessing concordance, we first established statistically significant disparities in centroid distances between concordant and discordant groups (1.98 ± .93 cm vs. 4.12 ± 2.66 cm, p < .05; Figure 4A), thereby validating the reliability of our classification framework. Although OPM‐MEG demonstrated almost perfect agreement overall, temporal lobe localization exhibited substantially lower agreement (80.1%, κ = .723) compared to extratemporal regions (92.0%, κ = .926). This observed reduction in temporal lobe concordance can be attributed to the larger effective sensor‐to‐source distance. ⁵⁰ According to the inverse‐square law, MEG signal strength decays with the square of the distance from the source. The mesial temporal structures, residing 4–6 cm deep within the cranial fossa, ⁵² are further shielded by the surrounding temporal bone. This anatomical configuration inherently increases the sensor‐to‐source distance, resulting in a lower SNR for deep sources, making temporal lobe signals more susceptible to being masked by background activity and contributing to the reduced spatial agreement. ⁵³ In addition, considering that the helmet employed in this study is a rigid spherical helmet with fixed sensor positions, due to the influence of the human head's shape, the distance between the temporal lobe region and the sensors may be slightly greater than that between the extratemporal lobe region and the sensors. This also serves as a potential cause of the aforementioned significant difference, which requires further verification by wearable OPM‐MEG. ⁴⁸

4.4. Diagnostic value and interpretation of classification systems

OPM‐MEG demonstrates diagnostic efficacy comparable to SQUID‐MEG, with significantly reduced operational costs, affirming its role as a cost‐effective and high‐fidelity tool in presurgical epilepsy evaluation. When interpreting prognostic outcomes, it is critical to acknowledge the inherent disparities between the Engel and ILAE classification systems, which underlie the variations in predictive values observed in our study (Table 3). As established in standards, ¹⁴ , ¹⁵ Engel class I defines success as freedom from disabling seizures, which includes patients with auras (simple partial seizures), whereas ILAE class 1 mandates complete seizure freedom, excluding auras. This fundamental distinction systematically results in a higher proportion of patients classified as Engel I compared to ILAE 1, as evidenced by studies such as Wieser et al. ⁵⁴ (66.9% vs. 57.1%) and Ozkara et al. ⁵⁵ (77.1% vs. 52.7%). In our cohort, this discrepancy is reflected in the differential sensitivity (ILAE: 85.7% vs. Engel: 73.0%) and PPV (ILAE: 75.0% vs. Engel: 84.4%) metrics, where the stricter ILAE criteria may capture a more stringent “cure” cohort, leading to higher specificity but potentially lower sensitivity for success. Subgroup analyses (Figure 6) further highlight these trends, particularly in SEEG and ECoG cohorts, where the Engel system's inclusivity of auras may inflate success rates in complex cases. Additionally, OPM‐MEG's dependence on interictal discharges for IZ localization, which may not fully overlap with the SOZ, ⁵⁶ could conservatively bias PPV estimates. By contextualizing these classification differences, our findings align with broader consensus that Engel and ILAE grades are complementary, with ILAE providing granularity for research and Engel emphasizing clinical functionality.

4.5. Considerations on patient cohort and reference standards

A key methodological consideration in this study pertains to the composition of our patient cohort and the implications for our reference standards. As detailed in Materials and Methods, a substantial portion of our cohort (31 patients) underwent SEEG‐guided thermocoagulation ablation without subsequent cortical resection. For these patients, the localization of the EZ based solely on SEEG findings carries inherent uncertainty, as SEEG implantation was itself necessitated by inconclusive noninvasive localization. Although the ablation zone was guided by the SEEG‐defined SOZ, the absence of a histologically confirmed resection margin introduces ambiguity regarding the true spatial extent of the EZ. Furthermore, the use of 1‐year seizure freedom after thermocoagulation as a surrogate endpoint for successful EZ localization must be interpreted with caution. Although short‐term seizure freedom is a valuable clinical indicator, thermocoagulation is associated with a well‐documented risk of delayed seizure recurrence beyond 12 months, which could lead to misclassification of long‐term outcomes. ⁵⁷ , ⁵⁸ Consequently, the diagnostic performance metrics derived from this subgroup should be viewed as reflecting the agreement between OPM‐MEG localization and short‐term procedural success rather than definitive validation against a gold standard, resected EZ. This design element, although reflective of real‐world clinical pathways for complex cases, represents a limitation that tempers the generalizability of our prognostic findings and underscores the need for validation in cohorts with uniform surgical resection.

4.6. Limitations

This study has several limitations. First, the analysis focused on OPM‐MEG/iEEG concordance without incorporating a direct comparative analysis against other modalities like MRI or PET, which may affect generalizability. In the future, EEG/OPM‐MEG synchronous monitoring will be incorporated into our study, and comprehensive multimodal analysis will be conducted. Second, the sample size in our study remains inadequate. For some subgroup analyses, such as the differences between adults and children, there are insufficient data for support. Finally, inherent selection bias from single‐center recruitment necessitates multicenter validation to enhance external validity.

AUTHOR CONTRIBUTIONS

Yuanzhong Shen and Chao You are co‐first authors. Kai Shu, Ming Ding, and Feng Hu are co‐corresponding authors. Yuanzhong Shen, Chao You, Yang Zhang, Nan Ji, Xu Zhao, Ming Ding, Feng Hu, and Kai Shu contributed to the conception and design of the study. Yuanzhong Shen, Chao You, Xu Zhao, Ping Zhang, Shanshan Huang, Huicong Kang, Xiaoyan Liu, Yuming Peng, Chang Sun, Bing Yan, Yixiang Zhang, Suiqiang Zhu, Wenzhen Zhu, Zhouping Tang, Ming Ding, Feng Hu, and Kai Shu contributed to the acquisition and analysis of data. Yuanzhong Shen, Chao You, Yuming Peng, Chang Sun, Ting Lei, Zhouping Tang, Ming Ding, Feng Hu, and Kai Shu contributed to drafting a significant portion of the manuscript and figures (for detailed information, please refer to the supplementary materials).

FUNDING INFORMATION

This work was supported by the Jian Dao (JD) Major Program of Hubei Province (SCZ2024008) and the National Key Research and Development Program of China (2022YFC2403905 and 2023YFC2510001).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. 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

Data S1. Supporting information.

Data S2. Supporting information.

Data S3. Standards for Reporting Diagnostic Accuracy 2015 checklist.

Figure S1. Automated Anatomical Labeling template for brain partitioning. The patient underwent stereoelectroencephalographic (SEEG) monitoring. The final results showed that the optically pumped magnetometer magnetoencephalography localization (red dots) was consistent with the SEEG localization (blue areas), both of which were located in the right temporal lobe.

Figure S2. Example of concordance judgment. (A) Electrocorticography (ECoG)–resection concordance. White/blue dipoles: optically pumped magnetometer magnetoencephalography (OPM‐MEG)‐derived epileptogenic zone (EZ) localization overlapping with resection cavity (red). Classification reflects agreement between presurgical OPM‐MEG and intraoperative ECoG‐guided resection margins. (B) ECoG–resection discordance. OPM‐MEG dipoles (white/blue) anatomically divergent from resection cavity (red) are shown. Classification indicates mismatch between noninvasive localization and surgically resected EZ boundaries defined by ECoG. (C) SEEG‐Ablation Concordance. OPM‐MEG dipoles (white/blue) colocalized with SEEG‐defined EZ (yellow). Classification demonstrates agreement between OPM‐MEG localization and SEEG‐guided stereotactic ablation targets. (D) Stereoelectroencephalography (SEEG)–ablation discordance. Spatial discrepancy between OPM‐MEG dipoles (white/blue) and SEEG‐defined EZ (yellow) is shown. Classification highlights discordance between modalities.

Figure S3. Surgical procedures for patients.

DATA AVAILABILITY STATEMENT

All data supporting the findings of this study are available from the corresponding authors upon reasonable request. The code used to generate the results that are reported in this study is available from the corresponding authors upon reasonable request.