Ictal and Interictal Source Imaging on Intracranial EEG Predicts Epilepsy Surgery Outcome in Children with Focal Cortical Dysplasia

Abstract

Objective:

To localize the seizure onset zone (SOZ) and irritative zone (IZ) using electric source imaging (ESI) on intracranial EEG (iEEG) and assess their clinical value in predicting epilepsy surgery outcome in children with focal cortical dysplasia (FCD).

Methods:

We analyzed iEEG data from 25 children with FCD-associated medically refractory epilepsy (MRE) who underwent surgery. We performed ESI on ictal onset to localize SOZ (ESI-SOZ) and on interictal discharges to localize IZ (ESI-IZ). We tested whether resection of ESI-SOZ and ESI-IZ predicted good surgical outcome (Engel 1). We further compared the prediction performance of ESI-SOZ and ESI-IZ to those of SOZ and IZ defined using conventional methods, i.e. by identifying iEEG-contacts showing ictal onsets (conventional-SOZ) or being the most interictally active (conventional-IZ).

Results:

The proximity of ESI-SOZ (p=0.043, odds-ratio=3.9) and ESI-IZ (p=0.011, odds-ratio=7.04) to resection has higher effect on patients’ outcome than proximity of conventional-SOZ (p=0.17, odds-ratio=1.7) and conventional-IZ (p=0.038, odds-ratio=2.6). Resection of ESI-SOZ and ESI-IZ presented higher discriminative power in predicting outcome (68% and 60%) than conventional-SOZ and conventional-IZ (48% and 53%).

Conclusions:

Localizing SOZ and IZ via ESI on iEEG offers higher predictive value compared to conventional-iEEG interpretation.

Significance:

iEEG-ESI may help surgical planning and facilitate prognostic assessment of children with FCD-associated MRE.

1. Introduction

Focal cortical dysplasia (FCD) is a leading cause of early onset medically refractory epilepsy (MRE) in children and adolescents (Knight et al., 2015; Bast et al., 2004; Krsek et al., 2009). For these patients, the surgical resection of the epileptogenic zone (EZ), which is the brain area indispensable for generating clinical seizures, represents the best possibility to achieve seizure freedom (Chassoux et al., 2000; Krsek et al., 2009). Magnetic resonance imaging (MRI) plays an important role within this context, since it facilitates the non-invasive detection of the FCD lesion. Advances in MRI techniques have improved the identification of FCD in many focal epilepsies, which were previously classified as cryptogenic, particularly in children (Bast et al., 2004; Bast et al., 2006; Krsek et al., 2009). Yet, the exact lesional boundaries are often hard to delimit by MRI and some FCDs still evade detection in conventional neuroimaging (Duncan, 2010; Duncan et al., 1997): 3-Tesla MRI still fails to localize up to 87% of FCD 1 and 33% of FCD 2 (Pittau et al., 2017; Tassi et al, 2012). Furthermore, the EZ does not always coincide with the FCD lesion, but it can involve only part of it or extend beyond the lesional boundaries (as they are visible on MRI) to a network of remote cortical areas (Aubert et al., 2009; Ramantani et al., 2013; Maillard et al., 2009). Therefore, corroborative electroencephalography (EEG) monitoring remains crucial to estimate the EZ before epilepsy surgery, since it allows delineating the two principal estimators of the EZ: (i) the irritative zone (IZ, i.e. the brain area where interictal spikes are generated) and (ii) the seizure onset zone (SOZ, i.e. the brain area where clinical seizures start) (Lüders et al., 2006, Rosenow et al., 2001). To this purpose, intracranial EEG (iEEG) represents the modality of choice.

During iEEG monitoring, experienced epileptologists typically review the recorded iEEG data with the goal of delineating the extent of the IZ and SOZ. The IZ is traditionally defined by the location of the iEEG contacts regarded as the most interictally active (Ramantani et al., 2013; Pellegrino et al., 2016), while the SOZ is defined by the location of the iEEG contacts showing the earliest change from background activity that is associated with each clinical seizure (Tamilia and Alhilani et al. 2019; Almubarak et al., 2014; Knowlton et al., 2006). Due to the limited spatial sampling of iEEG, this conventional approach may result in misleading localization since the location of the recording contacts may differ from the actual source of the recorded activity (Ramantani et al., 2013; Ebersole and Pedley, 2003). Thus, more advanced methods are needed to overcome the limitations of the traditional iEEG interpretation and improve the SOZ and IZ delineation for epilepsy surgery.

Few studies proposed the use of electric source imaging (ESI) on iEEG data (or iEEG-ESI) to localize interictal spikes, and thus the IZ (Tamilia and Alhilani et al., 2019, Ramantani et al., 2013; Nakasato et al., 1994). This has shown to improve the surgical planning in patients with frontal lobe epilepsy (Ramantani et al., 2013), and correlate with surgical outcome in children with MRE of different etiologies (Tamila and Alhilani et al., 2019). Yet, it is still unknown whether iEEG-ESI of the IZ can also facilitate surgical planning in FCD patients, nor whether it can also allow SOZ localization in these patients. ESI of ictal onset can be extremely challenging due to the low signal to noise ratio (SNR) of ictal data and possible contamination by muscular artifacts (Nemtsas et el., 2017). Recent scalp EEG studies showed that ictal ESI helps localizing the SOZ (Holmes et al., 2010; Plummer et al., 2019; Kovac et al., 2014; Rosenzweig et al., 2014; Pellegrino et al., 2016) and guiding the resection (Nemtsas et al., 2017; Beniczky et al., 2013; Kuo et al., 2018). However, it is still unclear whether ESI can also be used interpreting ictal iEEG data.

The goal of this study is to determine the surgical utility of ESI on ictal and interictal iEEG in children with MRE associated with FCD. We hypothesize that intracranial ESI of ictal and interictal activity delineates the IZ and SOZ with a higher prognostic value for surgical outcome compared to the conventional approach for iEEG interpretation. To test our hypothesis, we analyzed interictal and ictal iEEG data from children with FCD and localized the SOZ and IZ using both ESI and conventional clinical approach. To determine the clinical value of intracranial ESI within the surgical context, we estimated the prognostic value of ESI-IZ and ESI-SOZ for surgical outcome as compared to the conventionally defined IZ and SOZ.

2. Methods

2.1. Patients and data collection

We retrospectively reviewed pediatric patients with MRE who underwent epilepsy surgery at Boston Children’s Hospital (BCH) between June 2011 and July 2017. Patients were included if they met the following inclusion criteria: (i) histological confirmation of FCD; (ii) long-term monitoring with iEEG (grids, strips, and/or depth electrodes); (iii) availability of pre-surgical MRI, post-implantation computerized tomography (CT), and post-operative MRI; and (iv) at least one year of post-surgical follow up. The study protocol received approval by the BCH Institutional Review Board (IRB-P00022114), which waived the need for written informed consent due to the study’s retrospective character. The histological confirmation of FCD was carried out by microscopic analysis of the removed tissue, as part of our institution’s clinical practice. All specimens were classified according to the new International League Against Epilepsy (ILAE) classification system for FCD (Blümcke et al., 2011).

2.2. Long-term iEEG recordings

As part of our institution’s clinical practice, iEEG long-term monitoring (LTM) is performed after consensus of the multidisciplinary epilepsy surgery conference, when there is perceived need for a better definition of the SOZ and/or the functional anatomical regions, as described in Tamilia and Alhilani (2019). The iEEG implantation was tailored for each patient independently from this study, based on the available non-invasive information. Intracranial EEG was recorded with subdural grids and strips (2.3-mm exposure diameter, 10-mm distance; Ad-Tech., USA) and/or depth electrodes (10 linearly arranged contacts: 1.1-mm diameter, 3-5 mm inter-distance; Ad-Tech., USA) using XLTEK NeuroWorks (Natus Inc., USA). For the intracranial recordings performed with subdural grid or strips, the original reference was a four-contact strip facing the dura per clinical practice at our institution. For the intracranial recordings performed with depth electrodes, the location of the original reference was in the white matter contact was used as reference electrode. The location, number and type (grids, strips, or depths) of the electrodes were prospectively decided, independently from this study, by the epilepsy surgery conference, consisted of pediatric epileptologists, neurosurgeons, neuropsychologists, neuroradiologists, and other practitioners (Tamilia and Alhilani et al., 2019). Electrode implantation is typically planned in order to monitor all possible epileptogenic areas which are regarded “of interest” based on the patient’s comprehensive presurgical evaluation, which includes ictal video-EEG, MRI and interictal MEG, when available.

2.3. MRI and CT studies

MRI was obtained before and after surgical resection with magnetization-prepared rapid acquisition gradient-echo sequences (MPRAGE; TE=1.74 ms, TR= 2, 520 ms, voxel size = 1 x 1 x 1 mm) using a high resolution 3T scanner (TIM TRIO, Siemens AG, Erlangen, Germany) via a protocol that is described in detail in previous studies (Prabhu and Mahomed, 2015). After iEEG implantation, a CT scan was performed (voxel size = 0.5 x 0.5 x 0.5 mm) to document the actual location of electrodes. For each patient, presurgical MRI and post-implantation CT were fused to identify the anatomical location of each iEEG contact using Brainstorm (Tadel et al., 2011). The coordinate of each contact was determined through visual review of the co-registered CT-MRI image and mapped on the three-dimensional (3D) model of the patient’s cortical surface reconstructed from his/her preoperative MRI using FreeSurfer (Dale et al., 1999).

2.4. Localization of IZ and SOZ via intracranial ESI

2.4.1. Selection and Analysis of Interictal and Ictal data

The iEEG marking of ictal and interictal events was done independently of the surgical planning. The invasive recordings were analyzed in bipolar and average reference montages. Both ictal and interictal activity segments were visually identified and marked by two independent readers (LR and MA), who were blind to each other’s markings. The two reviewers had not been part of the multidisciplinary team during the patients’ prospective evaluation and surgical planning and were blind to the patients’ clinical history and outcome during the marking. In case of disagreement, the questionable activity was reviewed by a more senior expert (PP) who was blinded to the patient’s outcome.

2.4.2. Forward Model

We constructed a realistic head model from each patient’s MRI using OpenMEEG software (Gramfort et al., 2010) and extracted the cortical surfaces via Freesurfer. The realistic boundary elementary model (BEM) was created using a single layer consisting of the inner skull. The model comprising 1,442 vertices, which sampled the full brain volume to also account for deep brain regions, was generated using the adaptive integration method implemented in Brainstorm.

2.4.3. Marking of Interictal Activity

As part of BCH clinical practice, the interictal iEEG data had been prospectively reviewed by the attending epileptologists who extracted several 5-10 minute epochs of data with epileptiform interictal activity on a daily basis. This had been done independently from this study and from the patient’s vigilance state. Then, we retrospectively reviewed these epochs and randomly selected one for each patient that did not include major artifacts, ictal events or technical disruptions. Segments of polyspikes (>2 spikes per 200 ms) and beta or gamma activity were excluded from visual marking, so that only single interictal epileptiform discharges (IEDs) were considered. IED marking was performed according to the criteria established by Chatrian et al. (1974). All the iEEG data were pre-processed and the IEDs were marked offline on the filtered data (band pass filter: 1-70 Hz; notch filter: 60 Hz and harmonics), according to the following criteria: (i) paroxysmal occurrence, (ii) abrupt change in polarity, and (iii) duration < 200msec. IEDs were marked at the time point of their highest amplitude in Brainstorm. Figure 1A illustrates an interictal iEEG segment with IEDs markings (blue dashed lines).

Figure 1. Marking and ESI of Interictal and Ictal iEEG.

(A) Example of IEDs recorded from patient #21 (9-year-old girl). Ten seconds of interictal data recorded by a subset of the implanted subdural electrodes are shown. Blue dashed lines indicate the IED marking, performed at the timing of the earliest of highest amplitude peaks (negative polarity is upward). (B) ESI of the IEDs (or spikes) marked in A. The implanted subdural contacts are displayed (white dots) on the pre-operative MRI of the patient. Interictal ECDs were estimated at each marked IED (blue dashed line in A), and only ECDs with GOF above 60% are shown on MRI (cyan dipoles), since ECDs with GOF <60% were excluded from further analysis. (C) Example of ictal onset activity recorded from patient #3 (10-year-old girl). Ten seconds of ictal data recorded around clinical onset (red line) by a subset of the implanted subdural and depth electrodes are shown. The highlighted window (in cobalt) shows ~4 sec of rhythmic ictal activity from channel LP53, which is the one showing the earliest or most prominent ictal change. Blue dashed lines indicate the marking of the ictal activity, performed at the timing of the negative peaks (negative polarity is upward) on this channel. (D) ESI of the ictal onset activity marked in C. The implanted subdural and depth contacts are displayed on the pre-operative MRI of the patient (white and yellow dots). Contact LP53, highlighted in C, is indicated (cobalt arrow). Ictal ECDs were estimated at each marked timing (blue dashed line in A), and only ECDs with GOF above 60% are shown on the MRI (cyan dipoles).

ECD: equivalent current dipoles; ESI: electric source imaging; GOF: goodness-of-fit; IED: interictal epileptiform discharge; iEEG: intracranial EEG; MRI: magnetic resonance imaging.

2.4.4. ESI-IZ

To localize the marked IEDs, we used the Equivalent Current Dipoles (ECD) (Hämäläinen et al., 1993), since this is the only validated method for clinical use (Stefan et al., 2003; Knowlton et al., 1997). To avoid the risk of merging IEDs generated by different sources, we localized each individual IED, rather than their average (Rikir et al., 2014, Diekmann et al., 1998). Unconstrained source reconstruction was performed at the negative peak of each IED using the dipole scanning method (Tamilia and Alhilani et al., 2019). Only interictal ECDs with goodness-of-fit (GOF) higher than 60% were considered for further analysis (Tamilia and Alhilani et al., 2019), as shown in Figure 1B.

2.4.5. Marking of Ictal Activity

For each patient, we analyzed the clinical or subclinical seizures recorded during the iEEG LTM period. All data were pre-processed (band-pass filter: 1-70 Hz; notch-filter: 60 Hz and harmonics) and marked offline. We limited the marking and following source imaging to the epochs of early ictal changes based on previously described approaches (Beniczky et al., 2013; Pellegrino et al., 2016; Nemtsas et al., 2017; Plummer et al., 2019). The electrographic start of the ictal event was regarded as the time point of the earliest rhythmic or fast activity. Ictal onsets were classified as non-localizing when contaminated by artifacts affecting most of the ictal activity. Non-localizing events were excluded from further analysis.

Epochs of early ictal changes were analyzed by: (i) selecting the time period with the earliest ictal activity; and (ii) marking the negative peaks of the ictal waveform, by looking at the channel displaying the earliest and highest oscillatory amplitude. Figure 1C illustrates an iEEG segment where the ictal onset activity was marked (blue dashed lines) based on the peaks observed in channel LP53 (blue window).

2.4.6. ESI-SOZ

To localize ictal onsets, ECDs were estimated for each ictal peak that was marked as described above. Ictal ECDs with GOF >60% were considered for further analysis, as shown in Figure 1D. When no ictal ECDs showed a GOF >60%, we averaged ictal waveforms with similar voltage distribution into a single averaged waveform, localized all time points from the averaged peak up to 75 ms before, and selected the ECD with the highest GOF above 60% (Kovac et al., 2014; Beniczky et al., 2013; Holmes et al., 2010; Plummer et al., 2019).

2.5. Clinically Defined SOZ and IZ

The clinically defined SOZ and IZ were prospectively determined during each patient’s long-term iEEG monitoring, independently from this study. Per our institution’s practice, pediatric epileptologists reviewed all the ictal EEG segments of the patient throughout the LTM period and identified the SOZ (conventional-SOZ) as the iEEG contact/s showing the earliest change in activity associated with clinical or subclinical seizures (Pellegrino et al., 2016; Gotman et al., 1993). Similarly, the clinically defined IZ (conventional-IZ) was defined by pediatric epileptologists as the iEEG contacts recording interictal spikes after review of interictal activity over different days (Bartolomei et al., 2016). For each patient, the location and extent of the conventional-SOZ and conventional-IZ were determined by the coordinates of all the SOZ and IZ contacts, respectively. Figure 2A shows the conventional-IZ (red subdural contacts) and the ESI-IZ (cyan dipoles) from an 18-year-old boy who underwent left temporal resection. Similarly, Figure 2B shows the conventional-SOZ (red deep contacts) and the ESI-SOZ (cyan dipoles) from a 10-year-old boy who underwent left parietal-occipital resection.

Figure 2. Delineation of Irritative Zone (IZ) and Seizure Onset Zone (SOZ).

(A) Delineation of the IZ for patient #7 (18-year old boy, good outcome). Top panels show the conventional-IZ defined by the coordinates of most interictally active contacts (in red) overlaid on the patient’s pre-operative MRI and the surgically resected volume (in green), drawn after co-registration with post-op MRI. Bottom panels show the ESI-IZ defined by the coordinates of the interictal ECDs (cyan dipoles). The yellow boxes show the D_res (yellow arrow) for one contact of the conventional-IZ (top) as well as for one interictal ECD of the ESI-IZ. (B) Delineation of the SOZ for patient #18 (10-year old boy, good outcome). Top panels show the conventional-SOZ defined by the coordinates of the ictal onset contacts (in red) overlaid on the patient’s pre-operative MRI and resected volume (in green). Bottom panels show the ESI-SOZ defined by the coordinates of the ictal ECDs (cyan dipoles).

ECD: equivalent current dipoles; ESI: electric source imaging; D_res: resection distance; MRI: magnetic resonance imaging.

2.6. Concordance between MRI lesion and ESI solutions

To evaluate the spatial relationship between the MRI-defined lesion and the IZ-ESI or SOZ-IZ, we determined the anatomical location of every localized dipole. Each dipole was assigned to one of the following anatomical regions based on the individual Freesurfer cortical parcellation (using the ‘Desikan-Killiany’ atlas): (i) frontal; (ii) frontal pole; (iii) central, (iv) parietal; (v) temporal; (vi) occipital; and (vii) cingulate. For each patient, the anatomical region where most of the dipoles were localized determined the predominant region/s of the ESI solution. Similarly, the MRI abnormality was assigned to its anatomical region/s based on the MRI clinical evaluation performed at the time of the patient’s pre-surgical evaluation, independently from this study. Finally, we determined the concordance between the ESI solutions, for IZ and SOZ separately, and the location of the MRI lesion (Tamilia et al. 2019) as follows: (i) concordant specific (>50% of the ESI dipoles within the MRI-defined lobe/s); (ii) concordant non-specific (between 10% and 50% of the ESI dipoles within the MRI-defined lobe/s); and (iii) discordant (less than 10% of the ESI dipoles within the MRI-defined lobe/s).

2.7. Resection and Postoperative Follow-Up

Cortical resections were individually tailored according to anatomical and neurophysiological data, including the clinically defined EZ (SOZ contacts primarily, then IZ contacts). The ESI analysis was performed retrospectively; thus, the ESI findings did not influence the surgical planning. The corticectomy specimens were then processed for histology and immunohistochemistry examinations. To define the resection within the patient’s brain volume, we coregistered the preoperative and postoperative MRI using Brainstorm (Tadel et al., 2011). We defined the resected volume by marking all the volume points corresponding to the resection cavity on the co-registered image. The resected volume on MRI was defined by a member of the research team (ET) who was blind to the marking and ESI results as well as to the patient’s outcome. Figure 2 shows the resected volume (green volume) overlaid on the pre-operative MRI for two patients. The post-surgical clinical outcome was evaluated using the Engel classification system (Engel et al., 1993) based on the patient’s most-recent follow-up at least 12 months after surgery, as part of the clinical practice at our institution. Outcome was dichotomized into good (Engel 1) and poor (Engel ≥2).

2.8. Prognostic Value for Surgical Outcome

The distance of each interictal and ictal ECD from the resection was calculated as its minimum Euclidean distance from the closest resected volume point (D_res). Similarly, D_res of each iEEG contact belonging to the conventional-IZ and conventional-SOZ was calculated. Figure 2A shows an example of D_res (yellow arrow) for one contact of the conventional-IZ and one ECD of the ESI-IZ. For each zone (ESI-SOZ, ESI-IZ, conventional-SOZ and conventional-IZ), we estimated the resection percentage (i.e. proportion of ECDs or iEEG contacts with D_res <15 mm).

2.9. Statistical Analysis

We used Wilcoxon rank-sum test to compare continuous variables between good and poor outcome group. Nonparametric statistics was used since the variables did not follow standard normal distribution as established by Kolmogorov-Smirnov test. Categorical variables were compared between good and poor outcome group using the Fisher’s exact test. Binary logistic analysis was performed to test whether the resection distance (D_res) (mean distance per patient) was predictive of patient’s post-surgical outcome (good versus poor). Regression was performed for ESI-IZ, conventional-IZ, ESI-SOZ and conventional-SOZ.

Finally, we built receiver operating characteristic (ROC) curves on the resection percentage of each zone to assess whether removing ESI-SOZ, ESI-IZ, conventional-SOZ or conventional-IZ was able to predict the individual patient’s outcome. We considered good outcome following complete resection to be the ground truth, i.e. unambiguous proof of correct localization of the EZ. We regarded as: (i) true positives (TP), good-outcome patients with high resection percentage (i.e. complete resection); (ii) true negatives (TN), poor-outcome patients with low resection percentage (i.e. incomplete resection); (iii) false positives (FP), poor-outcome patients with complete resection; and (iv) false negatives (FN), good-outcome patients with incomplete resection. The optimum cut-off value that differentiated high versus low resection percentage (complete versus incomplete resection) was selected based on the ROC curve point showing the highest value of Youden’s index (J), defined as TP/(TP+FN) + TN/(TN+FP) − 1, which estimates the overall discriminative power in predicting outcome. Furthermore, to assess the performance of each zone in terms of outcome prediction, we also calculated positive predictive value [PPV=TP/(TP+FP)] and negative predictive value [NPV=TN/(TN+FN)]. For each ESI and clinical zone, Fisher’s exact test was used to assess whether there was association between resection (above optimal threshold) and seizure-outcome.

P-values <0.05 were considered significant. Results are expressed as mean ± standard deviation. MATLAB 2018a (The MathWorks, Inc) was used for statistical analysis.

3. Results

3.1. Patient cohort and resection

Eighty-four pediatric patients with FCD-associated MRE underwent epilepsy surgery in the selected period. Of these 84 patients, 59 did not meet our inclusion criteria (see Supplementary Figure S1) due to: (i) no iEEG long-term monitoring (n=35); (ii) no post-implantation CT (n=7); (iii) no post-operative MRI (n=8); and (iv) no one-year post-surgical follow up (n=9). Hence, 25 patients were included (13 females, mean age 12.8; range: 5-18 years). Six patients revealed histopathological hallmarks of FCD type I; 13 patients revealed FCD type II and three FCD type III. Three patients’ specimens exhibited dysplastic features and reactive changes indicating FCD, however more precise classification was not possible. The mean follow-up duration was 37 months (range 12-91 months). Only 33% of FCD I patients were seizure free postoperatively (2 of 6), compared to 77% of patients with FCD type II (10 of 13).

The mean resection volume was 22.5 ± 18.5 cm³ without difference between good and poor outcome (21.5 cm³ vs. 24 cm³; p = 0.85). In six cases, the hypothesized EZ overlapped with the eloquent cortex and/or vascular structures, thus was not completely resected during surgery. Such overlap with eloquent cortex and/or vascular structures was not associated with the outcome (p=0.65). Table 1 reports each patient demographics and presurgical clinical localization, as well as information about iEEG placement, resected area, postsurgical outcome and follow up period. Fifteen patients were seizure free (Engel 1).

Table 1:

Clinical characteristics of our cohort, including demographics, results of the non-invasive presurgical workup, and invasive recordings with post-surgical outcomes.

			Presurgical evaluation findings					Surgery and outcomes
	Age (y)	Sex	18FDG-PET hypometabolism areas	Ictal SPECT hyperperfusion areas)	Video EEG	#iEEG electrodes	Resected lobe(s)	Resection Volume (cm3)	Resection of intracranially defined EZ	Histology	Engel	Post-op follow up (months)
1	11	M	not done	L iT, aT	LF, LC	88S+10 D	L mF	8.24	Complete	FCD2a	1a	47
2	7	F	not done	LT	L F-C-T	90D	L sT	13.69	Complete	FCD2a	1a	26
3	14	F	LT	Interictal: LH	LH, LC	140D	LT	8.48	Complete	FCD (nc)	3	31
4	14	F	LT	L mT	LT	72S	LT	18.66	Partial	FCD2a	1a	48
5	16	F	LT	NL	LT	96S+8D	LT	23.08	Partial	FCD3c	1a	53
6	9	F	RP	RP	R C-P	80S+20 D	R C-P	18.95	Partial	FCD1	2a	70
7	18	M	L mT, iT	L mT, iT	L aT	64S	L aT	32.39	Complete	FCD3a	1a	26
8	7	M	LT-P	NL	LH	120S	LF	22.75	Partial	FCD2a	4	91
9	16	M	L mT	RT, LT	LT	88S	LT	10.9	Complete	FCD1a	2b	25
10	10	F	LT	LT	F	112S+10 D	R mF	25.21	Partial	FCD2b	1b	42
11	18	F	NL	LF	NL	144S+10 D	LF	7.38	Complete	FCD2b	2a	32
12	17	F	L mT	L aT, R aT	L C-P	72S	L P	8.18	Complete	FCD (nc)	1b	26
13	13	M	not done	NL	LT	136S	L F-T	47.26	Complete	FCD3d	2	48
14	18	M	LH	L mT	L F-C	64S+30 D	L F	21.86	Complete	FCD2b	1a	28
15	12	F	L F	LF, LT	L F-P	70D	LF	29.58	Complete	FCD2a	1a	29
16	12	M	LT	LT	L aT	96S	L T	4.55	Complete	FCD2a	1a	62
17	18	M	L iF, pF	L sT, mT	LT	238D	L mT	30.08	Complete	FCD1	3	25
18	10	M	L F-T	not done	L C-P	64S+60	L P-O	59.97	Complete	FCD1	1ba	12
19	10	F	L F-P	LF-P	LH	96S	L F-P	10.81	Partial	FCD (nc)	3	53
20	7	F	not done	RF	RH	72S+40 D	RF	8.91	Complete	FCD2b	2a	51
21	5	M	RF	RF pole	R F-C-P	128S+10 D	RF	74.89	Complete	FCD2b	1a	24
22	18	M	L LG	interictal: NL	L F-T	162D	LO	52.23	Complete	FCD2b	1a	33
23	8	F	L insula, sT	Interictal: LT)	L F-T	164D	LF	12.42	Complete	FCD1	2	24
24	15	M	R F-P	R aP	RC	102D	RP	7.64	Complete	FCD1	1aa	13
25	17	F	LF	LF	LF	72S+40 D	LF	5.42	Complete	FCD2b	1aa	16

The age (y) provided is at the time of the invasive long term recording.

M: male, F: female, F=Frontal, C=Central, L= Left, T=temporal, H=hemisphere, R=right, NL= non localizing or not lateralizing, SMA= supplementary motor cortex, P=parietal, N= normal, m=medial, i=inferior, p=Posterior, LG=lingual gyrus, s=superior, a=anterior, l=lateral, nc= FCD histologically confirmed but not further classifiable, S = subdural iEEG electrodes (grids and/or strips), D= deep electrodes

^aEngel class in this cases was established at the time of the most recent follow which occurred between one and two years after surgery

3.2. Interictal and Ictal activity

A total of 5,520 IEDs were marked across all 25 patients (mean = 221 per patient; rate = 37.1 ± 33.7 IEDs/min). A total of 1,318 ECDs showed GOF >60% and thus were further analyzed (mean: 52.7 ± 70.8 ECDs/patient). IED rates and number of ECDs did not differ between good and poor outcome patients (p=0.93; p=0.56). For seven patients (patients #1, 3, 4, 5, 17, 21 and 23), ictal data from LTM monitoring were not available for retrieval from BCH archive at the time of the study; thus, only the remaining 18 patients (11 good-outcome) were included in SOZ analyses. A total of 179 seizures were recorded; 11 seizures were excluded because ictal onsets were regarded as non-localizing due to presence of artifacts; thus, a total of 168 seizures were localized (mean = 9.3 ± 8.7 seizures/patient), without difference between good and poor outcome patients (p=0.2). The average time interval between the first clinical sign of the ictal event to the marked iEEG ictal rhythms was 8.2 s (ranging from 8.5 s before and 36 s after the first sign). An average of 3 ± 3 iEEG contacts per patient were excluded from the analysis due to the presence of artifacts. The number of the interictally active iEEG contacts belonging to the conventional-IZ was 22 ± 13, and the number of iEEG contacts showing ictal onset (i.e. part of the conventional-SOZ) was 15 ± 10 for each patient. The spatial extent of conventional-IZ and conventional-SOZ did not differ between good and poor outcome patients (p=0.14, p=0.8).

3.3. Spatial Concordance of ESI with MRI lesion

Two patients did not show any MRI-visible lesion at the time of the pre-surgical evaluation (see Table 2). Of all IZ dipoles from patients with lesional MRI, 52% (n=663) were localized within the anatomical location of the MRI lesion, with a mean percentage per patient of 48 ± 27%. The ESI-IZ was concordant specific with the MRI lesion region/s in 12 out of the 23 (52%) patients with MRI-visible lesions, concordant non-specific in eight (35%) and discordant in three (13%). For the ESI-SOZ, 70% (n=204) of all dipoles were localized within the anatomical location of the MRI lesion, with a mean percentage per patient of 55 ± 41%. The ESI-SOZ was concordant specific with the MRI lesion region/s in eight out of the 16 (50%) patients, concordant non-specific in four (25%) and discordant in four (25%). Table 2 reports the region/s of the MRI lesion, and the concordance of the ESI dipoles for each patient.

Table 2:

MRI lesion and ESI solutions.

	MRI-lesion Region/s	IZ			SOZ
		Main ESI region/s	% of dipoles in MRI-lesion region/s	Concor dance	Main ESI region/s	% of dipoles in MRI-lesion region/s	Concor dance
1	L F	LF RF	35%	C-ns	n/a	n/a	n/a
2	L T	L T	65%	C-s	L T	90%	C-s
3	L T	L T	51%	C-s	L T	90%	C-s
4	L T	L T	77%	C-s	n/a	n/a	n/a
5	L T	L T	52%	C-s	n/a	n/a	n/a
6	L P	R P/T LC	31%	C-ns	n/a	n/a	n/a
7	L T	LF	38%	D	n/a	n/a	n/a
8	L F	L T/C	25%	D	LO	0%	D
9	Negative	L T	n/a	n/a	L T/F/PF	n/a	n/a
10	R F/C	R C/F	63%	C-s	R C/P	50%	D
11	L F	L C	0%	D	L F	100%	C-s
12	L P	L P	71%	C-s	L T/C	0%	D
13	L T,F,P,C (MCA)	L F	64%	C-s	n/a	n/a	n/a
14	L C/P	L P/C	79%	C-s	L F/P/C	33%	C-ns
15	L F	L F	75%	C-s	L F	50%	C-s
16	L T	L T	64%	C-s	L T	100%	C-s
17	Negative	L F/T	n/a	n/a	L PF	n/a	n/a
18	L F/P	L T/C	32%	D	L T	13%	D
19	L C	L C/F	44%	C-ns	n/a	n/a	n/a
20	R F/Ins	R PF/Ins	43%	C-ns	R T	0%	D
21	R PF/F	R F	100%	C-s R	PF	96%	C-s
22	L O	L T/O	33%	C-ns	L O	73%	C-s
22	L O	L T/O	33%	C-ns	L O	73%	C-s
23	L F	L T/P	0%	D	L T	5%	D
24	R P	R C	0%	D	R P	84%	C-s
25	L F	L F	60%	C-s	L F	100%	C-s

F = Frontal, L = Left, C = Central, T = temporal, R = right, P = parietal, O = occipital; PF = Frontal pole; MCA = middle cerebral artery; Ins = Insula; C-s = concordant specific; C-ns = concordant non-specific; D = discordant

3.4. Distance from resection and Outcome

D_res in patients with good outcome was shorter than in patients with poor outcome for the ESI-IZ (p=0.003), ESI-SOZ (p=0.006), and conventional-IZ (p=0.01). For the IZ, regression analysis showed that both the ESI solution (p=0.011) and the conventional approach (p=0.038) were able to predict the patient’s outcome based on their distance from the resection, with the ESI presenting a higher effect (OR: 7.04; CI: 1.56-31.74) than the conventional approach (OR: 2.61; CI: 1.05-6.47). For the SOZ, regression analysis showed that only the ESI solution (p=0.043) was able to predict the patient’s outcome based on its distance from the resection, with an OR of 3.9 (CI: 1.04-14.4); while the conventional approach did not reach significance (p=0.17; OR: 1.7; CI: 0.8-3.9). Table 3 reports the regression analysis results.

Table 3:

Binary Logistic Regression on the distance from resection of SOZ and IZ (defined via ESI or conventional approach).

Zone	Approach	B	ORa	95% CI	p-value
SOZ	ESI	1.35	3.87	1.04-14.37	0.043*
	Conventional	0.56	1.75	0.78-3.9	0.17
IZ	ESI	1.95	7.04	1.56-31.74	0.011*
	Conventional	0.96	2.61	1.05-6.47	0.038*

^aOdds-Ratio (OR): odds of poor outcome over good outcome for every 1-cm increase in distance from resection

3.5. Predictive value of ictal and interictal activity

ROC curve analysis showed that the resection of the ESI-SOZ presented a higher discriminative power in predicting the patient’s outcome than the resection of the conventional-SOZ: (i) for ESI-SOZ, we found a PPV of 90%, NPV of 75%, and an overall diagnostic performance (J) of 68% (p=0.013, Figure 3A); (ii) for conventional-SOZ, we found a PPV of 77%, NPV of 80%, and J of 48% (p=0.047, Figure 3B). The optimal cut-off to define complete versus incomplete resection of the SOZ was 78% for ESI and 70% for conventional-SOZ (Table 4). Complete resection of the ESI-SOZ was found in nine out of 11 (88%) good-outcome patients and one out of seven (14%) poor-outcome patients. For conventional-SOZ, complete resection was found in 10 out of 11 (91%) good-outcome patients and three out of seven (43%) poor-outcome patients.

Figure 3. ROC curves of the Outcome Prediction and Performance Measures.

ROC curve results are shown for the resection of ESI-SOZ (A), conventional-ESI (B), ESI-IZ (C) and conventional-IZ (D). Each scenario shows the ROC curve on the left and the estimated PPV (blue bar), NPV (purple bar) and J (green bar) on the right. Dashed red line marks the Youden’s Index (J), which is graphically represented as the height above the chance level (black line). Diagonal segments of the ROC curve are produced by ties between the actual positives and the actual negatives state (trapezoidal rule was applied).

ESI: electric source imaging; IZ: irritative zone; ROC: receiver operating characteristic; SOZ: seizure onset zone.

Table 4:

Resection of SOZ and IZ (defined via ESI or conventional approach) and their Prognostic Value for Surgical Outcome (good versus poor).

Zone	Approach	Resection Cut-off a [%]	Surgically Resectedb	Patients [#]			PPV	NPV	J	P-valuec
				Tot	Poor	Good
SOZ	ESI	78%	No Yes	8 10	6 1	2 9	90%	75%	68%	0.0128
	Conventional	70%	No Yes	5 13	4 3	1 10	77%	80%	48%	0.0474
IZ	ESI	90%	No Yes	16 9	10 0	6 9	100%	63%	60%	0.0028
	Conventional	47%	No Yes	7 18	6 4	1 14	78%	86%	53%	0.0068

^aCut-off value on the resection percentage was defined by the ROC curve point presenting the highest Youden’s Index (J)

^bSurgically Resected = Resection Percentage above cut-off

^cFisher’s exact test

For interictal data, the resection of the ESI-IZ showed a higher discriminative power in predicting outcome than the resection of the conventional-IZ: (i) for ESI-IZ, we estimated a PPV of 100%, NPV of 63%, and J of 60% (p=0.003, Figure 3C); (ii) for conventional-IZ, we had a PPV of 78%, NPV of 63%, and J of 53% (p=0.007, Figure 3D). The optimal cut-off on the IZ resection percentage to predict outcome was 90% for ESI and 47% for conventional-IZ (Table 4). Complete resection of the ESI-IZ was found in nine out of 15 (60%) good-outcome patients and in none of the 10 poor-outcome patients. For conventional-IZ, complete resection was found in 14 out of 15 (93%) good-outcome patients and four out of 10 (40%) poor-outcome patients.

4. Discussion

In this study, we investigated the clinical value of interictal and ictal ESI applied to iEEG (or intracranial ESI) in children with FCD-associated MRE. We showed that localization of the IZ and SOZ via intracranial ESI has clinically relevant potential since it provides a better estimation of the EZ compared to the conventional approaches that are typically used in clinical practice to interpret iEEG. To our best knowledge, this study is the first to perform ESI on both interictal and ictal iEEG and to assess the clinical utility of intracranial ESI within the context of epilepsy surgery. Our main findings can be summarized as follows: (i) the proximity of ESI-SOZ and ESI-IZ to the resection presented a higher effect on the patient’s outcome than the proximity of conventional-SOZ and conventional-IZ; (ii) resection of the ESI-IZ predicted the patient’s surgical outcome with higher discriminative power (Youden’s Index or J=60%) than the conventional-IZ (J=53%); and (iii) resection of the ESI-SOZ presented higher discriminative power (J=68%) in terms of outcome prediction compared to the conventional SOZ (J=48%). These findings suggest that intracranial ESI improves the 3D demarcation of the SOZ and IZ by facilitating the localization of the actual epileptic generators, which underlie the interictal and ictal electrographic events observed at the contact level.

Our work adds to the few previous studies, which investigated the potential of intracranial ESI (Ramantani et al., 2013; Kim et al., 2010) by showing that interictal and ictal ESI offer augmented value in assessing individualized patient prognosis compared to conventional methods used to interpret iEEG and to delineate the IZ or SOZ. Our data indicated that the proximity of the ESI-IZ to the resection have a significant effect on the patient’s outcome with the ESI presenting a higher effect than the conventional approach: for the ESI-IZ, the odds of poor postsurgical outcome were seven times higher than the odds of good outcome for every 1-cm increase in the resection distance; while for the conventionally-defined IZ, the odds of poor outcome were 2.6 times higher than the odds of good outcome. For the SOZ, our data indicate that only the proximity of the ESI-SOZ, but not the conventional-SOZ, to the resection has a significant effect on the patient’s outcome. Of note, the lower sample size for the SOZ analysis, compared to the IZ analysis, limits the statistical power and the ability to detect small effect size. Furthermore, the ROC curve analysis on the percentage of resection of the different zones further corroborated the above mentioned results, by suggesting that the conventional approach is weaker in terms of prognostic power compared to the ESI approach: resection of the ESI-SOZ and ESI-IZ offered a discriminative power of 60% and 68% in predicting the patients’ outcome, which exceeded the prognostic performance of the clinically-defined zones in the same patients (53% and 48%). In particular, the resection of the ESI solutions seems to offer an extremely high PPV (90-100%) for post-surgical outcome. Surgical removal of the ESI-IZ never led to poor outcome in our cohort and was thus associated with a 100% chance of success, as opposed to removal of the conventional-IZ, which led to a poor outcome in 4 out of 14 (29%) patients. For the SOZ, resection of the ESI defined area led to poor outcome in one single case out of 10 (10%), thus offering a 90% chance of good outcome and outperforming the removal of the conventional SOZ, which was associated with a 77% of possibility of positive outcome. The 23-29% of chance of poor outcome, despite the removal of the conventional SOZ or IZ, points out the possibility that such zones may overlook actual epileptogenic foci. This can be possibly due to the lack of a strict spatial correspondence between the location of the recording iEEG contact and the real location of the epileptogenic focus, such as in the case of a contact lying on a gyral crown that records the activation of a source located in the depth of the sulcus (Ramantani et al., 2013). In addition, we observed a very high specificity of the ESI solutions (see ROC in Fig. 3): the ESI-SOZ and ESI-IZ was missed during resection in 86% and 100% of the poor-outcome cases (as opposed to 57% and 60% for the corresponding conventional zones). In these cases, including the ESI solutions as possible epileptogenic foci may have expanded the zones to consider for resection; yet, we cannot determine whether this would have changed surgical plan or postsurgical outcomes, given the study’s retrospective nature.

The lower performance of the conventionally-defined areas compared to the ESI solutions may be explained by the fact that the location of the subdural or deep contact recording a certain electrographic event may differ from the location of the actual underlying source responsible for its generation (Ramantani et al., 2013). To this regard, it is crucial to consider that the conventional method fully relies on the visual assessment to determine the location of the ictal onset or interictal generator (which is set to be equal to the location of the channel where the activity of interest is seen). In a similar way, the ESI procedure also depends on a visual icEEG assessment, since this is necessary to identify the timing of the electrographic event of interest. However, ESI relies on such visual marking solely to set the timing of interest, but then it locates the underlying generator (by solving the inverse problem) based on the data recorded from all channels at that time (regardless of the channel where the visual marking is performed). In this way, ESI determines the location of the ictal onset or interictal generator, which does not necessarily correspond to the coordinates of the contact where the human operator identified the event (although possibly biased by it).

Finally, the fact that we used less than 10 minutes of interictal iEEG data to localize the ESI-IZ represents an important aspect from a clinical practice perspective. A high workload is typically needed to identify the IZ during the invasive LTM since reviewing the interictal data over many days of recordings can be very time-consuming (Beniczky et al., 2013). Here, we showed that the predictive value gained by performing ESI on a randomly selected segment (~5 min long) of interictal data (ESI-IZ) outperformed the daily observation of interictal recordings during the LTM period (conventional IZ). This indicates that performing ESI on 5-10 min data can significantly impact the clinical practice by making the need of constantly reading interictal LTM recordings superfluous. This may potentially hold great benefit to decrease the complications that invasive LTM could cause, since longer periods of invasive recordings are one of the greatest risk factors (Harkness et al., 2010; Hamer et al., 2002).

Our study has some limitations which should be stated. The first limitation being the non-randomized, retrospective nature of the study design: although we could establish the statistical relationship between ESI results and seizure outcome in retrospection, we were not able to directly evaluate the effect that ESI had on resection planning, and thus, on seizure outcome. Also, the smaller sample size from the SOZ analysis played a role in restricting further direct comparisons between the SOZ and IZ localization methods. Furthermore, the single ECD model does not account for the possibility of multiple underlying generators overlapping in time (Iwasaki et al., 2002) and may be an abstract representation of an extended cortical area. In addition, the propagation of ictal onsets and IEDs from a primary epileptogenic generator toward less epileptogenic areas, may lead to a misallocation of the epileptogenic generator (Alarcon et al., 1997, Tamilia et al., 2018; Jiang et al., 2010). Thus, further studies are needed that will make use of multiple ECDs or distributed sources and further investigate IED and ictal propagation via iEEGESI. Finally, given the heterogeneity of our cohort in terms of histopathology of the FCD and the relatively small group of patients within each subtype group, our analysis did not discriminate between sporadic spikes or sharp waves and rhythmic epileptiform discharges (whose localizing value is largely established in FCD type II). Future studies on homogeneous cohorts are needed in order to investigate the specific benefit of ESI in FCD type II by discriminating between rhythmic epileptiform discharges and sporadic spikes and sharp wave.

5. Conclusion

This study shows that ESI applied to interictal and ictal iEEG data facilitates the estimation of the epileptogenic zone in children with FCD undergoing epilepsy surgery. Localization of the IZ and SOZ via intracranial ESI offers higher prognostic value compared to conventional strategies that are used for iEEG interpretation in current clinical practice. The use of ESI to interpret iEEG can potentially facilitate surgical planning and help predict the outcome of epilepsy surgery in patients with FCD-associated MRE undergoing iEEG monitoring.

Highlights.

• Irritative zone and seizure onset zone can be precisely localized with iEEG source imaging.

• Proximity of electric source imaging (ESI) solutions to resection predicts outcome better than traditional iEEG interpretation.

• iEEG ESI of irritative zone and seizure onset zone help outcome prediction in pediatric epilepsy surgery.