Automatic ictal onset source localization in presurgical epilepsy evaluation

Clinical Neurophysiology 129 (2018) 1291–1299

Contents lists available at ScienceDirect

Clinical Neurophysiology journal homepage: www.elsevier.com/locate/clinph

Automatic ictal onset source localization in presurgical epilepsy evaluation Johannes Koren a, Gerhard Gritsch c, Susanne Pirker a, Johannes Herta b, Hannes Perko c, Tilmann Kluge c, Christoph Baumgartner a,d,⇑ a Karl Landsteiner Institute for Clinical Epilepsy Research and Cognitive Neurology, Neurological Department, General Hospital Hietzing with Neurological Center Rosenhügel, Vienna, Austria b Department of Neurosurgery, Medical University of Vienna, Vienna, Austria c Austrian Institute of Technology GmbH (AIT), Safety & Security Department, Vienna, Austria d Department of Epileptology and Clinical Neurophysiology, Sigmund Freud University, Vienna, Austria

a r t i c l e

i n f o

Article history: Accepted 24 March 2018 Available online 6 April 2018 Keywords: Ictal Source localization Electric source imaging EpiSource Epilepsy surgery

h i g h l i g h t s  Ictal onset source localization (IOSL) showed high diagnostic accuracy during presurgical evaluation.  IOSL contributes to a correct localization of the seizure onset zone on a sublobar level.  IOSL can be obtained within 5 minutes per seizure and used in standard epilepsy monitoring settings.

a b s t r a c t Objective: To test the diagnostic accuracy of a new automatic algorithm for ictal onset source localization (IOSL) during routine presurgical epilepsy evaluation following STARD (Standards for Reporting of Diagnostic Accuracy) criteria. Methods: We included 28 consecutive patients with refractory focal epilepsy (25 patients with temporal lobe epilepsy (TLE) and 3 with extratemporal epilepsy) who underwent resective epilepsy surgery. Ictal EEG patterns were analyzed with a novel automatic IOSL algorithm. IOSL source localizations on a sublobar level were validated by comparison with actual resection sites and seizure free outcome 2 years after surgery. Results: Sensitivity of IOSL was 92.3% (TLE: 92.3%); specificity 60% (TLE: 50%); positive predictive value 66.7% (TLE: 66.7%); and negative predictive value 90% (TLE: 85.7%). The likelihood ratio was more than ten times higher for concordant IOSL results as compared to discordant results (p = 0.013). Conclusions: We demonstrated the clinical feasibility of our IOSL approach yielding reasonable high performance measures on a sublobar level. Significance: Our IOSL method may contribute to a correct localization of the seizure onset zone in temporal lobe epilepsy and can readily be used in standard epilepsy monitoring settings. Further studies are needed for validation in extratemporal epilepsy. Ó 2018 Published by Elsevier B.V. on behalf of International Federation of Clinical Neurophysiology.

Abbreviations: amTLR, anteromesial temporal lobe resection; ESI, electrical source imaging; IOSL, ictal onset source localization; LAURA, local autoregressive average; LORETA, low resolution electromagnetic tomography; MRI, magnetic resonance imaging; MUSIC, multiple signal classification; sAHE, selective amygdala-hippocampectomy; sLORETA, standardized low resolution electromagnetic tomography; SMAC, spherical model with anatomical constraints; SNR, signalto-noise ratio,; SOZ, seizure onset zone; SPECT, single-photon emission computed tomography; TLE, temporal lobe epilepsy; mTLE, mesial temporal lobe epilepsy; TLR, temporal lobe resection; vEEG, video electroencephalography. ⇑ Corresponding author at: Karl Landsteiner Institute for Clinical Epilepsy Research and Cognitive Neurology, Neurological Department, General Hospital Hietzing with Neurological Center Rosenhügel, Riedelgasse 5, 1130 Vienna, Austria. E-mail address: [email protected] (C. Baumgartner).

1. Introduction Epilepsy surgery is a valuable treatment option for patients with medically refractory epilepsy (Rosenow and Luders, 2001). Successful surgical treatment depends on a thorough presurgical evaluation localizing the epileptogenic zone and essential brain regions in each individual patient (Rosenow and Luders, 2001). Video-EEG monitoring is one of the cornerstones of each presurgical evaluation with interictal EEG providing information on the

https://doi.org/10.1016/j.clinph.2018.03.020 1388-2457/Ó 2018 Published by Elsevier B.V. on behalf of International Federation of Clinical Neurophysiology.

1292

J. Koren et al. / Clinical Neurophysiology 129 (2018) 1291–1299

irritative zone and ictal EEG localizing the seizure onset zone (SOZ). While routine clinical practice still relies on visual analysis of EEG data, electrical source imaging (ESI) facilitates attribution of epileptiform EEG discharges to the three-dimensional intracerebral location of their neuronal generators and thus significantly increases the localizing information of EEG (Brodbeck et al., 2011). Most studies on ESI concentrated on interictal EEG data and thus on better delineation of the irritative zone because spike averaging improves signal-to-noise ratio and therefore robustness of source modeling (Bast et al., 2006; Leijten and Huiskamp, 2008; Brodbeck et al., 2011; Scherg et al., 2012; Megevand et al., 2014). However, accurate localization of the SOZ by ESI of ictal EEG data (ictal onset source localization – IOSL) seems even more important for surgical planning (Merlet and Gotman, 2001). So far several studies using different ictal source localization techniques have reported concordance between 37.5% and 100% with the actual SOZ (Assaf and Ebersole, 1997; Blanke et al., 2000; Lantz et al., 2001; Jung et al., 2009; Lee et al., 2009; Stern et al., 2009; Holmes et al., 2010; Yang et al., 2011; Lu et al., 2012; Beniczky et al., 2013; Breedlove et al., 2014; Akdeniz 2016). Various reference standards for IOSL confirmation were used: interictal EEG (Boon and D’Have, 1995; Koutroumanidis et al., 2004; Valentin et al., 2014), intracranial EEG (Assaf and Ebersole, 1997; Lantz et al., 2001; Merlet and Gotman, 2001; Koessler et al., 2010), postsurgical outcome (Assaf and Ebersole, 1999; Lantz et al., 1999; Blanke et al., 2000; Jung et al., 2009; Lee et al., 2009; Lu et al., 2012; Breedlove et al., 2014), MRI (Worrell et al., 2000), ictal SPECT (Beniczky et al., 2006; Habib et al., 2016), decision of the multidisciplinary epilepsy surgery team (Beniczky et al., 2013) or a combination thereof (Boon et al., 2002; Ding et al., 2007; Stern et al., 2009; Holmes et al., 2010; Yang et al., 2011). Only one recent publication using a distributed source model (LAURA) to localize ictal activity reported precise performance measures (sensitivity of 70%, specificity of 76% and PPV of 92%) (Beniczky et al., 2013). Several drawbacks and difficulties regarding ictal source localization analysis of scalp EEG recordings have to be considered: possible low signal-to-noise ratio, lack of ictal EEG correlates in scalp recordings during seizure onset, rapid propagation or already propagated ictal patterns in scalp EEGs and artifacts obscuring EEG seizure patterns (Pacia and Ebersole, 1997; Alarcon et al., 2001; Foldvary et al., 2001; Rosenow and Luders, 2001; Boon et al., 2002; Beniczky et al., 2013). Most important no standard method of IOSL has been established so far (dipole modeling, LORETA, sLORETA, MUSIC, LAURA, etc.) and most methods require highly interactive analysis of ictal EEG patterns including individual parameter adjustments which complicates the use of IOSL in clinical practice (Koessler et al., 2010). We developed a new automatic algorithm which requires only visual selection of the EEG pattern at ictal onset. The algorithm then automatically performs source localization without further interactions and parameter adjustments by the user, making IOSL results easy to obtain, reproducible and objective. Solutions can be obtained within five minutes per seizure. Therefore the algorithm can be used in everyday clinical practice in the epilepsy monitoring unit. We tested the algorithm’s diagnostic accuracy in a standard long-term video-EEG monitoring setting following STARD (Standards for Reporting of Diagnostic Accuracy) criteria. Postoperative outcome two years after resective epilepsy surgery was used as reference standard. We hypothesized that IOSL results correctly localized the SOZ if IOSL matched the actual resection site on a sublobar level and patients were seizure free after epilepsy surgery.

2. Methods 2.1. Patient selection We searched our database for patients with refractory focal epilepsy who were admitted for presurgical evaluation in our center and subsequently underwent resective epilepsy surgery. We included all patients for whom raw EEG data was available and identified in this way 30 consecutive operated patients. All patients gave their informed consent prior to being admitted to long-term video EEG (vEEG) monitoring. The local ethics committee approved the study. 2.2. EEG data Long-term vEEGs were recorded from 23 electrodes placed according to the Extended International 10–20-system (including additional ‘true’ anterior temporal electrodes FT9/FT10) and TP9/ TP10 at a 256 Hz sampling rate using a Micromed EEG recording system (SystemPlus Evolution, Veneto, Italy). Patients’ EEG recordings were visually analyzed by board certified electroencephalographers (JK, SP and CB). We visually determined time, location and frequency of the EEG pattern at onset of every seizure defined according to criteria previously published (Foldvary et al., 2001). We excluded seizures obscured by artifacts from further analysis. All included seizures were anonymized and randomized. IOSL was applied to the pattern at onset of these seizures by an independent reviewer (GG) who was blinded to all clinical data. 2.3. Visual EEG seizure onset localization We systematically localized seizure onset zones visually according to criteria proposed by Foldvary et al. (2001). Specifically we distinguished between the following seizure onset localizations: 1. Generalized seizure onset: activity involving multiple electrodes over both hemispheres having a less than 2:1 amplitude predominance over one hemisphere; 2. Lateralized seizure onset: activity involving multiple electrodes over multiple lobes of a single hemisphere having a 2:1 or greater amplitude predominance over this hemisphere; 3. Regional or lobar seizure onset: activity involving electrodes overlying a single lobe having a 2:1 or greater amplitude predominance than that seen over other regions of the same hemisphere; 4. Focal or sublobar seizure onset: activity with a maximum at a single electrode with no more than 2 contiguous electrodes within 80% to 100% of the maximum amplitude (Foldvary et al., 2001). In temporal lobe seizures, we assigned a medio-basal seizure onset localization if ictal activity fulfilled these criteria with a maximum at electrodes FT9 or FT10, respectively and a lateral temporal seizure onset localization if ictal activity fulfilled these criteria with a maximum at electrodes T7 or T8, respectively. 2.4. Ictal onset source localization The core idea of our ictal onset source localization (IOSL) technique was to automatically determine the most dominant rhythmic EEG pattern within the earliest ictal activity, i.e. the first change in EEG time–frequency plots. Next, we implemented a frequency dependent time window which had to contain at least eight ictal waves or discharges (e.g. 4 Hz ictal activity = time window of 2 s; 8 Hz ictal activity = time window of 1 s) to the selected ictal activity. The spatial distribution of this rhythmic activity over all EEG electrodes was the basis for our source localization method, leading to an automatic localization approach. The inverse method used in our study was a frequency domain version of the minimum

J. Koren et al. / Clinical Neurophysiology 129 (2018) 1291–1299

variance beamformer (MVB) based on a Spherical Model with Anatomical Constraints (SMAC) (Spinelli et al., 2000) head model derived from the Colin 27 Average Brain (Montreal Neurological Institute) (Gritsch et al., 2011). The MVB tends to determine ictal activity as a more focal solution rather than a distributed one (Robinson et al., 1999; Gritsch et al., 2011). Finally, the spatial origin of ictal EEG activity was visualized in a graphical user interface, presenting the corresponding neural activity as color coded overlay in coronal, sagittal and axial slices on a standard MRI. No activity was represented as dark blue, activities in the range of 50% of the maximum strength were colored green and the region of maximum activity was visualized in red. We used a threshold of 50% for depicting IOSL results and concentrated on the maximum activity. With this commercially available software tool (EpiSource; www. encevis.com) all representative seizures were blindly analyzed by one co-author (GG). The graphical results of the calculated ictal activity at the time of seizure onset were saved. The complete source localization procedure took approximately five minutes per seizure. 2.5. Agreement of visual seizure onset localization, IOSL and surgical resection location We determined the agreement between visual seizure onset localization, IOSL and surgical resection location using the following location categories with respect to previous studies (Beniczky et al., 2013): left temporal medio-basal, right temporal mediobasal, left temporal lateral, right temporal lateral, left temporal medio-basal and lateral, right temporal medio-basal and lateral, extratemporal. We chose these categories because most subjects in our study suffered from temporal lobe epilepsy (TLE subgroup; 25 of 28 patients, 89%). 2.6. Postoperative outcome We determined two-years postoperative seizure outcome of all included patients according to the ILAE classification (Wieser et al., 2001). We divided patients into two groups based on postoperative outcome two years after epilepsy surgery:

1293

TN/[TN + FN]), positive likelihood ratio (LR+; sensitivity/[1–specifi city]) and negative likelihood ratio (LR-; [1–sensitivity]/specificity) were calculated. Performance measures for visual seizure onset localization were calculated in the same manner as for IOSL. 2.8. Statistical analysis We determined agreement between visual seizure onset localization, IOSL and surgical resection location with weighted Cohen’s Kappa. Values were interpreted according to the following criteria: no agreement (k  0), slight (0.01–0.2), fair (0.21–0.4), moderate (0.41–0.6), substantial (0.61–0.8) and almost perfect agreement (>0.8) (Landis et al., 1977). Sensitivity and specificity of the IOSL method was statistically tested using Chi-Square test. Yates’ correction for continuity was used if necessary. Level of significance was set to p = 0.05 for all statistical tests. We used Microsoft Excel 2007 and IBM SPSS Statistics 19 for calculations and statistical testing.

3. Results We included 28 consecutive operated patients in our study (Fig. 1). 25 subjects (89.2%) suffered from temporal lobe epilepsy (TLE subgroup). Patients showed the following demographic characteristics: mean age at surgical resection was 41 years (range 15–61) and 57.1% (16) were female. Individual clinical characteristics, visual seizure onset localization, IOSL localization, resection site and postoperative seizure outcome are shown in Table 1. We excluded two patients because all their recorded seizures were obscured by artifacts. The 28 included patients had a total of 146 seizures during long-term video EEG monitoring (range: 1–15 seizures per patient). We had to exclude 76 seizures from further analysis due to obscuration by artifacts. Thus, IOSL could be applied in 70 seizures (range: 1–6 seizures per patient).

(1) seizure-free group: patient achieved sustained complete seizure freedom (including freedom from auras) after epilepsy surgery (ILAE Class 1a) (2) seizure group: patient did not achieve complete seizure freedom after epilepsy surgery (ILAE Class 1 to 6). 2.7. Performance analysis of localization results We defined the following reference standard for seizure onset zone localization: seizure-freedom (including freedom from auras; ILAE Class 1a) for at least two years after epilepsy surgery. We compared IOSL results with our defined reference standard. If IOSL results matched surgical resection location exactly or if resection location was more extended but still contained the IOSL result completely in a seizure free patient, then IOSL was classified as true positive (TP). If a single patient got more than one seizure, IOSL results of every single seizure had to match the surgical resection location in a seizure free patient in order to be classified as TP. If IOSL and surgical resection location did not match in patients with ongoing seizures, IOSL was classified as true negative (TN). False positive (FP) localization results were defined as IOSL matching the actual resection location in a patient with ongoing seizures. False negative (FN) localization results were defined as IOSL not matching the actual resection location in a seizure free patient. Sensitivity (TP/[TP + FN]), specificity (TN/[TN + FP]), positive predictive value (PPV; TP/[TP + FP]), negative predictive value (NPV;

Fig. 1. Flowchart of the study; Concordant localization is a match between ictal onset source localization (IOSL) and the actual epilepsy surgery resection location at sublobar level. Discordant localization is no match between ictal onset source localization (IOSL) and the actual epilepsy surgery resection location at sublobar level.

1294

J. Koren et al. / Clinical Neurophysiology 129 (2018) 1291–1299

Table 1 Patients clinical characteristics. MRI: magnetic resonance imaging; L: left; R: right; FCD: focal cortical dysplasia; ILAE: International league against epilepsy. Patient

Age at surgery

MRI

Visual seizure onset localization

Ictal onset source localization

Resection site

Two-years postoperative outcome

#01

37

Incidental lesion right temporal

R hemispheric

R temporal lobe resection

ILAE class 4

#02

50

L hypoplastic anterior temporal lobe

R temporal

28

R lateral frontal FCD

R temporal

R anteromesial temporal lobe resection R lateral frontal lesionectomy

ILAE class 1a

#03

#04

45

Normal

53

L hippocampal sclerosis

#06

15

L hippocampal sclerosis

L hemispheric

#07

53

L hippocampal atrophy

L temporal

R lateral temporo-occipital resection L anteromesial temporal lobe resection L anteromesial temporal lobe resection L selective amygdalohippocampectomy

ILAE class 4

#05

R hemispheric and R parietal L temporal

R medio-basal and lateral temporal R temporo-parietal R medio-basal temporal R medio-basal and lateral temporal R insula R medial occipital lobe

#08

51

L medio-basal-and-lateral temporal posthemorrhagic lesion

L temporal

L temporal lobe resection

ILAE class 3

#09

28

Normal

R temporal

ILAE class 4

#10

37

Normal

#11

37

R hippocampal atrophy

R hemispheric and R temporal R temporal

#12

35

Normal

L temporal

#13

50

L hippocampal atrophy

L temporal

#14

29

L hippocampal atrophy

L temporal

#15

49

Normal

L temporal

#16

29

L hippocampal sclerosis

L hemispheric

#17

51

L hippocampal sclerosis

#18

44

L temporo-occipital cystic lesion

L hemispheric and L temporal L temporal

#19

61

R hippocampal atrophy

R hemispheric

#20

52

R hippocampal sclerosis

R temporal

#21

39

R hippocampal sclerosis

#22

52

L hippocampal sclerosis

R temporal, L temporal L hemispheric

#23

38

Normal

R temporal

#24

31

R hippocampal sclerosis

R temporal

#25

58

Normal

R temporal

R anteromesial temporal lobe resection R anteromesial temporal lobe resection R anteromesial temporal lobe resection L anteromesial temporal lobe resection L anteromesial temporal lobe resection L anteromesial temporal lobe resection L selective amygdalohippocampectomy L selective amygdalohippocampectomy L anteromesial temporal lobe resection L medial temporo-occipital lesionectomy R anteromesial temporal lobe resection R selective amygdalohippocampectomy R anteromesial temporal lobe resection L selective amygdalohippocampectomy R anteromesial temporal lobe resection R selective amygdalohippocampectomy R temporal lobe resection

#26

41

L medio-basal temporal ganglioglioma

L temporal

#27

31

R occipital postoperative lesion

R temporal

#28

37

R hippocampal sclerosis

R temporal

3.1. Agreement between visual seizure onset localization and surgical resection location Visual seizure onset localization revealed four patients with a lateralized, right hemispheric seizure onset (14.3%; within this group, one patient had also an additional regional, right temporal seizure onset and one patient an additional regional, right

L medio-basal temporal L medial frontal L insula L medio-basal temporal L lateral frontal L medio-basal and lateral temporal L insula R medio-basal temporal R medio-basal temporal L medio-basal temporal L medio-basal temporal L medio-basal temporal L medio-basal temporal L medio-basal temporal L medio-basal and lateral temporal L medio-basal temporal L medio-basal temporal R medio-basal temporal R medio-basal and lateral temporal R medio-basal temporal L medio-basal temporal R medio-basal temporal R medio-basal temporal R medio-basal and lateral temporal L medio-basal temporal R medio-basal temporal R medio-basal temporal

L medio-basal temporal lesionectomy R selective amygdalohippocampectomy R selective amygdalohippocampectomy

ILAE class 5

ILAE class 3 ILAE class 1a ILAE class 3

ILAE class 1a ILAE class 3 ILAE class 1a ILAE class 1a ILAE class 3 ILAE class 1a ILAE class 4 ILAE class 1a ILAE class 3 ILAE class 1a ILAE class 3 ILAE class 3 ILAE class 1a ILAE class 3 ILAE class 1a ILAE class 5 ILAE class 1a ILAE class 1a ILAE class 1a

parietal seizure onset), four patients with a lateralized, left hemispheric visual seizure onset (14.3%; within this group, one patient showed an additional regional, left temporal seizure onset), 10 subjects with a regional, right temporal seizure onset (35.7%), nine subjects with a regional, left temporal seizure onset (32.1%) and one patient with separate regional, right and left temporal seizure onset localizations (3.6%; Table 1). It needs to

J. Koren et al. / Clinical Neurophysiology 129 (2018) 1291–1299

be stressed that a sublobar localization based on our visual EEG analysis criteria could not be achieved in any seizure. Therefore, we could not assess an agreement on the sublobar level. 3.2. Agreement between ictal onset source localization and surgical resection location Kappa values showed substantial agreement between ictal onset source localization (IOSL) and actual resection site for all included patients (0.618) and the TLE subgroup (0.645). IOSL showed consistent localization results in 23 patients. 5 patients showed multiple different ictal onset source localizations for different seizures (Table 1 and Fig. 2). IOSL results were localized within the surgical resection site in 18 of 28 patients (64.3%) of the entire patient group and in 18 of 25 patients (72%) in the TLE subgroup (Fig. 1). 8 patients (28.6%) underwent selective amygdalohippocampectomy, 13 (46.4%) had an anteromesial temporal lobe resection, three (10.7%) had a temporal lobe resection and four (14.3%) had a lesionectomy (1 temporal, 3 extratemporal). The results of IOSL, surgical resection site and postoperative seizure outcome are shown in Fig. 2. In 19 of 28 patients (67.9%) IOSL localized the seizure onset zone exclusively to the medio-basal temporal compartment. In five of these patients (26.3%) a selective amygdala-hippocampectomy (sAHE) was performed, all of them became seizure free and therefore were classified as TP. In 12 patients (63.1%) an anteromesial temporal lobe resection (amTLR) was performed, six patients (50%) became seizure free and were classified as TP. Five patients (41.7%) were not seizure free after surgery and classified as FP. One patient (8.3%) was localized to the contralateral medio-basal temporal lobe, did not become seizure free and was classified as TN. In two patients (10.6%) a lesionectomy was performed: one subject had a medio-basal temporal ganglioglioma, was seizure free after surgery and therefore classified as TP. The other patient had a temporo-occipital cystic lesion, was not seizure free after surgery and therefore classified as TN. Three patients (10.7%) were localized to the medio-basal and the lateral temporal compartment. Two patients (66.7%) underwent a sAHE, did not became seizure free and thus were considered as TN. The other patient had a temporal lobe resection (TLR), was not seizure free after surgery and therefore was classified as FP. In one patient (3.6%) IOSL localized to a single extratemporal compartment, namely to the medial occipital lobe. This patient underwent a lateral temporo-occipital resection. He did not became seizure free and was therefore classified as TN. 5 patients (17.8%) showed multiple different ictal onset source localizations for different seizures: One patient showed a frontal medial and insular IOSL and underwent an amTLR. He became seizure free and thus considered as FN. In one patient with a mediobasal, lateral temporal and temporo-parietal IOSL a TLR was performed. In two patients IOSL localized to the medio-basal and the lateral temporal compartment as well as to the insular cortex, one patient underwent a TLR, the other one a lateral frontal lesionectomy. The fifth patient showed a medio-basal temporal and lateral frontal IOSL and underwent a sAHE. The latter four subjects did not become seizure free and therefore were classified as TN. Chi-Square test after Yates correction revealed statistical significance of our IOSL results (p = 0.013). Results shifted just slightly in the TLE subgroup but were not statistical significant anymore after Yates correction (p = 0.056). Fig. 3 shows an example of a concordant IOSL result: A 37-yearold male with right mesial TLE and hippocampal sclerosis, who underwent a selective amygdalohippocampectomy and was

1295

seizure free for at least two years after epilepsy surgery. IOSL localized to the right medio-basal temporal lobe. Fig. 4 shows an example of a discordant IOSL result: A 28-year-old female with focal cortical dysplasia (FCD) in the right frontal lobe who underwent a lesionectomy and was not seizure free after epilepsy surgery. IOSL were outside the lesions in the medio-basal and the lateral temporal compartment as well as in the insular cortex. Seven patients were MRI negative, meaning no lesion could be identified on preoperative MRI scans. In this specific patient subgroup IOSL results showed concordance with the actual resection site in six cases (85.7%). 3.3. Performance measures of visual seizure onset localization and IOSL For visual EEG seizure onset localization, we could not calculate performance measures on the sublobar level. We therefore calculated performance measures on the lobar level: eight TP (28.6%), six TN (21.4%), nine FP (32.1%) and five FN (17.9%); Sensitivity of visual EEG analysis for the entire patient group was 61.5%, specificity was 40%, PPV was 47.1% and NPV was 54.6%. For IOSL, we observed 12 TP (42.9%), nine TN (32.1%), six FP (21.4%) and one FN (3.6%) on the sublobar level (Fig. 2). Performance measures of IOSL for the entire patient group and for the TLE subgroup were as follows: – – – – – –

Sensitivity 92.3% (entire patient group); 92.3% (TLE subgroup) Specificity 60%; 50% Positive predictive value (PPV) 66.7%; 66.7% Negative predictive value (NPV) 90%; 85.7% Positive likelihood ratio (LR+) 2.31; 1.85 Negative likelihood ratio (LR ) 0.13; 0.15

4. Discussion We studied a novel automatic ictal onset source localization (IOSL) method (frequency domain version of the minimum variance beamformer) in 28 consecutive patients with refractory focal epilepsy who underwent resective epilepsy surgery applying STARD criteria. IOSL was performed on individual seizure data obtained during presurgical video-EEG monitoring using a standard head model. Sensitivity of our IOSL method was very high (92.3%) while specificity was moderate (60%). Assaf and Ebersole (1997) reported low sensitivity between 36% and 66% and high specificity between 92% and 96% in 40 patients with TLE using multiple fixed dipole modeling (Assaf and Ebersole, 1997). Other studies showed good agreement (63–100%) of IOSL with postsurgical outcome or intracranial EEG recordings in small patient cohorts (5–15 subjects) (Blanke et al., 2000; Lantz et al., 2001; Stern et al., 2009; Holmes et al., 2010; Yang et al., 2011; Akdeniz 2016). Boon et al. (2002) included 100 patients but IOSL could be applied in only 31 patients due to artifacts. Ultimately IOSL played a key role in the surgical decision process in 14 patients but performance measures were not reported. This publication is the only prospective IOSL study to date to the best of our knowledge (Boon et al., 2002). Only one previous blinded study reported performance measures following STARD criteria so far (Beniczky et al., 2013). In this study a sensitivity of 70% and a specificity of 76% was reported. Likelihood ratios were similar to our study: We found a positive likelihood ratio (LR+) of 2.31 (corresponding to concordant IOSL results) and a negative likelihood ratio of (LR ) of 0.13 (corresponding to discordant results), while Beniczky et al. (2013) reported a LR+ of 3 and a LR of 0.33. This significant difference of LR+ versus LR proves the clinical value of ictal onset source localization.

1296

J. Koren et al. / Clinical Neurophysiology 129 (2018) 1291–1299

Fig. 2. Flowchart of ictal onset source localization results, resection sites and postoperative seizure outcome; * = one patient was localized to the contralateral medio-basal temporal lobe, did not became seizure free after surgery and therefore was considered as true negative; # = one patient underwent a mesial temporal lobe lesionectomy and one patient had a medial temporo-occipital lesionectomy; à = this specific patient was localized to the medial occipital lobe and underwent a lateral temporo-occipital resection; sAHE: selective amygdala-hippocampectomy; amTLR: anteromesial temporal lobe resection; TLR: temporal lobe resection; TP: true positive; FP: false positive; TN: true negative; FN: false negative.

Our IOSL method showed a moderate positive predictive value (PPV) of 66.7% and a high negative predictive value (NPV) of 90%. We believe that the high NPV represents a main finding of our study with immediate clinical implications. If IOSL

does not match the planned resection site in a given patient, a reevaluation of the current surgical hypothesis needs to be undertaken before proceeding to surgery with a high risk of surgical failure.

J. Koren et al. / Clinical Neurophysiology 129 (2018) 1291–1299

1297

Fig. 3. Example of a patient with concordant ictal onset source localization (IOSL) results: 37-year-old male with right mesial temporal lobe epilepsy (TLE) and hippocampal sclerosis. The corresponding ictal EEG shows a high signal-to-noise ratio and high rhythmicity. The depicted IOSL images show coronal, sagittal and axial slices with highest seizure activity. The time-frequency plot detects changes in frequency (in Hz) over time of the analyzed seizure.

Fig. 4. Example of a patient with discordant ictal onset source localization (IOSL) results: 28-year-old female with focal cortical dysplasia in the right frontal lobe. The corresponding ictal EEG shows a low signal-to-noise ratio and rather weak rhythmicity. The depicted IOSL images show coronal, sagittal and axial slices with highest seizure activity. The time-frequency plot detects changes in frequency (in Hz) over time of the analyzed seizure. IOSL were outside the lesions in the medio-basal and the lateral temporal compartment as well as in the insular cortex.

Beniczky et al. (2013) reported a high PPV of 92% and very low NPV of 43%. We assume that this discrepancy can be explained by the different validation method used in our study. Beniczky et al.

(2013) validated IOSL results by multidisciplinary consensus of the epilepsy surgery team. On the contrary, we used resection site and postoperative seizure outcome two years after surgery to

1298

J. Koren et al. / Clinical Neurophysiology 129 (2018) 1291–1299

validate the results of IOSL for the following reasons: Ictal EEG and therefore IOSL delineate the seizure onset zone (SOZ). Because we did not use invasive recordings and visual analysis of scalp-EEG provides no accurate information on the location of the SOZ, we chose the theoretical concept of the epileptogenic zone, i.e. the area of brain which is indispensable to generate seizures, as reference standard for IOSL localizations (Rosenow and Luders, 2001). Usually the epileptogenic zone contains the SOZ. We assumed that the epileptogenic zone had been removed if a patient was completely seizure-free since surgery. However, it cannot be excluded, that the resection volume actually was larger than the epileptogenic zone and the resection contained but was not identical to the epileptogenic zone. We assumed that IOSL results were correct if they were contained within the resection site and the patient was completely seizure-free since surgery. It should be noted that IOSL results never exactly matched the resection volume and that our validation method therefore only represents a rough approximation for an accurate localization of the SOZ. Nevertheless, our approach seems reasonable from a clinical standpoint because it potentially provides additional information to guide resection in individual patients. Furthermore, this approach was used in several previous studies to validate electrical source imaging (ESI) (Assaf and Ebersole, 1997; Lantz et al., 1999; Blanke et al., 2000; Boon et al., 2002; Jung et al., 2009; Lee et al., 2009; Stern et al., 2009; Holmes et al., 2010; Yang et al., 2011; Lu et al., 2012; Breedlove et al., 2014). In the study of Beniczky et al., 20 out of 42 included patients were operated (Beniczky et al., 2013). They applied Engel’s classification and considered patients as seizure-free if they belonged to Engel class I after one year, i.e. were free of seizures with loss of awareness. We used the ILAE classification and classified patients as seizure-free only if they achieved Class 1a outcome, i.e. were completely seizure and aura free for two years. Therefore the proportion of 46% seizure free patients in our study is considerably lower compared to the 80% seizure free rate in Beniczky’s study. However, our surgical results compare very well to the results of de Tisi et al., who found an overall probability of 49% for remaining completely seizure and aura free by two years post-surgery (de Tisi et al., 2011). One major limitation of our study is that the majority (89.2%) of our patients suffered from temporal lobe epilepsy (TLE). Furthermore most TLE patients suffered from mesial TLE (mTLE) and thus subsequently underwent resections limited to the mesial temporal compartment. Overall 8 patients had a selective amygdalohippocampectomy and 13 an anteromesial temporal lobe resection, summing up to 75% of all surgical interventions in our cohort. Furthermore mTLE patients comprise nearly all patients with sustained seizure free outcome after surgery (11 out of 12 seizure free patients; 91%). Thus, our general conclusions only apply to temporal lobe epilepsy and cannot be readily extrapolated to extratemporal cases. Further studies are needed to validate our method in extratemporal epilepsy. The four lesionectomy cases (one temporal, three extratemporal) in our cohort showed the following results: one subject with a medio-basal temporal ganglioglioma showed a concordant IOSL results and was seizure free after surgery. The other three patients suffering from extratemporal lesions (one frontal, two occipital) showed discordant IOSL results, though IOSL localizations were in close proximity to the lesion and the actual resection site. None of these patients became seizure free after surgery, resulting in a perfect performance of the IOSL method in these patients. We demonstrated the added value of automatic IOSL in comparison to visual EEG seizure onset localization because a sublobar localization was only possible with IOSL and not with visual EEG analysis. Performance measures for visual seizure onset localization could not be calculated on the sublobar level and therefore

not directly compared to the sublobar performance measures of IOSL. Nevertheless, performance measures on the sublobar level of our automatic method were superior even to performance measures on the lobar level of visual analysis in our patient cohort. Some important methodological issues of ictal source localization need to be addressed: First, low signal-to-noise ratio (SNR) is a major drawback. Previous publications solved this issue by averaging ictal wave-forms (Assaf and Ebersole, 1997; Merlet and Gotman, 2001; Beniczky et al., 2006, 2013). We localized seizures with high SNR without averaging and excluded seizures with very low SNR in a first step. Furthermore we used frequency domain techniques implemented in the minimum variance beamformer allowing for a simple suppression of noise and artifacts by focusing only on frequency parts belonging to the desired ictal signal and thus inherently increasing SNR (Gritsch et al., 2011). Second, fast propagation of ictal activity in scalp EEG may cause false localization. We aimed to localize the most dominant rhythmic EEG pattern within the earliest ictal activity, i.e. the first change seen in EEG time-frequency plots, and used a frequency dependent time window. Therefore we cannot exclude modeling both activity in the seizure onset zone (SOZ) as well as early propagated activity. Nevertheless we believe that this approach is adequate due to the limited spatial resolution and the fact that only the center of gravity of the SOZ can be localized rather than detailed routes of early propagation on ictal scalp EEG. It should be noted that scalp-EEG cannot detect epileptic activity restricted to mesial temporal structures especially to the hippocampus (Wennberg and Cheyne, 2014). A recent EEG-fMRI study showed that interictal spikes visible on scalp-EEG were associated with weaker BOLD responses in temporal neocortex ipsilateral to stronger mesial temporal BOLD responses suggesting that scalp-recorded spikes result from mesial temporal spikes propagating to the temporal neocortex (Watanabe et al., 2017). Our medio-basal ictal onset source localizations therefore should be viewed as ictal activity propagated from the mesial to the basal temporal compartment rather than ictal activity restricted to the mesial temporal lobe. Third, source localization accuracy can be significantly improved with larger number of electrodes (Brodbeck et al., 2011; Lu et al., 2012; Beniczky et al., 2013). To date only four IOSL studies with relatively few patients (8–15) used more than 64 electrodes (Koessler et al., 2010; Yang et al., 2011; Lu et al., 2012; Akdeniz 2016) and only one study applied high-density EEG (256 electrodes) in 10 subjects (Holmes et al., 2010). We used a standard 10–20 electrode setup with additional true anterior temporal electrodes because the present study was designed to assess diagnostic accuracy of our IOSL method on a sublobar level and test its clinical feasibility in a standard epilepsy monitoring unit environment. Fourth, we used the Colin 27 Average Brain as basis for our head model. One large-scale study showed that localization accuracy is better if individual head models are used (Brodbeck et al., 2011). Fifth, there are several reasons for mislocalization depending on which inverse method is used. Our IOSL method for instance could not exploit the full ictal EEG information in order to perform a correct localization if weak ictal EEG rhythmicity or fast propagation of seizures was present. This dependence on rhythmicity explains the poor localization performance in patients with extratemporal epilepsy (Foldvary et al., 2001). In TLE patients, on the contrary, our IOSL tool showed good results because rhythmic ictal activity is frequently present in TLE. One advantage of our method is that it can be applied readily in a clinical setting without interactive parameter adjustments and that results can be provided and interpreted within a few minutes. So far only few studies reported on the time exposure of source localization analysis of ictal EEG data. IOSL methods were very time consuming in two studies, taking up to several hours (Boon et al., 2002; Holmes et al., 2010). Beniczky et al. reported an extra

J. Koren et al. / Clinical Neurophysiology 129 (2018) 1291–1299

time consumption of less than 30 min per patient (Beniczky et al., 2013). In conclusion, we demonstrated clinical feasibility of our ictal onset source localization approach yielding reasonable high performance measures in operated patients on a sublobar level. In particular we showed a very high NPV of our IOSL method, which may lead to further investigations in patients with discrepancies between IOSL and the planned resection site. We therefore believe that IOSL provide clinically relevant information in a standard epilepsy monitoring setting. Disclosures Johannes Koren and Johannes Herta were both partially supported by The Austrian Research Promotion Agency grant 826,816 (EpiMon). Algorithm development was conducted by the Austrian Institute of Technology including the authors Gerhard Gritsch, Hannes Perko and Tilmann Kluge. The Austrian Institute of Technology is the manufacturer of the EEG software package ‘‘Encevis”, which will include the EpiSource algorithms. 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. Acknowledgements We like to thank Sofija Kopitovic, Ingeborg Moser and Sandra Zeckl for their contribution and help during EEG data acquisition and processing. Statistical testing Johannes Koren had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Reference Akdeniz G. Electrical source localization by LORETA in patients with epilepsy: confirmation by postoperative MRI. Ann Indian Acad Neurol 2016;19:37–43. Alarcon G, Kissani N, Dad M, Elwes RD, Ekanayake J, Hennessy MJ, et al. Lateralizing and localizing values of ictal onset recorded on the scalp: evidence from simultaneous recordings with intracranial foramen ovale electrodes. Epilepsia 2001;42:1426–37. Assaf BA, Ebersole JS. Continuous source imaging of scalp ictal rhythms in temporal lobe epilepsy. Epilepsia 1997;38:1114–23. Assaf BA, Ebersole JS. Visual and quantitative ictal EEG predictors of outcome after temporal lobectomy. Epilepsia 1999;40:52–61. Bast T, Boppel T, Rupp A, Harting I, Hoechstetter K, Fauser S, et al. Noninvasive source localization of interictal EEG spikes: effects of signal-to-noise ratio and averaging. J Clin Neurophysiol 2006;23:487–97. Beniczky S, Lantz G, Rosenzweig I, Akeson P, Pedersen B, Pinborg LH, et al. Source localization of rhythmic ictal EEG activity: a study of diagnostic accuracy following STARD criteria. Epilepsia 2013;54:1743–52. Beniczky S, Oturai PS, Alving J, Sabers A, Herning M, Fabricius M. Source analysis of epileptic discharges using multiple signal classification analysis. Neuroreport 2006;17:1283–7. Blanke O, Lantz G, Seeck M, Spinelli L, Grave de Peralta R, Thut G, et al. Temporal and spatial determination of EEG-seizure onset in the frequency domain. Clin Neurophysiol 2000;111:763–72. Boon P, D’Have M. Interictal and ictal dipole modelling in patients with refractory partial epilepsy. Acta Neurol Scand 1995;92:7–18. Boon P, D’Have M, Vanrumste B, Van Hoey G, Vonck K, Van Walleghem P, et al. Ictal source localization in presurgical patients with refractory epilepsy. J Clin Neurophysiol 2002;19:461–8. Breedlove J, Nesland T, Vandergrift 3rd WA, Betting LE, Bonilha L. Probabilistic ictal EEG sources and temporal lobe epilepsy surgical outcome. Acta Neurol Scand 2014;130:103–10. Brodbeck V, Spinelli L, Lascano AM, Wissmeier M, Vargas MI, Vulliemoz S, et al. Electroencephalographic source imaging: a prospective study of 152 operated epileptic patients. Brain 2011;134:2887–97.

1299

de Tisi J, Bell GS, Peacock JL, McEvoy AW, Harkness WF, Sander JW, et al. The longterm outcome of adult epilepsy surgery, patterns of seizure remission, and relapse: a cohort study. Lancet 2011;378:1388–95. Ding L, Worrell GA, Lagerlund TD, He B. Ictal source analysis: localization and imaging of causal interactions in humans. Neuroimage 2007;34:575–86. Foldvary N, Klem G, Hammel J, Bingaman W, Najm I, Luders H. The localizing value of ictal EEG in focal epilepsy. Neurology 2001;57:2022–8. Gritsch GHM, Perko H, Fürbass F, Ossenblok P, Kluge T. Automatic detection of the seizure onset zone based on ictal EEG. Conf Proc IEEE Eng Med Biol Soc 2011:3901–4. Habib MA, Ibrahim F, Mohktar MS, Kamaruzzaman SB, Rahmat K, Lim KS. Ictal EEG source imaging for presurgical evaluation of refractory focal epilepsy. World Neurosurg 2016;88:576–85. Holmes MD, Tucker DM, Quiring JM, Hakimian S, Miller JW, Ojemann JG. Comparing noninvasive dense array and intracranial electroencephalography for localization of seizures. Neurosurgery 2010;66:354–62. Jung KY, Kang JK, Kim JH, Im CH, Kim KH, Jung HK. Spatiotemporospectral characteristics of scalp ictal EEG in mesial temporal lobe epilepsy with hippocampal sclerosis. Brain Res 2009;1287:206–19. Koessler L, Benar C, Maillard L, Badier JM, Vignal JP, Bartolomei F, et al. Source localization of ictal epileptic activity investigated by high resolution EEG and validated by SEEG. Neuroimage 2010;51:642–53. Koutroumanidis M, Martin-Miguel C, Hennessy MJ, Akanuma N, Valentin A, Alarcon G, et al. Interictal temporal delta activity in temporal lobe epilepsy: correlations with pathology and outcome. Epilepsia 2004;45:1351–67. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159–74. Lantz G, Michel CM, Seeck M, Blanke O, Landis T, Rosen I. Frequency domain EEG source localization of ictal epileptiform activity in patients with partial complex epilepsy of temporal lobe origin. Clin Neurophysiol 1999;110:176–84. Lantz G, Michel CM, Seeck M, Blanke O, Spinelli L, Thut G, et al. Space-oriented segmentation and 3-dimensional source reconstruction of ictal EEG patterns. Clin Neurophysiol 2001;112:688–97. Lee EM, Shon YM, Jung KY, Lee SA, Yum MK, Lee IK, et al. Low resolution electromagnetic tomography analysis of ictal EEG patterns in mesial temporal lobe epilepsy with hippocampal sclerosis. Clin Neurophysiol 2009;120:1797–805. Leijten FS, Huiskamp G. Interictal electromagnetic source imaging in focal epilepsy: practices, results and recommendations. Curr Opin Neurol 2008;21:437–45. Lu Y, Yang L, Worrell GA, He B. Seizure source imaging by means of FINE spatiotemporal dipole localization and directed transfer function in partial epilepsy patients. Clin Neurophysiol 2012;123:1275–83. Megevand P, Spinelli L, Genetti M, Brodbeck V, Momjian S, Schaller K, et al. Electric source imaging of interictal activity accurately localises the seizure onset zone. J Neurol Neurosurg Psychiatry 2014;85:38–43. Merlet I, Gotman J. Dipole modeling of scalp electroencephalogram epileptic discharges: correlation with intracerebral fields. Clin Neurophysiol 2001;112:414–30. Pacia SV, Ebersole JS. Intracranial EEG substrates of scalp ictal patterns from temporal lobe foci. Epilepsia 1997;38:642–54. Robinson S, Vrba J. Functional neuroimaging by synthetic aperture magnetometry (SAM). Recent advances in biomagnetism. Tohoku University Press; 1999. p. 302–5. Rosenow F, Luders H. Presurgical evaluation of epilepsy. Brain 2001;124:1683–700. Scherg M, Ille N, Weckesser D, Ebert A, Ostendorf A, Boppel T, et al. Fast evaluation of interictal spikes in long-term EEG by hyper-clustering. Epilepsia 2012;53:1196–204. Spinelli L, Andino SG, Lantz G, Seeck M, Michel CM. Electromagnetic inverse solutions in anatomically constrained spherical head models. Brain Topogr 2000;13:115–25. Stern Y, Neufeld MY, Kipervasser S, Zilberstein A, Fried I, Teicher M, et al. Source localization of temporal lobe epilepsy using PCA-LORETA analysis on ictal EEG recordings. J Clin Neurophysiol 2009;26:109–16. Valentin A, Alarcon G, Barrington SF, Garcia Seoane JJ, Martin-Miguel MC, Selway RP, et al. Interictal estimation of intracranial seizure onset in temporal lobe epilepsy. Clin Neurophysiol 2014;125:231–8. Watanabe S, Dubeau F, Zazubovits N, Gotman J. Temporal lobe spikes: EEG-fMRI contributions to the ‘‘mesial vs. lateral” debate. Clin Neurophysiol 2017;128:986–91. Wennberg R, Cheyne D. EEG source imaging of anterior temporal lobe spikes: validity and reliability. Clin Neurophysiol 2014;125:886–902. Wieser HG, Blume WT, Fish D, Goldensohn E, Hufnagel A, King D, et al. ILAE commission report. Proposal for a new classification of outcome with respect to epileptic seizures following epilepsy surgery. Epilepsia 2001;42:282–6. Worrell GA, Lagerlund TD, Sharbrough FW, Brinkmann BH, Busacker NE, Cicora KM, et al. Localization of the epileptic focus by low-resolution electromagnetic tomography in patients with a lesion demonstrated by MRI. Brain Topogr 2000;12:273–82. Yang L, Wilke C, Brinkmann B, Worrell GA, He B. Dynamic imaging of ictal oscillations using non-invasive high-resolution EEG. Neuroimage 2011;56:1908–17.