Acta Neurol Scand 2014: 130: 103–110 DOI: 10.1111/ane.12253
© 2014 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd ACTA NEUROLOGICA SCANDINAVICA
Probabilistic ictal EEG sources and temporal lobe epilepsy surgical outcome Breedlove J, Nesland T, Vandergrift WA 3rd, Betting LE, Bonilha L. Probabilistic ictal EEG sources and temporal lobe epilepsy surgical outcome. Acta Neurol Scand 2014: 130: 103–110. © 2014 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd. Objective – For patients with medication refractory medial temporal lobe epilepsy (MTLE), surgery offers the hope of a cure. However, up to 30% of patients with MTLE continue to experience disabling seizures after surgery. The reasons why some patients do not achieve seizure freedom are poorly understood. A promising theory suggests that epileptogenic networks are broadly distributed in surgically refractory MTLE, involving regions beyond the medial temporal lobe. In this retrospective study, we aimed to investigate the distribution of epileptogenic networks in MTLE using Bayesian distributed EEG source analysis from preoperative ictal onset recordings. This analysis has the advantage of generating maps of source probability, which can be subjected to voxel-based statistical analyses. Methods – We compared 10 patients who achieved post-surgical seizure freedom with 10 patients who continued experiencing seizures after surgery. Voxelbased Wilcoxon tests were employed with correction for multiple comparisons. Results – We observed that ictal EEG source intensities were significantly more likely to occur in lateral temporal and posterior medial temporal regions in patients with continued seizures post-surgery. Conclusions – Our findings support the theory of broader spatial distribution of epileptogenic networks at seizure onset in patients with surgically refractory MTLE.
Introduction
Anterior temporal lobectomy (ATL) is the most effective treatment for medial temporal lobe epilepsy (MTLE) in patients whose seizure control was not achieved with antiepileptic drugs (AEDs) (1, 2). Up to 75% (2) of patients with MTLE may become completely free of seizures after surgery. Unfortunately, 25–30% of patients who undergo surgical treatment continue to experience disabling seizures, in spite of a thorough presurgical evaluation. These patients often appear clinically identical to those who can become seizure free, with compatible clinical semiology, imaging, and ictal scalp electroencephalography (EEG) profiles. This discrepancy in outcome suggests that there may be subtle, yet clinically meaningful, differences in the disease mechanisms that are not identified by current clinical assessment tools.
J. Breedlove1, T. Nesland1, W. A. Vandergrift 3rd1, L. E. Betting2, L. Bonilha1 1 Department of Neurology, Comprehensive Epilepsy Center, Medical University of South Carolina, Charleston, SC, USA; 2Departamento de Neurologia, Psiquiatria e Psicologia, Faculdade de Medicina de Botucatu – UNESP, Botucatu, SP, Brazil
Key words: electroencephalography; source localization; temporal lobe epilepsy; epilepsy surgery Leonardo Bonilha MD PhD, Comprehensive Epilepsy Center, Division of Neurology, Medical University of South Carolina Tel: 843 792 3383 Fax: 842 792 8626 e-mail: [email protected] Accepted for publication March 14, 2014
A promising theory suggests that the location and distribution of the epileptogenic zone may be different in patients with surgical refractoriness. Patients who do not respond to surgery may exhibit a broader distribution of the epileptogenic networks (3). These networks may extend beyond the area of resection, and the post-surgical persistence of abnormal networks may continue to generate seizures after surgery (4, 5). This study sought to investigate the distribution of epileptogenic networks in preoperative MTLE patients using electroencephalographic source imaging (ESI) on ictal EEG recordings. Specifically, we applied a unique Bayesian approach to evaluate the probabilistic distribution of ictal EEG sources of presurgical seizures in a cohort of patients with MTLE who underwent surgical treatment. We hypothesized that the probabilistic EEG ictal sources would exhibit an anatomically broader 103
Breedlove et al. distribution in patients with poor ATL outcomes, when compared with those with favorable ATL outcomes. Methods Participants
We retrospectively studied twenty patients followed at the Medical University of South Carolina who were diagnosed with medication refractory MTLE according to the parameters defined by the International League Against Epilepsy – ILAE (6). All patients underwent ATL performed by the same neurosurgeon to treat epilepsy after a comprehensive presurgical investigation, which included a detailed neurological examination, high-resolution magnetic resonance imaging (MRI), and long-term scalp video EEG (VEEG). Additionally, some patients underwent complimentary studies such as single-photon emission computed tomography (SPECT) and intracranial EEG recording to ensure the lateralization of their seizures. All findings from the presurgical evaluations were discussed during a multidisciplinary epilepsy conference, and decisions regarding surgical treatment were achieved through consensus. Patients with VEEG recordings demonstrating bilateral seizure onset or MRI findings indicating large abnormalities (such as sizeable infarcts or mass lesions) were excluded from the study.
for all patients in the post-operative period. Outcomes were classified according to the Engel Surgical Outcome Scale, grouped into two main groups: (i) free of disabling seizures or seizure free (SF), equivalent to Engel Class I; or (ii) continued seizures (CS), equivalent to Engel Classes II, III, or IV. Ictal EEG selection and analysis
All patients included in this study demonstrated similar patterns of ictal EEG across their seizures. For each patient having multiple seizure recordings available, we selected an ictal EEG event that best demonstrated the patient’s typical seizure as indicated by chart history and VEEG reports. For example, if the patient’s history indicated that his or her seizures typically occurred during sleep, a seizure that arose out of sleep was chosen for source localization. The event was also selected on the basis that it was the one mostly free from artifact. Ictal onset was defined as the first visual rhythmic electrographic change from background pattern and was marked by the patient’s epileptologist prior to epilepsy surgery. We analyzed 30-s epochs that were manually extracted from these EEGs, 10 s to +20 s with respect to seizure onset. The frequency of scalp rhythmic discharges at ictal onset was determined through visual inspection for each electrographic seizure. Distributed source analysis
EEG recording
All patients were admitted to the epilepsy-monitoring unit for long-term scalp VEEG prior to surgery. EEGs were recorded to a vertex reference during acquisition using 21 electrodes. The array selected included 19 of the standard 10–20 system positions (excluding ear references, A1 and A2), with the addition of FT9 and FT10 to record anterior portions of the temporal lobe. Patients were monitored using a XLTEK acquisition station and an EMU40 head box (XLTEK, Natus, Ontario, Canada). Data were acquired at a sampling rate of 256 Hz, a band pass of 1–70 Hz, and all electrodes were placed with impedances below 5 kOhms in accordance with ACNS guidelines (7). Surgical outcome
Surgical outcome was defined with at least 1 year of post-surgical follow-up. None of the patients had intra- or perioperative complications, and the preoperative antiepileptic regimen was continued 104
To reconstruct ictal activity into 3D voxel space, all ictal epochs were submitted to source analysis using the distributed imaging method provided by SPM8 (http://www.fil.ion.ucl.ac.uk/spm/software/ spm8/), in accordance with the steps outlined below. This workflow is illustrated in Fig. 1. Preprocessing, coregistration, and forward modeling
Each raw epoched data set was run through a 1–35 Hz band pass filter to reduce artifact from 60 Hz noise, movement, and muscle contraction. Sensor locations were loaded in a MATLAB file containing the MNI coordinates of each sensor and coregistered using a file containing the locations of 3 fiducials (nasion, right preauricular, and left preauricular). A realistic boundary element method (BEM) was used for forward modeling. Inversion
For each ictal epoch, a custom greedy search (GS) was implemented using the entire 30-s time
Ictal EEG source and surgical outcome
Figure 1. Workflow of distributed EEG source analysis. The raw EEG data (including the electrographic onset plus 5 s) are submitted to preprocessing. The electrodes are coregistered into standard space allowing for the calculation of scalp field maps for each millisecond. The forward model is created using the realistic boundary element method (BEM) and inversed using a window created around the ictal frequency. Resulting is a 3D image displaying the probabilistic source intensities. The example shown here is a left-onset temporal lobe seizure.
window, low-frequency filer (LFF) of 1 Hz, and high-frequency filter (HFF) of 48 Hz. Solutions were restricted to the temporal lobes using MNI coordinates corresponding to the areas of interest. This restriction simplified the model and was based on prior knowledge that seizures were originating from within the temporal lobes, as per the combined presurgical data including semiology, VEEG, and/or MRI for each patient. For each ictal event, the resulting source amplitudes were estimated and written as a single 3D NIfTI (Neuroimaging Informatics Technology Initiative) image. This was accomplished using a time window set at ictal onset plus 5 s. To focus on seizure activity and avoid unintentional analysis of background patterns, images were created using a 3 Hz frequency window surrounding the patient’s specific ictal frequency at onset (frequency of rhythmic discharges 1.5 Hz).
was 39 10 years. There were no significant differences in age (T = 1.38, P = 0.19) or gender distribution (v2 = 3.53, P = 0.06) between the CS and SF groups. Patients in both groups were found to have similar seizure frequencies (P = 0.71), age of onset of epilepsy (P = 0.65), and duration of epilepsy (P = 0.59). Thirteen patients exhibited medial temporal sclerosis (MTS) during routine diagnostic MRI (65%), eleven of which were unilateral and concordant with the side of ictal onset demonstrated by VEEG, while two exhibited bilateral MTS. The MRI results of seven patients (35%) were interpreted as normal without lesion or hippocampal sclerosis that could explain seizures. These findings are fairly representative of previous literature demonstrating that an estimated 30% of nonlesional TLE patients show no evidence of hippocampal sclerosis on MRI (9). The clinical information from all subjects is summarized in Table 1.
Statistical analyses of source amplitudes
Distributed EEG source analyses
Voxel-based Wilcoxon tests were performed on the distributed source NIfTI images to investigate the presence of voxel-wise differences between optimal vs suboptimal surgical outcome. These comparisons were performed using the software NPM (http://www.cabiatl.com/mricro/npm/). A statistical threshold of P < 0.05 was considered significant, adjusted for multiple comparisons through false discovery rate (FDR) correction (8). We performed separate analyses for left and right MTLE, comparing CS vs SF patients within each group.
Mean source intensities for both groups separated by left and right-sided MTLE are demonstrated in Fig. 2. For left and right SF and CS groups, EEG source distributions had greatest intensities in the ipsi-, then contra-lateral temporal poles, spreading posteriorly with diminishing intensities through parahippocampal and middle temporal structures.
Results Subjects
The mean post-surgical follow-up was 2 years 5 months. The mean age of patients at surgery
Voxel-based analyses
Voxel-based analyses were performed comparing EEG source images in SF patients vs CS patients. We observed a significant increase in both intensity and distribution of EEG sources in CS patients. Left CS MTLE patients exhibited a broader area of sources on both ipsi- and contralateral sides involving more lateral and posterior mid-temporal structures when compared with the 105
106
F
F M F F F F F F
F
F F F F
7
8 9 10 11 12 13 14 15
16
17 18 19 20
18 20 1 38
47
13 26 39 16 9 8 25 12
21
1 23 1 1 27 14
Age of onset (year)
30 36 51 39
51
22 35 50 45 38 44 51 37
51
39 37 19 42 46 19
Age at surgery (year)
12 16 50 1
4
9 9 11 29 29 36 26 25
30
38 14 18 41 19 5
Duration (year)
240 24 182 52
8
36 182 182.5 48 104 216 12 24
10
30 48 442 130 24 18
Seizures/ year (avg)
Normal R MTS Normal R MTS
Normal
Normal Normal L MTS Normal L MTS Bilateral MTS Bilateral MTS Normal R MTS R MTS L MTS L MTS L MTS L MTS L MTS
MRI
Yes Yes No Yes
Yes
No Yes Yes No Yes Yes Yes Yes
Yes
Yes Yes Yes Yes Yes Yes
Secondary generalized seizures
3 20 4 1
4
5 6 3 4 3 3 5 6
4
2 4 5 13 4 13
# Ictal VEEG recordings temp temp temp temp temp temp
spikes spikes, slow slowing spikes sharps spikes
None L>R temp spikes R temp spikes L>R temp spikes L temp spikes L temp sharps, spikes L temp spikes L>R sharps, L temp slow R temp spikes, L & R temp slow R temp sharps, spikes R>L temp spikes R>L temp spikes R>L temp spikes
R temp spikes
L L L L L L
Interictal EEG
MTLE MTLE MTLE MTLE MTLE MTLE
R R R R
MTLE MTLE MTLE MTLE
R MTLE
R MTLE R MTLE R MTLE L MTLE L MTLE L MTLE L MTLE L MTLE
R MTLE
L L L L L L
Epilepsy
RATL RATL RATL RATL
RATL
RATL RATL RATL LATL LATL LATL LATL LATL
RATL
LATL LATL LATL LATL LATL LATL
Surgery
Temp, temporal lobe; MRI, magnetic resonance imaging; MTS, medial temporal sclerosis; MTLE, medial temporal lobe epilepsy; ATL, anterior temporal lobectomy; CS, continued seizures; SF, seizure free.
M F F M F F
Gender
1 2 3 4 5 6
Patient No.
Table 1 Demographic and clinical characteristics of patients
2.63 2.63 1.46 2.46
2.5
1.42 1.12 2.45 2.27 2.78 3.51 4.8 1.78
2.84
1.05 2.84 1.46 3.95 1.5 3.24
Followup (year)
SF SF SF SF
SF
CS CS CS SF SF SF SF SF
CS
CS CS CS CS CS CS
Surgical outcome
Breedlove et al.
Ictal EEG source and surgical outcome
Figure 2. Average distributed source analysis on ictal EEGs (for electrographic onset plus 5 s) for MTLE patients with postsurgical seizure freedom (SF) and MTLE patients with continued seizures (CS) after surgery. Regions are color-coded based on the probabilistic intensity of EEG sources. The results are overlaid onto an anatomical template.
SF group. The most prominent increase in source intensity in this group was noted in the ipsilateral middle temporal gyrus just superior and posterior to the estimated line of resection and in the contralateral superior gyrus. The right CS MTLE group demonstrated a similar pattern of increased source intensity; however, this involved less contralateral involvement and posterior spread compared with the left MTLE groups. These results are demonstrated in Fig. 3. We did not observe an increase in intensity or distribution of ictal sources in the SF group with left or right MTLE. Discussion
In this study, we employed a probabilistic technique to investigate the location of ictal onset and early ictal spread in patients with MTLE. We observed that patients who did not achieve
seizure freedom with surgery exhibited a broader anatomical distribution of EEG sources within the temporal lobe; corroborating the hypothesis that the epileptogenic network may be less restricted to the medial temporal region in patients with surgical refractoriness. Non-invasive EEG is known to have accurate temporal resolution but a limited capability to resolve the spatial location of cerebral activity. Developments in ESI have attempted to overcome scalp EEG’s poor spatial resolution and have allowed for its implementation in a more exact study of seizure generation, propagation, and maintenance in partial epilepsy. Conceptually, ESI is a computational technique that strives to solve the EEG inverse problem, that is, estimating which of a nearly infinite combination of sources within the brain contributes to a given pattern measured at the surface. ESI has been shown to provide significant information about the sources of ictal epileptic activity when 107
Breedlove et al.
Figure 3. Voxel-based results comparing SF vs CS MTLE patients. Regions where a statistically significant increase in EEG source intensities was observed in CS patients compared with SF patients are color-coded based on the corresponding Z score and P-value (as demonstrated in the color-bars). The typical area resected in ATL surgery is indicated by red outlines. The results are overlaid onto an anatomical template in standard space. All results are corrected for multiple comparisons using False Discovery Rate.
measured against invasive intracranial EEG (10–12), MRI (13), and (SPECT) (14). However, it should be recognized that solving the EEG inverse problem is non-trivial, and to ensure that its solution is unique, prior assumptions or constraints must be made. We adopted strategies to overcome the methodological limitations of ESI using scalp EEG: namely we employed a Bayesian probabilistic mapping of EEG sources and employed a voxelbased analysis of EEG sources. In contrast to the equivalent current dipole (ECD) and multiple dipole methods, this distributed method provides an inverse solution that does not require a prior assumption on the number of dipoles contributing to electrical potentials detected at the scalp. It assumes that the activity is made up of a large number of mini-dipoles simultaneously activated across the cortical sheet. These dipolar sources’ locations and orientations are fixed; only their strengths are variable. Constraints specific to distributed source methods are required to determine the most likely combination of sources, such as those assuming coherence in activity between neighboring neurons over non-neighboring neurons. The technique utilized in this study is distinct from common distributed methods in that its inversion uses Bayesian model comparison 108
to estimate the relative importance of multiple and varied constraints (15). There remains debate, however, as to which inversion technique is the most reliable, and it has been suggested that combining dipole and distributed models provides a better solution than either alone in localizing inter-ictal sources (16). Regarding ictal sources, the dipole method has yet to be compared with the probabilistic method used in this study, and it is possible that there may be utility in combining these two techniques. In this study, we aimed to target ictal source patterns rather than interictal for several reasons. Firstly, one of most difficult challenges of ESI involves dealing with the fact that the electrographic activity from a given source can propagate within milliseconds to distant cortical regions (17). This makes it particularly difficult to define an accurate source for an event that lasts only milliseconds itself. One interesting attribute of distributed ictal source analysis is that it can represent the range of cortical areas involved in early ictal spread. In other words, this method can identify the distribution of early ictal networks across the cortex, rather than a single point that may fail to capture the desired source. Additionally, the area contributing to interictal spikes could be anatomically independent from
Ictal EEG source and surgical outcome the onset area. Some patients continue to demonstrate interictal spikes on EEG even following years of complete seizure freedom following surgery. Moreover, for the distributed method that we chose to explore, it has been suggested that a time window of at least 100 ms should be submitted to source reconstruction to allow for enough time for adequate extraction of information (12). It is noteworthy that this study was performed using conventional clinical EEG, with a limited number of electrodes (n = 21). Certainly, this method could be replicated with data obtained from high-density EEG recordings. This study utilized low-density recordings because this is the standard recording for ictal investigation due to its relative availability and practicality over highdensity setups. Generally, patients can tolerate standard electrodes for much longer than electrode nets, allowing for enough time to capture ictal events and increasing the chance of catching multiple seizure types. Moreover, hippocampal subtypes, which may have important implications regarding epileptogenic networks (18), were not available for this study. Future studies combining the investigation of hippocampal pathologies with probabilistic distributed source analysis could help elucidate the relationships existing between hippocampal sclerosis, epileptogenic networks, and surgical outcomes in MTLE. As explained in the methods, we investigated only the most representative seizure from each patient. A larger sampling of seizures from each patient could have potentially refined the results by addressing intrasubject variability (such as in morphologies, seizure semiologies, and sleep– wake states) and yielding a higher confidence interval for the recording. To assess the probabilistic ESI method’s ability to reproduce similar sources across seizures, we have provided supplemental data demonstrating the source reconstruction for six different seizures in two separate patients. Specifically, we analyzed the source distribution of three ictal events in a post-operatively seizure-free patient and three in a patient with continued seizures after surgery. The results of these reconstructions are summarized in Figure S1. Visual inspection of this figure reveals a similar pattern of distribution of source intensities across seizures within subjects. Of interest, this similar pattern is demonstrated despite that fact that some of the seizures occurred during different sleep–wake states and with varying semiologies. However, the population size of this subanalysis is very small, and comparisons between seizures are only qualitative; therefore, the results of this supplemental data
should be interpreted within the context of these limitations. It is also noteworthy that the source localization analyses were restricted to the temporal lobes. Within the Bayesian framework, the method of restricting source solutions is valuable for incorporating prior knowledge about the areas that are most likely generating the observed discharge. If these assumptions are correct and the designated areas can account for the signal, then the reconstruction model will be greatly simplified. However, if the assumptions are incorrect, the restriction will negatively impact the resulting solutions. Indeed, prior literature has demonstrated extra-temporal abnormalities in patients with MTLE (19). It is therefore possible that other neocortical areas may also be involved in the ictal networks and differ between surgical outcome groups (20) and the results of the reconstructions should be interpreted with this limitation in mind. Taken in the context of the limitations above, our results suggest that the source of the early ictal scalp EEG pattern has a broader anatomical distribution within the temporal lobe in patients with surgical refractory MTLE. These results suggest that there may be a subtle but meaningful difference in the spatial distribution of epileptogenic networks in patients with MTLE. These networks cannot be fully identified by the routine clinical presurgical assessment, but they can influence surgical outcome. Ethical statement
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 The authors have no acknowledgments to declare.
Conflict of interest and sources of funding The authors report no financial or non-financial conflict of interests associated with this study. No funding was received for this study other than that of the authors’ institution.
Supporting Information Additional Supporting Information may be found in the online version of this article. Figure S1. Seizure-related information and the corresponding distributed source analyses on three separate ictal EEGs (electrographic onset plus 5 s) for each of two MTLE patients: (A) a patient from the seizure-free (SF) group with
109
Breedlove et al. right-sided MTLE and (B) a patient from the continued seizures (CS) group with left-sided MTLE. Regions are colorcoded based on the probabilistic intensity of EEG sources. The results are overlaid onto an anatomical template.
References 1. WIEBE S, BLUME WT, GIRVIN JP, ELIASZIW M. A randomized, controlled trial of surgery for temporal-lobe epilepsy. N Engl J Med 2001;345:311–8. 2. ENGEL J Jr, WIEBE S, FRENCH J et al. Practice parameter: temporal lobe and localized neocortical resections for epilepsy. Epilepsia 2003;44:741–51. 3. BONILHA L, MARTZ GU, GLAZIER SS, EDWARDS JC. Subtypes of medial temporal lobe epilepsy: influence on temporal lobectomy outcomes? Epilepsia 2012;53:1–6. 4. SPENCER SS. Neural networks in human epilepsy: evidence of and implications for treatment. Epilepsia 2002;43:219–27. 5. ARRUDA F, CENDES F, ANDERMANN F et al. Mesial atrophy and outcome after amygdalohippocampectomy or temporal lobe removal. Ann Neurol 1996;40: 446–50. 6. Commission on Classification and Terminology of the International League Against Epilepsy. Proposal for revised classification of epilepsies and epileptic syndromes. Epilepsia 1989;30:389–99. 7. American Clinical Neurophysiology Society. Guideline 1: minimum technical requirements for performing clinical electroencephalography. J Clin Neurophysiol 2006;23: 86–91. 8. GENOVESE CR, LAZAR NA, NICHOLS T. Thresholding of statistical maps in functional neuroimaging using the false discovery rate. Neuroimage 2002;15:870–8. 9. CASCINO GD, JACK CR Jr, PARISI JE et al. Magnetic resonance imaging-based volume studies in temporal lobe epilepsy: pathological correlations. Ann Neurol 1991;30: 31–6. 10. ASSAF BA, EBERSOLE JS. Continuous source imaging of scalp ictal rhythms in temporal lobe epilepsy. Epilepsia 1997;38:1114–23.
110
11. 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. 12. KOESSLER L, BENAR C, MAILLARD L et al. Source localization of ictal epileptic activity investigated by high resolution EEG and validated by SEEG. Neuroimage 2010;51:642–53. 13. WORRELL GA, LAGERLUND TD, SHARBROUGH FW 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. 14. 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. 15. PHILLIPS C, MATTOUT J, RUGG MD, MAQUET P, FRISTON KJ. An empirical Bayesian solution to the source reconstruction problem in EEG. Neuroimage 2005;24:997– 1011. 16. PLUMMER C, WAGNER M, FUCHS M, HARVEY AS, COOK MJ. Dipole versus distributed EEG source localization for single versus averaged spikes in focal epilepsy. J Clin Neurophysiol 2010;27:141–62. 17. ALARCON G, GUY CN, BINNIE CD, WALKER SR, ELWES RD, POLKEY CE. Intracerebral propagation of interictal activity in partial epilepsy: implications for source localisation. J Neurol Neurosurg Psychiatry 1994;57:435–49. 18. BLUMCKE I, CORAS R, MIYATA H, OZKARA C. Defining clinico-neuropathological subtypes of mesial temporal lobe epilepsy with hippocampal sclerosis. Brain Pathol 2012;22:402–11. 19. BONILHA L, EDWARDS JC, KINSMAN SL et al. Extrahippocampal gray matter loss and hippocampal deafferentation in patients with temporal lobe epilepsy. Epilepsia 2010;51:519–28. 20. BONILHA L, HELPERN JA, SAINJU R et al. Presurgical connectome and postsurgical seizure control in temporal lobe epilepsy. Neurology 2013;81:1704–10.
© 2014 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd ACTA NEUROLOGICA SCANDINAVICA
Probabilistic ictal EEG sources and temporal lobe epilepsy surgical outcome Breedlove J, Nesland T, Vandergrift WA 3rd, Betting LE, Bonilha L. Probabilistic ictal EEG sources and temporal lobe epilepsy surgical outcome. Acta Neurol Scand 2014: 130: 103–110. © 2014 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd. Objective – For patients with medication refractory medial temporal lobe epilepsy (MTLE), surgery offers the hope of a cure. However, up to 30% of patients with MTLE continue to experience disabling seizures after surgery. The reasons why some patients do not achieve seizure freedom are poorly understood. A promising theory suggests that epileptogenic networks are broadly distributed in surgically refractory MTLE, involving regions beyond the medial temporal lobe. In this retrospective study, we aimed to investigate the distribution of epileptogenic networks in MTLE using Bayesian distributed EEG source analysis from preoperative ictal onset recordings. This analysis has the advantage of generating maps of source probability, which can be subjected to voxel-based statistical analyses. Methods – We compared 10 patients who achieved post-surgical seizure freedom with 10 patients who continued experiencing seizures after surgery. Voxelbased Wilcoxon tests were employed with correction for multiple comparisons. Results – We observed that ictal EEG source intensities were significantly more likely to occur in lateral temporal and posterior medial temporal regions in patients with continued seizures post-surgery. Conclusions – Our findings support the theory of broader spatial distribution of epileptogenic networks at seizure onset in patients with surgically refractory MTLE.
Introduction
Anterior temporal lobectomy (ATL) is the most effective treatment for medial temporal lobe epilepsy (MTLE) in patients whose seizure control was not achieved with antiepileptic drugs (AEDs) (1, 2). Up to 75% (2) of patients with MTLE may become completely free of seizures after surgery. Unfortunately, 25–30% of patients who undergo surgical treatment continue to experience disabling seizures, in spite of a thorough presurgical evaluation. These patients often appear clinically identical to those who can become seizure free, with compatible clinical semiology, imaging, and ictal scalp electroencephalography (EEG) profiles. This discrepancy in outcome suggests that there may be subtle, yet clinically meaningful, differences in the disease mechanisms that are not identified by current clinical assessment tools.
J. Breedlove1, T. Nesland1, W. A. Vandergrift 3rd1, L. E. Betting2, L. Bonilha1 1 Department of Neurology, Comprehensive Epilepsy Center, Medical University of South Carolina, Charleston, SC, USA; 2Departamento de Neurologia, Psiquiatria e Psicologia, Faculdade de Medicina de Botucatu – UNESP, Botucatu, SP, Brazil
Key words: electroencephalography; source localization; temporal lobe epilepsy; epilepsy surgery Leonardo Bonilha MD PhD, Comprehensive Epilepsy Center, Division of Neurology, Medical University of South Carolina Tel: 843 792 3383 Fax: 842 792 8626 e-mail: [email protected] Accepted for publication March 14, 2014
A promising theory suggests that the location and distribution of the epileptogenic zone may be different in patients with surgical refractoriness. Patients who do not respond to surgery may exhibit a broader distribution of the epileptogenic networks (3). These networks may extend beyond the area of resection, and the post-surgical persistence of abnormal networks may continue to generate seizures after surgery (4, 5). This study sought to investigate the distribution of epileptogenic networks in preoperative MTLE patients using electroencephalographic source imaging (ESI) on ictal EEG recordings. Specifically, we applied a unique Bayesian approach to evaluate the probabilistic distribution of ictal EEG sources of presurgical seizures in a cohort of patients with MTLE who underwent surgical treatment. We hypothesized that the probabilistic EEG ictal sources would exhibit an anatomically broader 103
Breedlove et al. distribution in patients with poor ATL outcomes, when compared with those with favorable ATL outcomes. Methods Participants
We retrospectively studied twenty patients followed at the Medical University of South Carolina who were diagnosed with medication refractory MTLE according to the parameters defined by the International League Against Epilepsy – ILAE (6). All patients underwent ATL performed by the same neurosurgeon to treat epilepsy after a comprehensive presurgical investigation, which included a detailed neurological examination, high-resolution magnetic resonance imaging (MRI), and long-term scalp video EEG (VEEG). Additionally, some patients underwent complimentary studies such as single-photon emission computed tomography (SPECT) and intracranial EEG recording to ensure the lateralization of their seizures. All findings from the presurgical evaluations were discussed during a multidisciplinary epilepsy conference, and decisions regarding surgical treatment were achieved through consensus. Patients with VEEG recordings demonstrating bilateral seizure onset or MRI findings indicating large abnormalities (such as sizeable infarcts or mass lesions) were excluded from the study.
for all patients in the post-operative period. Outcomes were classified according to the Engel Surgical Outcome Scale, grouped into two main groups: (i) free of disabling seizures or seizure free (SF), equivalent to Engel Class I; or (ii) continued seizures (CS), equivalent to Engel Classes II, III, or IV. Ictal EEG selection and analysis
All patients included in this study demonstrated similar patterns of ictal EEG across their seizures. For each patient having multiple seizure recordings available, we selected an ictal EEG event that best demonstrated the patient’s typical seizure as indicated by chart history and VEEG reports. For example, if the patient’s history indicated that his or her seizures typically occurred during sleep, a seizure that arose out of sleep was chosen for source localization. The event was also selected on the basis that it was the one mostly free from artifact. Ictal onset was defined as the first visual rhythmic electrographic change from background pattern and was marked by the patient’s epileptologist prior to epilepsy surgery. We analyzed 30-s epochs that were manually extracted from these EEGs, 10 s to +20 s with respect to seizure onset. The frequency of scalp rhythmic discharges at ictal onset was determined through visual inspection for each electrographic seizure. Distributed source analysis
EEG recording
All patients were admitted to the epilepsy-monitoring unit for long-term scalp VEEG prior to surgery. EEGs were recorded to a vertex reference during acquisition using 21 electrodes. The array selected included 19 of the standard 10–20 system positions (excluding ear references, A1 and A2), with the addition of FT9 and FT10 to record anterior portions of the temporal lobe. Patients were monitored using a XLTEK acquisition station and an EMU40 head box (XLTEK, Natus, Ontario, Canada). Data were acquired at a sampling rate of 256 Hz, a band pass of 1–70 Hz, and all electrodes were placed with impedances below 5 kOhms in accordance with ACNS guidelines (7). Surgical outcome
Surgical outcome was defined with at least 1 year of post-surgical follow-up. None of the patients had intra- or perioperative complications, and the preoperative antiepileptic regimen was continued 104
To reconstruct ictal activity into 3D voxel space, all ictal epochs were submitted to source analysis using the distributed imaging method provided by SPM8 (http://www.fil.ion.ucl.ac.uk/spm/software/ spm8/), in accordance with the steps outlined below. This workflow is illustrated in Fig. 1. Preprocessing, coregistration, and forward modeling
Each raw epoched data set was run through a 1–35 Hz band pass filter to reduce artifact from 60 Hz noise, movement, and muscle contraction. Sensor locations were loaded in a MATLAB file containing the MNI coordinates of each sensor and coregistered using a file containing the locations of 3 fiducials (nasion, right preauricular, and left preauricular). A realistic boundary element method (BEM) was used for forward modeling. Inversion
For each ictal epoch, a custom greedy search (GS) was implemented using the entire 30-s time
Ictal EEG source and surgical outcome
Figure 1. Workflow of distributed EEG source analysis. The raw EEG data (including the electrographic onset plus 5 s) are submitted to preprocessing. The electrodes are coregistered into standard space allowing for the calculation of scalp field maps for each millisecond. The forward model is created using the realistic boundary element method (BEM) and inversed using a window created around the ictal frequency. Resulting is a 3D image displaying the probabilistic source intensities. The example shown here is a left-onset temporal lobe seizure.
window, low-frequency filer (LFF) of 1 Hz, and high-frequency filter (HFF) of 48 Hz. Solutions were restricted to the temporal lobes using MNI coordinates corresponding to the areas of interest. This restriction simplified the model and was based on prior knowledge that seizures were originating from within the temporal lobes, as per the combined presurgical data including semiology, VEEG, and/or MRI for each patient. For each ictal event, the resulting source amplitudes were estimated and written as a single 3D NIfTI (Neuroimaging Informatics Technology Initiative) image. This was accomplished using a time window set at ictal onset plus 5 s. To focus on seizure activity and avoid unintentional analysis of background patterns, images were created using a 3 Hz frequency window surrounding the patient’s specific ictal frequency at onset (frequency of rhythmic discharges 1.5 Hz).
was 39 10 years. There were no significant differences in age (T = 1.38, P = 0.19) or gender distribution (v2 = 3.53, P = 0.06) between the CS and SF groups. Patients in both groups were found to have similar seizure frequencies (P = 0.71), age of onset of epilepsy (P = 0.65), and duration of epilepsy (P = 0.59). Thirteen patients exhibited medial temporal sclerosis (MTS) during routine diagnostic MRI (65%), eleven of which were unilateral and concordant with the side of ictal onset demonstrated by VEEG, while two exhibited bilateral MTS. The MRI results of seven patients (35%) were interpreted as normal without lesion or hippocampal sclerosis that could explain seizures. These findings are fairly representative of previous literature demonstrating that an estimated 30% of nonlesional TLE patients show no evidence of hippocampal sclerosis on MRI (9). The clinical information from all subjects is summarized in Table 1.
Statistical analyses of source amplitudes
Distributed EEG source analyses
Voxel-based Wilcoxon tests were performed on the distributed source NIfTI images to investigate the presence of voxel-wise differences between optimal vs suboptimal surgical outcome. These comparisons were performed using the software NPM (http://www.cabiatl.com/mricro/npm/). A statistical threshold of P < 0.05 was considered significant, adjusted for multiple comparisons through false discovery rate (FDR) correction (8). We performed separate analyses for left and right MTLE, comparing CS vs SF patients within each group.
Mean source intensities for both groups separated by left and right-sided MTLE are demonstrated in Fig. 2. For left and right SF and CS groups, EEG source distributions had greatest intensities in the ipsi-, then contra-lateral temporal poles, spreading posteriorly with diminishing intensities through parahippocampal and middle temporal structures.
Results Subjects
The mean post-surgical follow-up was 2 years 5 months. The mean age of patients at surgery
Voxel-based analyses
Voxel-based analyses were performed comparing EEG source images in SF patients vs CS patients. We observed a significant increase in both intensity and distribution of EEG sources in CS patients. Left CS MTLE patients exhibited a broader area of sources on both ipsi- and contralateral sides involving more lateral and posterior mid-temporal structures when compared with the 105
106
F
F M F F F F F F
F
F F F F
7
8 9 10 11 12 13 14 15
16
17 18 19 20
18 20 1 38
47
13 26 39 16 9 8 25 12
21
1 23 1 1 27 14
Age of onset (year)
30 36 51 39
51
22 35 50 45 38 44 51 37
51
39 37 19 42 46 19
Age at surgery (year)
12 16 50 1
4
9 9 11 29 29 36 26 25
30
38 14 18 41 19 5
Duration (year)
240 24 182 52
8
36 182 182.5 48 104 216 12 24
10
30 48 442 130 24 18
Seizures/ year (avg)
Normal R MTS Normal R MTS
Normal
Normal Normal L MTS Normal L MTS Bilateral MTS Bilateral MTS Normal R MTS R MTS L MTS L MTS L MTS L MTS L MTS
MRI
Yes Yes No Yes
Yes
No Yes Yes No Yes Yes Yes Yes
Yes
Yes Yes Yes Yes Yes Yes
Secondary generalized seizures
3 20 4 1
4
5 6 3 4 3 3 5 6
4
2 4 5 13 4 13
# Ictal VEEG recordings temp temp temp temp temp temp
spikes spikes, slow slowing spikes sharps spikes
None L>R temp spikes R temp spikes L>R temp spikes L temp spikes L temp sharps, spikes L temp spikes L>R sharps, L temp slow R temp spikes, L & R temp slow R temp sharps, spikes R>L temp spikes R>L temp spikes R>L temp spikes
R temp spikes
L L L L L L
Interictal EEG
MTLE MTLE MTLE MTLE MTLE MTLE
R R R R
MTLE MTLE MTLE MTLE
R MTLE
R MTLE R MTLE R MTLE L MTLE L MTLE L MTLE L MTLE L MTLE
R MTLE
L L L L L L
Epilepsy
RATL RATL RATL RATL
RATL
RATL RATL RATL LATL LATL LATL LATL LATL
RATL
LATL LATL LATL LATL LATL LATL
Surgery
Temp, temporal lobe; MRI, magnetic resonance imaging; MTS, medial temporal sclerosis; MTLE, medial temporal lobe epilepsy; ATL, anterior temporal lobectomy; CS, continued seizures; SF, seizure free.
M F F M F F
Gender
1 2 3 4 5 6
Patient No.
Table 1 Demographic and clinical characteristics of patients
2.63 2.63 1.46 2.46
2.5
1.42 1.12 2.45 2.27 2.78 3.51 4.8 1.78
2.84
1.05 2.84 1.46 3.95 1.5 3.24
Followup (year)
SF SF SF SF
SF
CS CS CS SF SF SF SF SF
CS
CS CS CS CS CS CS
Surgical outcome
Breedlove et al.
Ictal EEG source and surgical outcome
Figure 2. Average distributed source analysis on ictal EEGs (for electrographic onset plus 5 s) for MTLE patients with postsurgical seizure freedom (SF) and MTLE patients with continued seizures (CS) after surgery. Regions are color-coded based on the probabilistic intensity of EEG sources. The results are overlaid onto an anatomical template.
SF group. The most prominent increase in source intensity in this group was noted in the ipsilateral middle temporal gyrus just superior and posterior to the estimated line of resection and in the contralateral superior gyrus. The right CS MTLE group demonstrated a similar pattern of increased source intensity; however, this involved less contralateral involvement and posterior spread compared with the left MTLE groups. These results are demonstrated in Fig. 3. We did not observe an increase in intensity or distribution of ictal sources in the SF group with left or right MTLE. Discussion
In this study, we employed a probabilistic technique to investigate the location of ictal onset and early ictal spread in patients with MTLE. We observed that patients who did not achieve
seizure freedom with surgery exhibited a broader anatomical distribution of EEG sources within the temporal lobe; corroborating the hypothesis that the epileptogenic network may be less restricted to the medial temporal region in patients with surgical refractoriness. Non-invasive EEG is known to have accurate temporal resolution but a limited capability to resolve the spatial location of cerebral activity. Developments in ESI have attempted to overcome scalp EEG’s poor spatial resolution and have allowed for its implementation in a more exact study of seizure generation, propagation, and maintenance in partial epilepsy. Conceptually, ESI is a computational technique that strives to solve the EEG inverse problem, that is, estimating which of a nearly infinite combination of sources within the brain contributes to a given pattern measured at the surface. ESI has been shown to provide significant information about the sources of ictal epileptic activity when 107
Breedlove et al.
Figure 3. Voxel-based results comparing SF vs CS MTLE patients. Regions where a statistically significant increase in EEG source intensities was observed in CS patients compared with SF patients are color-coded based on the corresponding Z score and P-value (as demonstrated in the color-bars). The typical area resected in ATL surgery is indicated by red outlines. The results are overlaid onto an anatomical template in standard space. All results are corrected for multiple comparisons using False Discovery Rate.
measured against invasive intracranial EEG (10–12), MRI (13), and (SPECT) (14). However, it should be recognized that solving the EEG inverse problem is non-trivial, and to ensure that its solution is unique, prior assumptions or constraints must be made. We adopted strategies to overcome the methodological limitations of ESI using scalp EEG: namely we employed a Bayesian probabilistic mapping of EEG sources and employed a voxelbased analysis of EEG sources. In contrast to the equivalent current dipole (ECD) and multiple dipole methods, this distributed method provides an inverse solution that does not require a prior assumption on the number of dipoles contributing to electrical potentials detected at the scalp. It assumes that the activity is made up of a large number of mini-dipoles simultaneously activated across the cortical sheet. These dipolar sources’ locations and orientations are fixed; only their strengths are variable. Constraints specific to distributed source methods are required to determine the most likely combination of sources, such as those assuming coherence in activity between neighboring neurons over non-neighboring neurons. The technique utilized in this study is distinct from common distributed methods in that its inversion uses Bayesian model comparison 108
to estimate the relative importance of multiple and varied constraints (15). There remains debate, however, as to which inversion technique is the most reliable, and it has been suggested that combining dipole and distributed models provides a better solution than either alone in localizing inter-ictal sources (16). Regarding ictal sources, the dipole method has yet to be compared with the probabilistic method used in this study, and it is possible that there may be utility in combining these two techniques. In this study, we aimed to target ictal source patterns rather than interictal for several reasons. Firstly, one of most difficult challenges of ESI involves dealing with the fact that the electrographic activity from a given source can propagate within milliseconds to distant cortical regions (17). This makes it particularly difficult to define an accurate source for an event that lasts only milliseconds itself. One interesting attribute of distributed ictal source analysis is that it can represent the range of cortical areas involved in early ictal spread. In other words, this method can identify the distribution of early ictal networks across the cortex, rather than a single point that may fail to capture the desired source. Additionally, the area contributing to interictal spikes could be anatomically independent from
Ictal EEG source and surgical outcome the onset area. Some patients continue to demonstrate interictal spikes on EEG even following years of complete seizure freedom following surgery. Moreover, for the distributed method that we chose to explore, it has been suggested that a time window of at least 100 ms should be submitted to source reconstruction to allow for enough time for adequate extraction of information (12). It is noteworthy that this study was performed using conventional clinical EEG, with a limited number of electrodes (n = 21). Certainly, this method could be replicated with data obtained from high-density EEG recordings. This study utilized low-density recordings because this is the standard recording for ictal investigation due to its relative availability and practicality over highdensity setups. Generally, patients can tolerate standard electrodes for much longer than electrode nets, allowing for enough time to capture ictal events and increasing the chance of catching multiple seizure types. Moreover, hippocampal subtypes, which may have important implications regarding epileptogenic networks (18), were not available for this study. Future studies combining the investigation of hippocampal pathologies with probabilistic distributed source analysis could help elucidate the relationships existing between hippocampal sclerosis, epileptogenic networks, and surgical outcomes in MTLE. As explained in the methods, we investigated only the most representative seizure from each patient. A larger sampling of seizures from each patient could have potentially refined the results by addressing intrasubject variability (such as in morphologies, seizure semiologies, and sleep– wake states) and yielding a higher confidence interval for the recording. To assess the probabilistic ESI method’s ability to reproduce similar sources across seizures, we have provided supplemental data demonstrating the source reconstruction for six different seizures in two separate patients. Specifically, we analyzed the source distribution of three ictal events in a post-operatively seizure-free patient and three in a patient with continued seizures after surgery. The results of these reconstructions are summarized in Figure S1. Visual inspection of this figure reveals a similar pattern of distribution of source intensities across seizures within subjects. Of interest, this similar pattern is demonstrated despite that fact that some of the seizures occurred during different sleep–wake states and with varying semiologies. However, the population size of this subanalysis is very small, and comparisons between seizures are only qualitative; therefore, the results of this supplemental data
should be interpreted within the context of these limitations. It is also noteworthy that the source localization analyses were restricted to the temporal lobes. Within the Bayesian framework, the method of restricting source solutions is valuable for incorporating prior knowledge about the areas that are most likely generating the observed discharge. If these assumptions are correct and the designated areas can account for the signal, then the reconstruction model will be greatly simplified. However, if the assumptions are incorrect, the restriction will negatively impact the resulting solutions. Indeed, prior literature has demonstrated extra-temporal abnormalities in patients with MTLE (19). It is therefore possible that other neocortical areas may also be involved in the ictal networks and differ between surgical outcome groups (20) and the results of the reconstructions should be interpreted with this limitation in mind. Taken in the context of the limitations above, our results suggest that the source of the early ictal scalp EEG pattern has a broader anatomical distribution within the temporal lobe in patients with surgical refractory MTLE. These results suggest that there may be a subtle but meaningful difference in the spatial distribution of epileptogenic networks in patients with MTLE. These networks cannot be fully identified by the routine clinical presurgical assessment, but they can influence surgical outcome. Ethical statement
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 The authors have no acknowledgments to declare.
Conflict of interest and sources of funding The authors report no financial or non-financial conflict of interests associated with this study. No funding was received for this study other than that of the authors’ institution.
Supporting Information Additional Supporting Information may be found in the online version of this article. Figure S1. Seizure-related information and the corresponding distributed source analyses on three separate ictal EEGs (electrographic onset plus 5 s) for each of two MTLE patients: (A) a patient from the seizure-free (SF) group with
109
Breedlove et al. right-sided MTLE and (B) a patient from the continued seizures (CS) group with left-sided MTLE. Regions are colorcoded based on the probabilistic intensity of EEG sources. The results are overlaid onto an anatomical template.
References 1. WIEBE S, BLUME WT, GIRVIN JP, ELIASZIW M. A randomized, controlled trial of surgery for temporal-lobe epilepsy. N Engl J Med 2001;345:311–8. 2. ENGEL J Jr, WIEBE S, FRENCH J et al. Practice parameter: temporal lobe and localized neocortical resections for epilepsy. Epilepsia 2003;44:741–51. 3. BONILHA L, MARTZ GU, GLAZIER SS, EDWARDS JC. Subtypes of medial temporal lobe epilepsy: influence on temporal lobectomy outcomes? Epilepsia 2012;53:1–6. 4. SPENCER SS. Neural networks in human epilepsy: evidence of and implications for treatment. Epilepsia 2002;43:219–27. 5. ARRUDA F, CENDES F, ANDERMANN F et al. Mesial atrophy and outcome after amygdalohippocampectomy or temporal lobe removal. Ann Neurol 1996;40: 446–50. 6. Commission on Classification and Terminology of the International League Against Epilepsy. Proposal for revised classification of epilepsies and epileptic syndromes. Epilepsia 1989;30:389–99. 7. American Clinical Neurophysiology Society. Guideline 1: minimum technical requirements for performing clinical electroencephalography. J Clin Neurophysiol 2006;23: 86–91. 8. GENOVESE CR, LAZAR NA, NICHOLS T. Thresholding of statistical maps in functional neuroimaging using the false discovery rate. Neuroimage 2002;15:870–8. 9. CASCINO GD, JACK CR Jr, PARISI JE et al. Magnetic resonance imaging-based volume studies in temporal lobe epilepsy: pathological correlations. Ann Neurol 1991;30: 31–6. 10. ASSAF BA, EBERSOLE JS. Continuous source imaging of scalp ictal rhythms in temporal lobe epilepsy. Epilepsia 1997;38:1114–23.
110
11. 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. 12. KOESSLER L, BENAR C, MAILLARD L et al. Source localization of ictal epileptic activity investigated by high resolution EEG and validated by SEEG. Neuroimage 2010;51:642–53. 13. WORRELL GA, LAGERLUND TD, SHARBROUGH FW 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. 14. 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. 15. PHILLIPS C, MATTOUT J, RUGG MD, MAQUET P, FRISTON KJ. An empirical Bayesian solution to the source reconstruction problem in EEG. Neuroimage 2005;24:997– 1011. 16. PLUMMER C, WAGNER M, FUCHS M, HARVEY AS, COOK MJ. Dipole versus distributed EEG source localization for single versus averaged spikes in focal epilepsy. J Clin Neurophysiol 2010;27:141–62. 17. ALARCON G, GUY CN, BINNIE CD, WALKER SR, ELWES RD, POLKEY CE. Intracerebral propagation of interictal activity in partial epilepsy: implications for source localisation. J Neurol Neurosurg Psychiatry 1994;57:435–49. 18. BLUMCKE I, CORAS R, MIYATA H, OZKARA C. Defining clinico-neuropathological subtypes of mesial temporal lobe epilepsy with hippocampal sclerosis. Brain Pathol 2012;22:402–11. 19. BONILHA L, EDWARDS JC, KINSMAN SL et al. Extrahippocampal gray matter loss and hippocampal deafferentation in patients with temporal lobe epilepsy. Epilepsia 2010;51:519–28. 20. BONILHA L, HELPERN JA, SAINJU R et al. Presurgical connectome and postsurgical seizure control in temporal lobe epilepsy. Neurology 2013;81:1704–10.
