- Split View
-
Views
-
Cite
Cite
Stefan Rampp, Hermann Stefan, Xintong Wu, Martin Kaltenhäuser, Burkhard Maess, Friedhelm C Schmitt, Carsten H Wolters, Hajo Hamer, Burkhard S Kasper, Stefan Schwab, Arndt Doerfler, Ingmar Blümcke, Karl Rössler, Michael Buchfelder, Magnetoencephalography for epileptic focus localization in a series of 1000 cases, Brain, Volume 142, Issue 10, October 2019, Pages 3059–3071, https://doi.org/10.1093/brain/awz231
- Share Icon Share
Abstract
The aim of epilepsy surgery in patients with focal, pharmacoresistant epilepsies is to remove the complete epileptogenic zone to achieve long-term seizure freedom. In addition to a spectrum of diagnostic methods, magnetoencephalography focus localization is used for planning of epilepsy surgery. We present results from a retrospective observational cohort study of 1000 patients, evaluated using magnetoencephalography at the University Hospital Erlangen over the time span of 28 years. One thousand consecutive cases were included in the study, evaluated at the University Hospital Erlangen between 1990 and 2018. All patients underwent magnetoencephalography as part of clinical workup for epilepsy surgery. Of these, 405 underwent epilepsy surgery after magnetoencephalography, with postsurgical follow-ups of up to 20 years. Sensitivity for interictal epileptic activity was evaluated, in addition to concordance of localization with the consensus of presurgical workup on a lobar level. We evaluate magnetoencephalography characteristics of patients who underwent epilepsy surgery versus patients who did not proceed to surgery. In operated patients, resection of magnetoencephalography localizations were related to postsurgical seizure outcomes, including long-term results after several years. In comparison, association of lesionectomy with seizure outcomes was analysed. Measures of diagnostic accuracy were calculated for magnetoencephalography resection and lesionectomy. Sensitivity for interictal epileptic activity was 72% with significant differences between temporal and extra-temporal lobe epilepsy. Magnetoencephalography was concordant with the presurgical consensus in 51% and showed additional or more focal involvement in an additional 32%. Patients who proceeded to surgery showed a significantly higher percentage of monofocal magnetoencephalography results. Complete magnetoencephalography resection was associated with significantly higher chances to achieve seizure freedom in the short and long-term. Diagnostic accuracy was significant in temporal and extra-temporal lobe cases, but was significantly higher in extra-temporal lobe epilepsy (diagnostic odds ratios of 4.4 and 41.6). Odds ratios were also higher in non-lesional versus lesional cases (42.0 versus 6.2). The results show that magnetoencephalography provides non-redundant information, which significantly contributes to patient selection, focus localization and ultimately long-term seizure freedom after epilepsy surgery. Specifically in extra-temporal lobe epilepsy and non-lesional cases, magnetoencephalography provides excellent accuracy.
See Burgess (doi:10.1093/brain/awz281) for a scientific commentary on this article.
Introduction
In about 30% of patients suffering from focal epilepsies, pharmacotherapy with anti-epileptic drugs is insufficiently effective (Kwan et al., 2010). Persisting seizures, anti-epileptic drug side effects, as well as psychiatric comorbidities considerably impact quality of life (Taylor et al., 2011). Furthermore, the associated direct and indirect costs generate a significant burden for society (Strzelczyk et al., 2017). A safe and cost-effective alternative therapy option in appropriately selected patients is epilepsy surgery (Picot et al., 2016; West et al., 2016). Depending on the specific aetiology, seizure-freedom rates of ∼70% 1 year after surgery can be achieved in contrast to around 6% with further anti-epileptic drug therapy (Brodie et al., 2012; Blumcke et al., 2017).
However, there are considerable unresolved issues. Although current consensus advises prompt referral (Wiebe and Jette, 2012), delays between first diagnosis of epilepsy and evaluation for surgery remain in the range of one to two decades (Benbadis et al., 2003; Martínez-Juárez et al., 2017). Furthermore, success of epilepsy surgery is limited in specific populations, such as patients with unremarkable MRI findings (Blumcke et al., 2017). Probably the most significant issue is the recurrence of seizures within 2–5 years after surgery in more than 40% of patients (de Tisi et al., 2011).
As an advanced diagnostic modality, magnetoencephalography (MEG) may contribute to resolve some of these issues. The clinical utilization of MEG especially in the field of epileptology and epilepsy surgery has significantly evolved since the first MEG recordings almost 50 years ago (Cohen, 1968, 1970). While technical and methodological developments dominated initially, the technique was introduced early to clinical neurophysiology. Evaluation of the clinical value followed in the late 1980s and early 1990s (Sutherling et al., 1987, 1991; Stefan et al., 1990). The high temporal and good spatial resolution, as well as the relative insensitivity of MEG localization to conductivity differences e.g. of the skull or brain tissues (Scheler et al., 2007; Güllmar et al., 2010; Vorwerk et al., 2014) provided incentives especially for utilization for epileptic focus localization as well as functional mapping (Coolen et al., 2018).
An extensive and growing number of studies have evaluated MEG-based source analysis (or ‘magnetic source imaging’, MSI) in patients with focal epilepsies. MSI allows early identification of surgery candidates (Ossenblok et al., 2007). It improves planning and results of invasive recordings (Sutherling et al., 2008; Murakami et al., 2016). MEG yields non-redundant information in up to about 30% of cases and is confirmatory in an additional 50% (Stefan et al., 2003). It also yields valuable information in complex cases (Nissen et al., 2016). Application is viable also in young and very young children (Garcia-Tarodo et al., 2018). Seizure-freedom rates after epilepsy surgery are higher if MEG findings are taken into account (Vadera et al., 2013; Mu et al., 2014; Englot et al., 2015; Kasper et al., 2018). Furthermore, re-evaluation of patients with recurrent seizures after surgery, up to 40% (de Tisi et al., 2011), is facilitated and enables successful second surgery (Muthaffar et al., 2017).
The emergence of clinical MEG-focused associations, as well as the endorsement by existing societies, now also provide a professional framework and has led to the definition of guidelines for clinical MEG applications (Hashimoto et al., 2005; Bagic et al., 2009; De Tiège et al., 2017; Hari et al., 2018). The recommended methods are based on many studies and decades of experience, yielding reliable and valid results.
In the present study, we investigate the role of such clinical application of MEG in presurgical workup for epilepsy surgery. In the largest series to date of 1000 patients and more than 400 surgical procedures over the time span of 28 years, we evaluate the hypothesis that MEG supports patient selection, contributes to identification of the epileptogenic zone in presurgical evaluation and impacts long-term seizure outcomes after epilepsy surgery.
Materials and methods
Patients
This retrospective observational cohort study evaluates consecutive patients with focal epilepsies who underwent MEG recordings for focus localization from the installation of the first clinical MEG system at the University Hospital Erlangen in 1990 to the time of writing in 2018.
Decision for an MEG recording was based on individual clinical considerations. While there was no canonical set of indications, patients with clear and concordant structural, semiological and electrophysiological findings tended not to be referred to MEG. Exclusion criterion was MEG incompatibility (metal implants, pacemaker, etc.).
All recordings, source imaging and interpretation were performed prospectively before surgery. Findings were available to the medical and surgical team and were presented and discussed in patient management conferences. Localizations were also provided for neuronavigation for epilepsy surgery and implantation of invasive electrodes if requested.
All patients, or their parents in the case of children or juveniles, gave their written informed consent according to the Declaration of Helsinik to participate in the MEG/EEG recordings as part of the clinical workup and to use their anonymized data for scientific purposes and publication.
MEG recordings
Three different MEG systems were used: from 1990 to 1994, a 37-channel gradiometer system (Krenikon, Siemens); from 1995 to 2010 a 2 × 37 channel axial gradiometer Magnes II system (Biomagnetic Technologies, Inc.) and from 2010 a 248-channel whole-head magnetometer 3600WH system (Biomagnetic Technologies, Inc.). During the time period from 1995 to 2010, individual patients with complex epilepsies were measured at the Max Planck Institute for Human Cognitive and Brain Sciences in Leipzig, Germany, using a 148-channel axial gradiometer 2500WH system (Biomagnetic Technologies, Inc.). During the transition time from the Magnes II to the 3600WH system, individual patients were also recorded at the Department of Neurology, University Hospital Magdeburg, Germany using a 3600WH system.
Patients were recorded in supine and/or seated position, with their eyes closed for 40–60 min. The Krenikon system only allowed measurement from a limited region of interest with a diameter of ∼20 cm. The Magnes II system consisted of two sensors in separate dewars, which allowed recording from two regions of interest simultaneously. For complete coverage of the brain, sensor positions were varied in subsequent runs. Clinical information was used to optimize positioning and recording time per position.
Recordings usually included simultaneous EEG. However, the number of channels was initially limited to 20–33 electrodes. With the 3600WH system, EEG was recorded with 64 electrodes. EEG data were used to identify artefacts and normal variants, as well as epileptic activity for separate source analysis and detection of MEG correlates. Because of the scope of the present study, as well as the limitations of only low resolution EEG (Brodbeck et al., 2011) in a substantial percentage of patients, we did not perform a comparison with the results of ESI (electric source imaging). However, we have reported previously on such comparisons in patient subgroups (Paulini et al., 2007; Scheler et al., 2007; Heers et al., 2010a, b).
Interictal epileptic discharges
Findings used for the presented study were interictal epileptic discharges (IEDs), i.e. spikes and sharp waves. IED detection was performed visually by experienced interpreters routinely involved in presurgical evaluation. In general, at least five IEDs with similar morphology and topography were necessary for further analysis and reporting. Only in cases of highly specific signals and robust localizations were fewer IEDs considered sufficient.
Source localization
All IEDs were localized using dipole analysis. Software included the manufacturer’s software, as well as Curry (versions 4.6 to 7, Compumedics Neuroscan, Hamburg, Germany) and BESA (versions 5.3 to 6.1, BESA GmbH, Gräfeling, Germany). Preprocessing included band-pass filtering of 1–70 Hz and notch filtering of 50 Hz.
Two main localization approaches were used: single, and averaged IED analysis. Single IED analysis fitted single dipoles to the rising flank of the IED pattern. Averaged spike analysis fitted regional dipoles at the onset and at the peak of the pattern. Both approaches stressed robust results in early time segments.
Dipoles were selected according to statistical criteria, such as correlation of measured and estimated signal, confidence volume and source amplitude. Plausibility was evaluated subjectively (e.g. no erratic propagation, exclusion of deep dipoles in the centre of the head, etc.). The specific statistics and minimum required criteria changed over time and also depended on the specific software and MEG system used. In general, they were, and currently are, in line with published clinical practice guidelines (Bagić et al., 2011). In case of multifocal activity during single IEDs, multiple dipoles were fitted either sequentially or simultaneously. Stable localizations from early time points were then reported. Furthermore, additional distributed inverse solutions were calculated (see below).
Localizations used single or multiple spheres, a single shell adapted to the shape of the inner skull surface or individual BEM (boundary element model) for volume conductor models. In a study comparing some of these approaches, differences were found to be minimal for MEG (Scheler et al., 2007). The choice of volume conductor model was therefore secondary to other considerations, i.e. technical details and availability of the respective software and did not depend on patients’ characteristics. Since the Scheler et al. (2007) study, routine evaluation applied a spherical model for MEG analysis using Curry or BESA software.
Starting in 2001, additional inverse solutions were calculated using Curry and BESA software, including LORETA/sLORETA, minimum norms and CLARA. This was added to dipole analysis to ensure that the results were not significantly affected by erroneous assumptions, e.g. the use of single dipoles, etc. Discordant results were checked for problems in IED selection, averaging or preprocessing. If the issues could not be resolved, we reported the inconsistencies and the potentially limited validity of the findings.
All results were superimposed on MRIs of the individual patient, presented as part of patient management conferences and/or provided as written reports, including clinical discussion and interpretation.
Localizations
For statistical evaluation in the presented study, localizations were classified according to side and region (frontal, parietal, temporal, occipital and insular). Localizations reported as central were distinguished into frontal and parietal.
Localizations were classified regarding their spatial distribution. A group of localizations positioned closely to each other was termed a ‘cluster’ in contrast to diffuse distributions with no clear local maxima. The classification was performed during clinical routine and relied on visual interpretation, as widely accepted standards for the definition are not available (Bagić et al., 2011). Focal localizations show one single cluster of activity. Multifocal findings consist of several distinct groups of localizations, based on different pattern morphology and field topography. Finally, diffuse localizations are scattered over one or several regions without clear clustering.
Reference standard
MEG results were related to findings of presurgical evaluation and results of epilepsy surgery. Diagnostic accuracy of MEG, here referring to the ability to define the area for successful surgery, was compared to lesionectomy in patients with lesional epilepsies. In case of missing data, the respective cases were excluded from the specific statistical analysis. Operated cases were only excluded from Kaplan-Meier survival statistics if no follow-up data were available, e.g. because of surgery at another institution.
Presurgical workup
MEG results were compared to presurgical evaluation, including MRI, long-term video-EEG, as well as single-photon emission computed tomography (SPECT) in most cases, as well as ictal SPECT, PET and invasive EEG recordings. Specifics of the different modality depended on availability and standards at the time of investigation. For example, MRI field strength was initially 1.5 T, whereas patients since 2008 had 3 T MRI. Results were classified into regions as described above for MEG localizations. Concordance of MEG results and presurgical consensus were evaluated on a lobar level using χ2 tests. The degree of concordance was classified into the categories concordant, consistent and discordant findings. Concordant MEG findings exactly reproduce the presurgical synopsis. Consistent findings have an overlap of at least one lobe, but also show differences. Concordant findings were not included in the group of consistent findings. Finally, discordant findings do not show an overlap.
Epilepsy surgery
Association of seizure outcome after surgery with MEG/focus hypothesis concordance, as well as MEG resection, was evaluated using χ2 tests. Outcomes were graded according to the Engel classification (Engel et al., 1993). Outcomes were compared using non-parametric Mann-Whitney U-tests.
A detailed analysis of the relation of MEG findings to surgery and outcome was performed by visually evaluating whether the respective MEG localizations had been resected completely, partially or not at all, similar to previous studies (Vadera et al., 2013; Mu et al., 2014). For clusters of single spike dipole localizations, a tolerance of 1 cm within the border was included to define a complete resection. Spurious outlier localizations were not taken into consideration.
In case of averaged spike analysis, which yields only a single localization, complete resection was assumed if source localization was contained in the resection or within 1 cm of the border. This tolerance was defined before the evaluation and was aimed to account for coregistration and localization inaccuracies, as well as effects of brain shift when comparing presurgical MEG findings with MRIs acquired several months after surgery. The strategy follows previous studies (Lau et al., 2008; Kasper et al., 2018), which, however, use less restrictive tolerances of ∼2 cm.
Partial resection was assumed when the averaged source localization or the dipole cluster (in whole or in part) was located outside the resection but within 2–3 cm of the border. This analysis was only possible when adequate postoperative imaging or a detailed description by the surgeon was available. It addresses the validity of MEG findings on a finer level than the lobar comparisons to findings of presurgical workup.
Parameters of diagnostic accuracy [sensitivity, specificity, positive and negative predictive value (PPV, NPV), diagnostic odds ratio (DOR), positive and negative likelihood ratio (PLR, NLR)] were calculated for the association of complete MEG resection with the reference standard of: (i) postoperative Engel 1; and (ii) postoperative Engel 1A outcomes at last follow-up after at least 1 year. Additionally, a second analysis focused on long-term outcomes after at least 5 years after surgery.
Kaplan-Meier survival analysis was calculated to evaluate the influence of MEG resection on persistence of an Engel 1 outcome after surgery. Recurring seizures classified as Engel ≥ 2 were considered as events in terms of Kaplan-Meier analysis. Differences between degrees of MEG resection were evaluated using a Tarone-Ware test.
Lesionectomy
Resection size could potentially confound the influence of MEG resection on postsurgical seizure freedom, i.e. any benefit of complete MEG resection may be caused only by increasing the resection size and only secondarily result in a higher percentage of seizure-free cases. Seizure freedom in this group would then mainly rely on complete or extended lesionectomy. Therefore, we compared resection extent in lesional cases with seizure outcome. Reports and/or postoperative MRI were evaluated to classify lesionectomies into categories: no, partial, complete or extended resection of the suspected epileptogenic lesion. Multiple subpial transections and callosotomy in lesional cases were categorized as operative procedures without lesionectomy (category ‘no resection’). Extension of a previous resection was counted as extended lesionectomy. In cases with mesial temporal lobe epilepsies, selective amygdala-hippocampectomy and tailored resection were classified as complete resection constrained to lesion/structural alteration. Standard anterior 2/3-lobectomies were categorized as extended, as the resection exceeds the structural alteration. If multiple subpial transection was performed in addition to resection, the extent of the resection was used.
Extent was compared to the last available seizure outcome using χ2 tests, with follow-ups after at least 1 and 5 years. The analysis was performed (i) in all lesional surgeries; and (ii) limited to cases with IEDs on MEG findings, to enable a comparison with MEG resection in the same patient group. Additionally, parameters of diagnostic accuracy were calculated.
Data availability
Anonymized evaluation data are available upon request.
Results
Patients
Of the 1000 patients, 475 were female. Mean age on the day of MEG recording was 32.4 years (±12.5, range 3–70 years). A total of 114 patients were below the age of 18. Mean age at epilepsy onset was 13.9 years (±11.7 years, range 0–60 years). Mean duration of epilepsy at the time of first MEG recordings was 18.0 years (±12.0 years, range 0–66 years).
Putative location of the epileptic focus was determined from the consensus of available clinical information, which is referred to in the following as ‘focus hypothesis’ (overview in Supplementary Table 1).
MEG was performed after previous surgery in 117 patients, corresponding to 165 recordings. A total of 405 patients underwent epilepsy surgery after MEG; however, outcome data were not available for 19 procedures.
MEG recordings
A total of 1274 MEG recordings in 1000 patients were evaluated. Of these, 184 patients had repeated MEG because of different reasons, e.g. no IEDs in the first recordings or due to technical factors. This also included measurements after epilepsy surgery in case of persisting seizures. Before 2000, 283 patients were investigated with 371 recordings (24% repeated recordings). Between 2000 and 2009, 470 patients were evaluated with 616 recordings (24% repeated recordings) and after 2009, 247 patients with 287 recordings (14%). The difference in the number of repeated measurements was statistically significant (P = 0.002, χ2 = 12.6, χ2 test, P < 0.001, χ2 = 12.6, comparing before and after 2009).
IED detection
A total of 1231 recordings could be evaluated for IEDs; the remaining 43 (3%) were unusable due to strong artefacts, 35 of these were first recordings, while eight were repeated.
MEG detected IEDs in 883 of 1231 recordings (72%). Supplementary Table 2 provides more details. Recordings before 2000 yielded IEDs in 74.4%, between 2000 and 2009 in 64% and after 2009 in 73.2% (P = 0.001, χ2 = 13.4, χ2 test).
The difference between IED detection rates in temporal lobe epilepsy (TLE) versus extra-temporal lobe epilepsy (ETLE) was statistically significant (P < 0.001, χ2 = 17.9, χ2test).
Earlier onsets of epilepsy were related to an increased likeliness to detect IEDs in the first recordings, also when duration of epilepsy, as well as TLE versus ETLE, was taken into account (logistic regression, P < 0.001 for onset). Patients with ETLE showed significantly earlier onsets (11.0 ± 11.0 versus 14.6 ± 12.4 years, P = 0.002, t = 5.7, two-sided t-test). In contrast, duration of epilepsy did not show a significant influence or differences between TLE and ETLE groups.
MEG localizations
MEG localizations included the frontal lobe in 379 recordings (Fig. 1 shows an example), the temporal lobe in 480, the parietal lobe in 189, the occipital lobe in 50 and the insula in eight recordings. Information about focality of spike localizations was available in 876 recordings. Localizations were monofocal in 619 recordings with spikes (70.7%), multifocal in 185 (21%) and diffuse in 74 (8.4%).
Comparison to presurgical workup
In 115 of 802 recordings (14%), MEG findings involved fewer lobes than the clinical focus hypothesis. In a total of 635 recordings (79%), MEG and focus hypothesis involved an equal number of lobes. In the remaining 167 recordings, MEG yielded more extended results (20%). Consistent findings were slightly more frequent when a lesion was present (78% versus 71%, P = 0.024, χ2 = 5.1, df = 3, χ2 test).
Recordings after 2009, corresponding to the use of a whole-head MEG in our lab, yielded more focal (P < 0.001, χ2 = 14.5, df = 3, χ2 test) and confirmatory findings (57% versus 49%, P = 0.058, χ2 = 3.6, df = 3) compared to earlier recordings.
MEG findings were concordant with the consensus of presurgical evaluation in 405 of 802 cases with IEDs and known focus hypothesis (50.5%). They were consistent in an additional 254 patients (32%) and discordant in 143 (18%).
Concordant findings were less frequent in ETLE versus TLE (44% versus 59%, P < 0.001, χ2 test); whereas consistent findings were more frequent (36% versus 26%, P < 0.001, χ2 test). A total of 254 patients with spikes presented with multilobar findings in presurgical evaluation. MEG provided monofocal localizations in 93 (37%).
MEG and epilepsy surgery
A total of 405 surgeries after MEG were performed. Outcome data were available in 386 and MEG data were usable in 371 cases (92%). Last available outcome was Engel 1 in 215 patients (Engel 1A in 152), Engel 2 in 69, Engel 3 in 59 and Engel 4 in 43 patients. Median follow-up after surgery was 4.0 years (first quartile 1.0 years, third quartile 6.0 years). In 325 patients, outcome after at least 1 year was available. Long-term outcomes after at least 2 years were available in 278, after at least 5 years in 188 and after 10 years or more (20 years maximum) in 61. IEDs were detected in 256 of 371 recordings (69%). Localizations were monofocal in 203 recordings (79%), multifocal in 39 (15%) and diffuse in 14 (6%). Two patients underwent callosotomy and were excluded from comparisons with the area of surgery.
Patients undergoing epilepsy surgery had a significantly higher percentage of monofocal MEG localizations than patients who did not proceed to surgery (79% versus 67%, P < 0.001, χ2, 876 patients with spikes). This was reflected by more focal presurgical evaluation results (P < 0.001, χ2 = 12.7, df = 3, χ2, 895 patients with focus hypothesis).
In the subgroup of operated patients with IEDs in MEG, MEG localization and presurgical focus hypothesis were concordant or consistent in 219 of 254 cases (86.2%). Such findings were favourable regarding seizure outcome when Engel 1 or 2 outcomes were evaluated (χ2, P = 0.014, χ2 = 6.0, last available outcomes, P = 0.11, χ2 = 2.6, after at least 1 year, P = 0.019, χ2 = 5.5, after at least 2 years, df = 3, not significant thereafter). Comparison to Engel 1 outcomes did not reveal a significant association. MEG findings and focus hypothesis were concordant in 160 (63%), without clear association with postoperative seizure outcomes.
Resection extent of MEG localizations could be evaluated in 174 operated patients with IEDs, limited by the availability of adequate MRIs and documentation. Follow-up data after at least 1 year were available in 149 patients, after at least 5 years in 78.
MEG resection extent was significantly related to seizure freedom at the last available follow-up after at least 1 year (Engel 1 P < 0.001, χ2 = 35.1, Engel 1A P < 0.004, χ2 = 10.9, df = 3, χ2, Supplementary Table 3) and after at least 5 years (Engel 1 P < 0.001, χ2 = 22.7, Engel 1A P < 0.005, χ2 = 10.6, df = 3, χ2, Supplementary Table 3). Comparison of recordings with whole-head versus non-whole head systems did not reveal a significant difference (Kaplan-Meier survival analysis, all resection extents P = 0.91, only complete resections P = 0.32),
Sensitivity of complete MEG resection for an Engel 1 outcome (≥1 year) was 66% (57–73%); specificity was 83% (76–89%). PPV amounted to 83% (76–89%) and NPV to 65% (57–73%) (Table 2 and Supplementary Table 4). Long-term outcomes showed similar results (Table 2). The ability to predict the postsurgical outcome was considerably better for ETLE versus TLE cases. In ETLE cases, complete MEG resection reached a PLR of 7.55 (6.25–9.12) (Table 3). With an odds ratio of 42.0 (4.3–408.7) non-lesional cases benefitted more from MEG compared to patients with epileptogenic lesions with an odds ratio of 6.2 (2.6–14.7).
Complete MEG resection . | . | |
---|---|---|
Follow-up . | ≥1 year . | ≥5 years . |
n | 149 | 78 |
Engel 1 | 84 | 47 |
Sensitivity | 65% (57–73%) | 75% (63–84%) |
Specificity | 83% (76–89%) | 74% (63–83%) |
PPV | 83% (76–89%) | 81% (71–89%) |
NPV | 65% (57–73%) | 66% (54–76%) |
DOR | 9.3 (4.2–20.5) | 8.4 (3.0–23.7) |
PLR | 3.9 (3.2–4.7) | 2.9 (2.2–3.8) |
NLR | 0.4 (0.3–0.5) | 0.3 (0.3–0.5) |
Engel 1A | 55 | 27 |
Sensitivity | 62% (54–70%) | 74% (63–83%) |
Specificity | 66% (58–73%) | 55% (43–66%) |
PPV | 52% (43–60%) | 47% (35–58%) |
NPV | 75% (67–81%) | 80% (69–88%) |
DOR | 3.1 (1.6–6.3) | 3.5 (1.3–9.7) |
PLR | 1.8 (1.4–2.4) | 1.6 (1.2–2.4) |
NLR | 0.6 (0.4–0.8) | 0.5 (0.3–0.7) |
Complete MEG resection . | . | |
---|---|---|
Follow-up . | ≥1 year . | ≥5 years . |
n | 149 | 78 |
Engel 1 | 84 | 47 |
Sensitivity | 65% (57–73%) | 75% (63–84%) |
Specificity | 83% (76–89%) | 74% (63–83%) |
PPV | 83% (76–89%) | 81% (71–89%) |
NPV | 65% (57–73%) | 66% (54–76%) |
DOR | 9.3 (4.2–20.5) | 8.4 (3.0–23.7) |
PLR | 3.9 (3.2–4.7) | 2.9 (2.2–3.8) |
NLR | 0.4 (0.3–0.5) | 0.3 (0.3–0.5) |
Engel 1A | 55 | 27 |
Sensitivity | 62% (54–70%) | 74% (63–83%) |
Specificity | 66% (58–73%) | 55% (43–66%) |
PPV | 52% (43–60%) | 47% (35–58%) |
NPV | 75% (67–81%) | 80% (69–88%) |
DOR | 3.1 (1.6–6.3) | 3.5 (1.3–9.7) |
PLR | 1.8 (1.4–2.4) | 1.6 (1.2–2.4) |
NLR | 0.6 (0.4–0.8) | 0.5 (0.3–0.7) |
DOR = diagnostic odds ratio; PLR/NLR = positive/negative likelihood ratio.
Complete MEG resection . | . | |
---|---|---|
Follow-up . | ≥1 year . | ≥5 years . |
n | 149 | 78 |
Engel 1 | 84 | 47 |
Sensitivity | 65% (57–73%) | 75% (63–84%) |
Specificity | 83% (76–89%) | 74% (63–83%) |
PPV | 83% (76–89%) | 81% (71–89%) |
NPV | 65% (57–73%) | 66% (54–76%) |
DOR | 9.3 (4.2–20.5) | 8.4 (3.0–23.7) |
PLR | 3.9 (3.2–4.7) | 2.9 (2.2–3.8) |
NLR | 0.4 (0.3–0.5) | 0.3 (0.3–0.5) |
Engel 1A | 55 | 27 |
Sensitivity | 62% (54–70%) | 74% (63–83%) |
Specificity | 66% (58–73%) | 55% (43–66%) |
PPV | 52% (43–60%) | 47% (35–58%) |
NPV | 75% (67–81%) | 80% (69–88%) |
DOR | 3.1 (1.6–6.3) | 3.5 (1.3–9.7) |
PLR | 1.8 (1.4–2.4) | 1.6 (1.2–2.4) |
NLR | 0.6 (0.4–0.8) | 0.5 (0.3–0.7) |
Complete MEG resection . | . | |
---|---|---|
Follow-up . | ≥1 year . | ≥5 years . |
n | 149 | 78 |
Engel 1 | 84 | 47 |
Sensitivity | 65% (57–73%) | 75% (63–84%) |
Specificity | 83% (76–89%) | 74% (63–83%) |
PPV | 83% (76–89%) | 81% (71–89%) |
NPV | 65% (57–73%) | 66% (54–76%) |
DOR | 9.3 (4.2–20.5) | 8.4 (3.0–23.7) |
PLR | 3.9 (3.2–4.7) | 2.9 (2.2–3.8) |
NLR | 0.4 (0.3–0.5) | 0.3 (0.3–0.5) |
Engel 1A | 55 | 27 |
Sensitivity | 62% (54–70%) | 74% (63–83%) |
Specificity | 66% (58–73%) | 55% (43–66%) |
PPV | 52% (43–60%) | 47% (35–58%) |
NPV | 75% (67–81%) | 80% (69–88%) |
DOR | 3.1 (1.6–6.3) | 3.5 (1.3–9.7) |
PLR | 1.8 (1.4–2.4) | 1.6 (1.2–2.4) |
NLR | 0.6 (0.4–0.8) | 0.5 (0.3–0.7) |
DOR = diagnostic odds ratio; PLR/NLR = positive/negative likelihood ratio.
Complete MEG resection, follow-up ≥1 year . | |||
---|---|---|---|
. | ETLE . | TLE . | mTLE . |
n | 67 | 86 | 49 |
Engel 1A | 31 | 55 | 33 |
Sensitivity | 84% (72–92%) | 56% (45–67%) | 64% (49–77%) |
Specificity | 89% (78–95%) | 77% (67–86%) | 69% (54–81%) |
PPV | 87% (76–94%) | 82% (71–89%) | 81% (67–91%) |
NPV | 87% (75–94%) | 50% (39–61%) | 48% (34–62%) |
DOR | 41.60 (10.1–170.9) | 4.4 (1.6 – 12.0) | 3.9 (1.1–13.8) |
PLR | 7.6 (6.3–9.1) | 2.5 (1.8–3.4) | 2.0 (1.3–3.2) |
NLR | 0.2 (0.2–0.2) | 0.6 (0.4–0.8) | 0.5 (0.3–0.8) |
Complete MEG resection, follow-up ≥1 year . | |||
---|---|---|---|
. | ETLE . | TLE . | mTLE . |
n | 67 | 86 | 49 |
Engel 1A | 31 | 55 | 33 |
Sensitivity | 84% (72–92%) | 56% (45–67%) | 64% (49–77%) |
Specificity | 89% (78–95%) | 77% (67–86%) | 69% (54–81%) |
PPV | 87% (76–94%) | 82% (71–89%) | 81% (67–91%) |
NPV | 87% (75–94%) | 50% (39–61%) | 48% (34–62%) |
DOR | 41.60 (10.1–170.9) | 4.4 (1.6 – 12.0) | 3.9 (1.1–13.8) |
PLR | 7.6 (6.3–9.1) | 2.5 (1.8–3.4) | 2.0 (1.3–3.2) |
NLR | 0.2 (0.2–0.2) | 0.6 (0.4–0.8) | 0.5 (0.3–0.8) |
DOR = diagnostic odds ratio; mTLE = mesial TLE; PLR/NLR = positive/negative likelihood ratio.
Complete MEG resection, follow-up ≥1 year . | |||
---|---|---|---|
. | ETLE . | TLE . | mTLE . |
n | 67 | 86 | 49 |
Engel 1A | 31 | 55 | 33 |
Sensitivity | 84% (72–92%) | 56% (45–67%) | 64% (49–77%) |
Specificity | 89% (78–95%) | 77% (67–86%) | 69% (54–81%) |
PPV | 87% (76–94%) | 82% (71–89%) | 81% (67–91%) |
NPV | 87% (75–94%) | 50% (39–61%) | 48% (34–62%) |
DOR | 41.60 (10.1–170.9) | 4.4 (1.6 – 12.0) | 3.9 (1.1–13.8) |
PLR | 7.6 (6.3–9.1) | 2.5 (1.8–3.4) | 2.0 (1.3–3.2) |
NLR | 0.2 (0.2–0.2) | 0.6 (0.4–0.8) | 0.5 (0.3–0.8) |
Complete MEG resection, follow-up ≥1 year . | |||
---|---|---|---|
. | ETLE . | TLE . | mTLE . |
n | 67 | 86 | 49 |
Engel 1A | 31 | 55 | 33 |
Sensitivity | 84% (72–92%) | 56% (45–67%) | 64% (49–77%) |
Specificity | 89% (78–95%) | 77% (67–86%) | 69% (54–81%) |
PPV | 87% (76–94%) | 82% (71–89%) | 81% (67–91%) |
NPV | 87% (75–94%) | 50% (39–61%) | 48% (34–62%) |
DOR | 41.60 (10.1–170.9) | 4.4 (1.6 – 12.0) | 3.9 (1.1–13.8) |
PLR | 7.6 (6.3–9.1) | 2.5 (1.8–3.4) | 2.0 (1.3–3.2) |
NLR | 0.2 (0.2–0.2) | 0.6 (0.4–0.8) | 0.5 (0.3–0.8) |
DOR = diagnostic odds ratio; mTLE = mesial TLE; PLR/NLR = positive/negative likelihood ratio.
Stability of Engel 1 outcome over time was evaluated using Kaplan-Meier survival analysis (Fig. 2). The results show that rates of Engel 1 outcomes are significantly higher and more stable over the course of up to 10 years after surgery if MEG localizations are completely resected in comparison to both partial and no resection (P < 0.001, Tarone-Ware test). Similar results are observed if Engel 1–2 or Engel 1–3 are evaluated (P < 0.001 and P = 0.001).
Comparison with lesionectomy
Information about the extent of lesionectomy was available for 284 surgical procedures for lesional epilepsy, 134 of which with IEDs on MEG and information about MEG resection.
All lesionectomies
Extended lesionectomies were performed in 130 of 284 cases (46%). Complete lesionectomies were constrained to the lesion in 124 (44%). Only partial removal of the lesion was performed in 13 (5%). The lesion was not resected in 17 cases (6%). Median follow-up in this group was 4 years (first quartile 1.0 year, third quartile 5.0 years).
Complete or extended lesionectomy was significantly related to Engel 1 outcome (P < 0.001, χ2 = 15.0, after at least 1 year, χ2 test, df = 3). Outcome differences between complete and extended resections were not significant (P = 0.10, Mann-Whitney U-test).
In the subgroup of patients with temporal lobe lesions, degree of lesionectomy was not related to Engel 1 outcome (P = 0.11). Complete versus extended resections also yielded no significant differences (P = 0.77). In patients with temporal mesial epilepsy, selective amygdalo-hippocampectomy or tailored resection showed comparable results in comparison to standard resection including the anterior 2/3 of the temporal lobe (P = 0.74).
Sensitivity of complete or extended resection for an Engel 1 outcome (≥1 year) was 96% (95% CI: 93–98%); specificity was 19% (95% CI: 14–25%). PPV amounted to 64% (95% CI: 57–70%) and NPV to 78% (95% CI: 72–83%). Diagnostic performance was similar for long-term outcomes (≥5 years) (Supplementary Table 5).
Lesionectomies in patients with MEG resection data
In the group with operated patients with IEDs on MEG, extended lesionectomies were performed in 75 (56%). Lesions were completely resected without extension in 52 (39%), only partially in six (5%) and not at all in one case (1%). Median follow-up in this group was 3.8 years (first quartile 1.0 year, third quartile 5.0 years).
Here, extent of lesionectomy was not related to an Engel 1 outcome (P = 0.63, χ2 = 0.2, after at least 1 year, df = 3, χ2 test). There was also no significant difference between complete and extended resections regarding long-term outcome (P = 0.22, Kaplan-Meier analysis, Tarone-Ware test). Complete or extended resection of temporal lesions showed a tendency for an Engel 1 outcome in comparison to partial or no resection. Further restrictions to patients with mesial temporal lobe epilepsy yielded no significant differences between selective amygdala-hippocampectomy or tailored resection versus standard resection.
In contrast, complete MEG resection was associated with an Engel 1 outcome (P < 0.001, χ2 = 22.6, last outcome and χ2 = 18.7, after at least 1 year), which was also supported by Kaplan-Maier survival analysis (P = 0.001, Tarone-Ware test). This association remained when the analysis was restricted to patients with temporal lesions only (P = 0.0097).
In the subgroup of patients with temporal mesial involvement, both degree of lesionectomy and MEG resection showed no relation to Engel 1 outcome (P = 0.47 and P = 0.10). Sensitivity of complete or extended lesionectomy for an Engel 1 outcome (≥1 year) was 97% (92–100%); specificity was 5% (2–11%). PPV amounted to 62.04% (52–71%) and NPV to 50% (41–60%) (Table 4 and Supplementary Table 6). Diagnostic parameters were comparable or worse for long-term outcomes (≥5 years) (Table 4).
Lesionectomies in cases with MEG resection data . | ||
---|---|---|
Follow-up . | ≥1 year . | ≥5 years . |
n | 112 | 58 |
Engel 1 | 69 | 37 |
Sensitivity | 97% (91–100%) | 97% (88–100%) |
Specificity | 5% (2–11%) | 0% (0.2–6%) |
PPV | 62% (52–71%) | 63% (49–75%) |
NPV | 50% (40–60%) | 0.00% (0.2–6%) |
DOR | 1.6 (0.2–12.0) | 0.0 (nc) |
PLR | 1.0 (0.4–2.7) | 1.0 (nc) |
NLR | 0.6 (0.2–1.7) | nc |
Engel 1A | 44 | 20 |
Sensitivity | 100% (96–100%) | 100% (92–100%) |
Specificity | 6% (3–12%) | 3% (0.3–11%) |
PPV | 41% (32–50%) | 35% (23–49%) |
NPV | 100% (96–100%) | 100% (92–100%) |
DOR | nc | nc |
PLR | 1.1 (0.9–1.3) | 1.0 (0.7–1.5) |
NLR | 0.0 (0.0–0.0) | 0.0 (0.0–0.0) |
Lesionectomies in cases with MEG resection data . | ||
---|---|---|
Follow-up . | ≥1 year . | ≥5 years . |
n | 112 | 58 |
Engel 1 | 69 | 37 |
Sensitivity | 97% (91–100%) | 97% (88–100%) |
Specificity | 5% (2–11%) | 0% (0.2–6%) |
PPV | 62% (52–71%) | 63% (49–75%) |
NPV | 50% (40–60%) | 0.00% (0.2–6%) |
DOR | 1.6 (0.2–12.0) | 0.0 (nc) |
PLR | 1.0 (0.4–2.7) | 1.0 (nc) |
NLR | 0.6 (0.2–1.7) | nc |
Engel 1A | 44 | 20 |
Sensitivity | 100% (96–100%) | 100% (92–100%) |
Specificity | 6% (3–12%) | 3% (0.3–11%) |
PPV | 41% (32–50%) | 35% (23–49%) |
NPV | 100% (96–100%) | 100% (92–100%) |
DOR | nc | nc |
PLR | 1.1 (0.9–1.3) | 1.0 (0.7–1.5) |
NLR | 0.0 (0.0–0.0) | 0.0 (0.0–0.0) |
Only cases with IEDs on MEG and information on MEG resection were considered, corresponding to Table 2. nc = not computable due to zeros in the denominator, etc. DOR = diagnostic odds ratio; PLR/NLR = positive/negative likelihood ratio.
Lesionectomies in cases with MEG resection data . | ||
---|---|---|
Follow-up . | ≥1 year . | ≥5 years . |
n | 112 | 58 |
Engel 1 | 69 | 37 |
Sensitivity | 97% (91–100%) | 97% (88–100%) |
Specificity | 5% (2–11%) | 0% (0.2–6%) |
PPV | 62% (52–71%) | 63% (49–75%) |
NPV | 50% (40–60%) | 0.00% (0.2–6%) |
DOR | 1.6 (0.2–12.0) | 0.0 (nc) |
PLR | 1.0 (0.4–2.7) | 1.0 (nc) |
NLR | 0.6 (0.2–1.7) | nc |
Engel 1A | 44 | 20 |
Sensitivity | 100% (96–100%) | 100% (92–100%) |
Specificity | 6% (3–12%) | 3% (0.3–11%) |
PPV | 41% (32–50%) | 35% (23–49%) |
NPV | 100% (96–100%) | 100% (92–100%) |
DOR | nc | nc |
PLR | 1.1 (0.9–1.3) | 1.0 (0.7–1.5) |
NLR | 0.0 (0.0–0.0) | 0.0 (0.0–0.0) |
Lesionectomies in cases with MEG resection data . | ||
---|---|---|
Follow-up . | ≥1 year . | ≥5 years . |
n | 112 | 58 |
Engel 1 | 69 | 37 |
Sensitivity | 97% (91–100%) | 97% (88–100%) |
Specificity | 5% (2–11%) | 0% (0.2–6%) |
PPV | 62% (52–71%) | 63% (49–75%) |
NPV | 50% (40–60%) | 0.00% (0.2–6%) |
DOR | 1.6 (0.2–12.0) | 0.0 (nc) |
PLR | 1.0 (0.4–2.7) | 1.0 (nc) |
NLR | 0.6 (0.2–1.7) | nc |
Engel 1A | 44 | 20 |
Sensitivity | 100% (96–100%) | 100% (92–100%) |
Specificity | 6% (3–12%) | 3% (0.3–11%) |
PPV | 41% (32–50%) | 35% (23–49%) |
NPV | 100% (96–100%) | 100% (92–100%) |
DOR | nc | nc |
PLR | 1.1 (0.9–1.3) | 1.0 (0.7–1.5) |
NLR | 0.0 (0.0–0.0) | 0.0 (0.0–0.0) |
Only cases with IEDs on MEG and information on MEG resection were considered, corresponding to Table 2. nc = not computable due to zeros in the denominator, etc. DOR = diagnostic odds ratio; PLR/NLR = positive/negative likelihood ratio.
Discussion
In this study, we evaluate application of MEG for presurgical epileptic focus localization in 1000 patients over the course of 28 years. The results show that MEG provides non-redundant information, which significantly contributes to long-term seizure freedom after epilepsy surgery.
Sensitivity for IEDs
MEG detected IEDs in 72% of recordings. Detection rates in cases with suspected ETLE were higher in comparison to TLE (77% versus 68%) and were comparable between first and repeated recordings. These findings are compatible with previous reports on subgroups from the evaluated population (Stefan et al., 2003; Paulini et al., 2007), as well as other groups (Knake et al., 2006; Englot et al., 2015), which range between 70% and 78%. No systematic difference was seen comparing the different extratemporal compartments.
Earlier onset of epilepsy was significantly related to higher detection rates. Onsets were also earlier in patients with ETLE, suggesting that the difference in IED detection may be related to the different aetiologies in predominantly extra-temporal versus temporal localizations.
Detection rates varied over the years, probably due to usage of a whole-head system after 2009, evaluator experience and methodology. Between 1990 and 2000, more recordings presented with IEDs in comparison to 2001–09, although the Krenikon system had more restrictions than later systems. This apparent higher sensitivity may be a result of overinterpretation of e.g. benign epileptiform variants when the technique was still new and evaluators had limited experience. This is supported by the lower number of focal findings in the first decades, as e.g. normal variants or unspecific patterns show more distributed generators (Rampp et al., 2018) and thus result in diffuse localizations.
Comparison with presurgical diagnostics
Concordant presurgical findings are predictors of good postsurgical outcome (West et al., 2015). Therefore, we evaluated overlap of MEG and focus hypothesis, the consensus of presurgical diagnostics. MEG and focus hypothesis were consistent in 82%. The percentage was higher in TLE compared to ETLE cases (86% versus 80%, P = 0.014). Findings were completely concordant in 50%, with also a higher percentage in patients with putative TLE versus ETLE (49% versus 44%); however, the difference did not achieve statistical significance.
In cases in which MEG and presurgical evaluation were not completely concordant, MEG may suggest involvement of additional areas, which were not indicated by other methods. This interpretation is supported by studies comparing MEG and invasive EEG (Sutherling et al., 2008; Knowlton et al., 2009; Murakami et al., 2016). In a prospective study (Sutherling et al., 2008), MEG indicated additional areas in 13% and changed invasive EEG coverage in 23% of 69 patients. MEG changed the surgical decision in 20%. Murakami et al. (2016) similarly report that patients were more likely to become seizure free when MEG findings were resected completely and when stereo-EEG completely sampled the area suggested by MEG. In this light, our results suggest that the degree of added information is greater in ETLE cases, reflecting the known role of MEG in this subgroup (Knowlton, 2007).
Lower degrees of MEG concordance may be interpreted as lower redundancy, and thus the information added by MEG would be higher in cases with ETLE. This interpretation is supported by better outcomes after resection of focal localizations. However, the higher number of lobes included in the definition of ETLE may confound these results. Taken together, the available evidence suggests that MEG provides highly relevant information especially in ETLE.
MEG and epilepsy surgery
Selection of candidates for epilepsy surgery
Operated patients presented with more focal presurgical findings. As a circumscribed focus in an accessible region is a prerequisite for successful epilepsy surgery, this difference is not surprising. However, the aspect that the MEG results could be used as a surrogate for selection of surgical candidates may have practical consequences. MEG and source localization can be performed early in the diagnostic process to evaluate eligibility for epilepsy surgery (Ossenblok et al., 2007; Colon et al., 2009; Heers et al., 2010a, b) and is also viable in complex cases (Nissen et al., 2016). A focal finding at an early time point may inform the further diagnostic evaluation, enable more effective workflows and shorter delays to surgery, as implied by Knowlton (2006), for example. MEG findings consistent with the consensus of presurgical evaluation indicated patients with sustained good surgical outcome, as also demonstrated, for example, by Englot et al. (2015).
Surgical outcome
The relation of complete MEG resection and an Engel 1 outcome could be evaluated in a subgroup of 174 patients. The data shows that complete resections of focal MEG localizations are significantly related to Engel 1 (freedom of disabling seizures, P < 0.001, Table 1). This holds true even when Engel 1A outcomes are considered (complete seizure freedom, P = 0.004 and P = 0.005; Table 1). Only partial resections do not show this association. This finding is supported by a recent study (Murakami et al., 2016). The authors compared MEG localizations with stereo-EEG (sEEG) and surgical outcome. They similarly showed that complete resection of densely clustered spikes, adequately sampled by sEEG, was significantly related to seizure freedom in comparison to partial or no resection. Seizure outcome was restricted to a 12-month follow-up. The long follow-ups of our data now suggest that results may also be indicative of long-term outcomes. Comparison of MEG findings and the consensus of presurgical evaluation on a lobar level provided a less clear contrast, possibly due to the coarser comparison. Although MEG and the focus hypothesis may indicate the same lobe, a resection may not necessarily include the MEG findings.
Aetiology . | Cases . | Percentage of cases with MRI findings . |
---|---|---|
No lesion | 375 | – |
Lesional/MRI finding | 625 | – |
Hippocampal sclerosis | 86 | 14% |
FCD 2 | 25 | 4% |
Other dysplasia | 89 | 14% |
Malformations (incl. polymicrogyria, pachygyria, heterotopia, tuberous sclerosis and double cortex) | 80 | 13% |
Tumour | 112 | 18% |
Cystic lesions (except tumours) | 11 | 2% |
Inflammatory | 52 | 8% |
Traumatic brain injury | 45 | 7% |
Intracranial bleeding or ischaemia (incl. perinatal) | 38 | 6% |
Vascular malformations | 71 | 11% |
Other | 18 | 3% |
Unclear lesions | 16 | 3% |
Cases with multiple lesions/ aetiologies | 59 | 9% |
Missing data | 17 | 3% |
Aetiology . | Cases . | Percentage of cases with MRI findings . |
---|---|---|
No lesion | 375 | – |
Lesional/MRI finding | 625 | – |
Hippocampal sclerosis | 86 | 14% |
FCD 2 | 25 | 4% |
Other dysplasia | 89 | 14% |
Malformations (incl. polymicrogyria, pachygyria, heterotopia, tuberous sclerosis and double cortex) | 80 | 13% |
Tumour | 112 | 18% |
Cystic lesions (except tumours) | 11 | 2% |
Inflammatory | 52 | 8% |
Traumatic brain injury | 45 | 7% |
Intracranial bleeding or ischaemia (incl. perinatal) | 38 | 6% |
Vascular malformations | 71 | 11% |
Other | 18 | 3% |
Unclear lesions | 16 | 3% |
Cases with multiple lesions/ aetiologies | 59 | 9% |
Missing data | 17 | 3% |
FCD = focal cortical dysplasia.
Aetiology . | Cases . | Percentage of cases with MRI findings . |
---|---|---|
No lesion | 375 | – |
Lesional/MRI finding | 625 | – |
Hippocampal sclerosis | 86 | 14% |
FCD 2 | 25 | 4% |
Other dysplasia | 89 | 14% |
Malformations (incl. polymicrogyria, pachygyria, heterotopia, tuberous sclerosis and double cortex) | 80 | 13% |
Tumour | 112 | 18% |
Cystic lesions (except tumours) | 11 | 2% |
Inflammatory | 52 | 8% |
Traumatic brain injury | 45 | 7% |
Intracranial bleeding or ischaemia (incl. perinatal) | 38 | 6% |
Vascular malformations | 71 | 11% |
Other | 18 | 3% |
Unclear lesions | 16 | 3% |
Cases with multiple lesions/ aetiologies | 59 | 9% |
Missing data | 17 | 3% |
Aetiology . | Cases . | Percentage of cases with MRI findings . |
---|---|---|
No lesion | 375 | – |
Lesional/MRI finding | 625 | – |
Hippocampal sclerosis | 86 | 14% |
FCD 2 | 25 | 4% |
Other dysplasia | 89 | 14% |
Malformations (incl. polymicrogyria, pachygyria, heterotopia, tuberous sclerosis and double cortex) | 80 | 13% |
Tumour | 112 | 18% |
Cystic lesions (except tumours) | 11 | 2% |
Inflammatory | 52 | 8% |
Traumatic brain injury | 45 | 7% |
Intracranial bleeding or ischaemia (incl. perinatal) | 38 | 6% |
Vascular malformations | 71 | 11% |
Other | 18 | 3% |
Unclear lesions | 16 | 3% |
Cases with multiple lesions/ aetiologies | 59 | 9% |
Missing data | 17 | 3% |
FCD = focal cortical dysplasia.
Lesionectomy
Complete resection of an epileptogenic lesion was also related to a better outcome, compared to cases with partial or no resection. Extension to include neighbouring tissue did not result in different seizure outcomes. In the subgroup of operated patients with MEG findings, however, degree of lesionectomy was not related to outcome. This lack of influence is explained by almost all patients in this group having complete or extended lesionectomy. Only 5% presented with no or partial resection. MEG thus provides additional information above and beyond the degree of lesionectomy, which itself is based on the consensus of presurgical evaluation. This is further supported by the long-term results of Engel 1 outcomes in the subgroup of patients with complete or extended removal of an epileptogenic lesion: While there was no significant difference between complete or extended lesionectomy, complete MEG resection was associated with higher chances of a persisting Engel 1 outcome.
These data correlate well with the concept of epileptogenic lesions (Jehi, 2018): while the lesion plays a significant role in the generation of seizures and thus removal is necessary for seizure control, the epileptogenic zone itself may not be constrained to the lesion. At least in our data, this seems to be the rule, rather than the exception. By definition, the non-lesional part of the epileptogenic zone cannot be identified by structural features but only by its pathological function. MEG seems to be well suited to provide this functional information, at least when robust, focal results are available.
Methodological considerations
The spread of MEG dipoles significantly impacts how these can be taken into account for resection. Scattered localizations may not be amenable to complete resection and may be a sign of extended and potentially multifocal generation (Fischer et al., 2005). However, while truly extended epileptic areas may generate scattered activity, noise and subsequent low signal-to-noise ratio fabricate the false impression of extent (Bast et al., 2006). This may explain why partial resection of MEG also did not necessarily imply persisting seizures after surgery. Noise may have led to scattered localizations, which could not be resected completely. However, if the true epileptogenic zone was in fact focal, epilepsy surgery may of course nevertheless be successful. Because of this reasoning, robust, focal findings are most informative, as tight clusters of localizations or high quality averaged results are indirect evidence of both focal generation and low noise.
Averaging may reduce the influence of noise and thus provide more focal results in case of truly focal generation. In addition, averaging provides higher signal-to-noise ratio especially at the IED onset, which has been shown to more closely reflect the epileptogenic zone due to limited propagation at early time points (Bast et al., 2006; Mălîia et al., 2016; Kasper et al., 2018). However, averaging relies on the subjective classification of IEDs into morphologically similar groups. Investigators in our study using averaging techniques were experienced and compared morphology and topography of individual IEDs with the respective average patterns.
Subgroups
Brodbeck et al. (2011) evaluated the use of EEG based-source imaging for epilepsy surgery in 152 patients, 52 of whom had high-density EEG. High density EEG showed better diagnostic accuracy in comparison to low-density EEG, as well as PET, SPECT and MRI. Mean follow-up of 4 years captured seizure recurrence in the 2–5 years after surgery (de Tisi et al., 2011). Sensitivity and specificity were 80% and 88%, respectively, when Engel 1 and 2 outcomes after surgery were considered as reference standard. In ETLE patients, these results were considerably lower, e.g. sensitivity amounted to 75%. In contrast, TLE results were respectively better, with a sensitivity of 92%. This represents the complement to our MEG results, which provide higher diagnostic accuracy in ETLE (84% sensitivity, 89% specificity) and lower in TLE (56% and 77%), although more strictly evaluated for Engel 1 outcomes. These results underline the frequent suggestion to combine EEG and MEG to cover the complete spectrum of focal epilepsies.
This aspect is further supported by the limited diagnostic value in patient with mesial TLE. In these, MEG did not show a significant influence on postsurgical outcomes. Our data corroborate previous results on this well-known limitation (Leijten et al., 2003).
The association of Engel 1 outcomes with complete MEG resections was stronger in non-lesional versus lesional cases. This finding is concordant with previous studies highlighting the clinical value of MEG in this subgroup (Schneider et al., 2012; Englot et al., 2015). In the absence of a structural correlate, MEG localizations can provide crucial evidence for further evaluation e.g. for re-evaluation of MRI (Moore et al., 2002) or planning of invasive EEG (Schneider et al., 2012). Our results are based solely on interictal epileptic activity, reflecting the irritative zone. Beniczky et al. (2013) have rigorously evaluated source imaging of ictal EEG using presurgical evaluation as reference standard. They report a PPV of 92% and a NPV of 43% in 20 operated patients with mostly TLE. The likelihood ratio for matching source imaging and presurgical evaluation was nine times higher than non-matching results. This compares to a factor of 4.45 (PLR/NLR) in TLE cases in our data (PPV 82%, NPV 50%). Source imaging of ictal EEG seems superior to interictal MEG localization. However, in ETLE, this factor amounts to 41.9 (PPV 87%, NPV 86%), which clearly exceeds ictal EEG imaging.
Limitations
While MEG analysis was performed before surgery, evaluations regarding outcome, resection volumes and presurgical diagnostics were performed retrospectively, implicating the limitations of a retrospective study. Specifically, patient selection was not based on considerations regarding the study to ensure comparability, avoid bias, etc. Furthermore, due to the span of almost three decades, evaluation procedures and integration into clinical routine varied over the years, however always included dipole localizations. In general, a heterogeneous spectrum of methods may hinder broader adoption of MEG. This aspect is being addressed with the definition and implementation of clinical practice guidelines (Hashimoto et al., 2005; Bagic et al., 2009; De Tiège et al., 2017; Hari et al., 2018). The resulting comparability and quality standards may then also contribute to better worldwide availability of reimbursement.
The long time span also limited the available level or accuracy of MEG findings. A portion of the reports did not specify exact localization coordinates in a structured manner, which for example limited the eligibility for the finer comparison with resection volumes. The manuscript therefore illustrates the clinical value rather than fine localization accuracy, which could potentially be addressed with direct comparison to invasive EEG.
While complete removal of an epileptogenic lesion was associated with better postsurgical outcomes, extending the resection did not show better results. However, the group of extended lesionectomies includes rather different surgical strategies, depending on the operated lobe and lesion. For example, removal of a hemosiderin ring around a cavernoma was considered as extended lesionectomy, just like a standard temporal lobe resection. This may have led to less clear results in regard to seizure outcome.
It is also expected that the type of pathology influences surgical outcome (Blumcke et al., 2017). We did not investigate this in detail due to the long time span covered by the study, during which imaging and classifications of epileptogenic lesions have changed considerably.
Conclusions
Evaluation of the largest cohort with the longest follow-up to date revealed that MEG provides non-redundant information, which may be utilized for selection of epilepsy surgery candidates, adds to presurgical focus localization and significantly contributes to long-term seizure freedom after epilepsy surgery. In ETLE and non-lesional cases, MEG provides optimal accuracy.
Abbreviations
- (E)TLE
(extra)temporal lobe epilepsy
- IED
interictal epileptic discharges
- MEG
magnetoencephalography
- NPV
negative predictive value
- PPV
positive predictive value
Acknowledgements
We thank Mrs Martina Rzonsa for her invaluable work with recording data and unearthing clinical information from decades of clinical work.
Funding
This study was supported by the Deutsche Forschungsgemeinschaft (RA 2062/1–1, WO 1425/7–1). X.W. was supported by the ‘Förderverein Neurochirurgische Forschung’ of the University Hospital Erlangen, Germany.
Competing interests
The authors report no competing interests.