Clinical Neurophysiology 127 (2016) 108–116
Contents lists available at ScienceDirect
Clinical Neurophysiology journal homepage: www.elsevier.com/locate/clinph
Application of 256-channel dense array electroencephalographic source imaging in presurgical workup of temporal lobe epilepsy Rui Feng a, Jie Hu a,b,⇑, Li Pan a,⇑, Jinsong Wu a, Liqin Lang a, Shize Jiang a, Xin Gu b, Jun Guo b, Liangfu Zhou a a b
Department of Neurosurgery, Huashan Hospital, Fudan University, Shanghai, China Department of Neurosurgery, Jing’an Branch of Huashan Hospital, Fudan University, Shanghai, China
a r t i c l e
i n f o
Article history: Accepted 13 March 2015 Available online 23 March 2015 Keywords: 256-channel dense array electroencephalographic source imaging Temporal lobe epilepsy Positron emission tomography Conventional electroencephalography Magnetic resonance imaging Epileptogenic zone Finite difference model
h i g h l i g h t s 256-channel dense array electroencephalographic source imaging (dESI) showed the best sensitivity
(>90%) and specificity (75%) with comparison to conventional tools. Analysis of the relationship between the 256-channel dESI source patterns and surgical outcome
showed that cases with ‘‘single source’’ had better prognosis than with ‘‘multiple sources.’’ The resection of dESI sources representative of the irritative zone is related to good surgical prognosis
in temporal lobe epilepsy (TLE).
a b s t r a c t Objective: This study evaluated the localization precision of 256-channel dense array electroencephalographic source imaging (dESI) in comparison to conventional noninvasive tools. In addition, the study was designed to analyze the relationship between the 256-channel dESI source patterns and surgical outcome. Methods: Forty-three patients with temporal lobe epilepsy (TLE) who underwent one-stage resective surgeries were recruited in this study. We compared dESI with other noninvasive evaluation methods by comparing results with resections that eliminate or significantly reduced seizures according to sublobule and lobule criteria. Sensitivity and specificity of multiple evaluation methods were calculated. Kaplan–Meier analysis was performed to evaluate the relationship between source patterns and surgical outcome. Results: dESI showed the best sensitivity (sub-lobule, 91.4%; lobule, 97.1%) and specificity (75%) for both sub-lobule and lobule criteria. The Kaplan–Meier survival analysis showed that cases with ‘‘single source’’ had better prognosis than with ‘‘multiple sources’’ (p < 0.05); cases of ‘‘sources within resection’’ showed better surgical prognosis than cases of ‘‘sources outside resection’’ (p < 0.05). Conclusion: In this study, 256-channel dESI provided a higher clinical yield than the other most broadly used noninvasive presurgical workup tools. dESI results with ‘‘single source’’ correlated strongly to good prognosis, while cases with ‘‘multiple sources’’ may cautiously be considered as candidates for one-stage resective surgeries. The resection of the irritative zone identified by interictal epileptiform discharges (IEDs) was related to good surgical prognosis in TLE. Significance: In the presurgical workup of TLE, the clinical yield of 256-channel dESI is high. Patterns of dESI results are related to surgical prognosis, and they can be instructive for presurgical planning. The resection of the irritative zone can be related to good surgical prognosis in TLE. Ó 2015 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
⇑ Corresponding authors at: Department of Neurosurgery, Huashan Hospital, Fudan University, 12 Wulumuqi Zhong Road, Shanghai 200040, China. Tel.: +86 13361913896 (J. Hu), +86 13601683163 (L. Pan). E-mail addresses: [email protected] (J. Hu), [email protected] (L. Pan). http://dx.doi.org/10.1016/j.clinph.2015.03.009 1388-2457/Ó 2015 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
R. Feng et al. / Clinical Neurophysiology 127 (2016) 108–116
109
1. Introduction
2.2. Semiology evaluation and cEEG analysis
Temporal lobe epilepsy (TLE) is the most common type of intractable epilepsy, and it can be divided into three subgroups: hippocampal sclerosis, which constitutes at least 60–70% of TLE (Cascino et al., 1991; Carne et al., 2004), lesional TLE with various kinds of epileptogenic structural lesions (Carne et al., 2004), and TLE without observable abnormalities on structural magnetic resonance imaging (MRI) (MRI-negative TLE, Bien et al., 2009; LoPinto-Khoury et al., 2012). Invasive intracranial electroencephalographic (icEEG) recordings are currently required in clinical practice for the determination of the epileptogenic zone (EZ). Unfortunately, localization accuracy with icEEG is still far from 100%, due to the fact that icEEG electrodes practically cannot be placed to cover the entire brain. Moreover, the invasive procedure is expensive and of high risk. Because of these limitations, noninvasive tools including conventional electroencephalography (cEEG), semiology evaluation, structural MRI, and interictal fluorodeoxyglucose positron emission tomography (FDG-PET) are usually considered for a comprehensive evaluation. Recent studies have shown that dense array EEG source imaging (dESI) can significantly contribute to epilepsy presurgical evaluation (Brodbeck et al., 2011). The present study extends prior dESI for epilepsy evaluation research by comparing EZ localization accuracy, with respect to surgical outcome, of 256-channel dESI performed with an atlas head model against PET, MRI, cEEG, and semiology.
Semiology details were analyzed based on high-definition videos recorded simultaneously with EEG. cEEG examinations always include both waking and sleeping EEG, and they were performed using a 16-channel EEG system (Bio-logic, US). Electrode placement followed the international 10–20 system (Fp1/2, F3/4, F7/8, C3/4, T3/4, T5/6, P3/P4, O1/O2, Fz, Cz, and Oz), and the data were sampled at 512 Hz with a 0.1–100-Hz band-pass filter. The off-line cEEG analysis was performed by two experienced EEG experts.
2. Methods Forty-three patients with TLE were included in this study, which was approved by the ethics committee of Huashan Hospital (2013 Lin Shen No. 286). Informed consent was obtained from each patient. 2.1. Patient selection We selected patients with refractory epilepsy from a pool of patients who were candidates for one-stage resective surgeries. Our comprehensive presurgical workup for the determination of EZ includes semiology evaluation, cEEG analysis, structural MRI, interictal FDG-PET, and dESI. 2.1.1. Inclusion criteria The following were the inclusion criteria: 1. All patients were treated with multiple antiepilepsy drugs (AEDs) for at least 1 year, which yielded poor control of epilepsy. 2. TLE should fundamentally be supported by semiology indicating epilepsy and scalp cEEG, which at least showed interictal epileptic discharges (IEDs). 2.1.2. Exclusion criteria The following were the exclusion criteria: 1. Patients with malignancy-related TLE, TLE with huge epileptogenic lesions, clearly identified bilateral TLE, or multifocal epilepsy. 2. Patients whose noninvasive evaluation did not permit the determination of EZ focus, and who underwent invasive evaluation. A comprehensive presurgical workup in this study for the determination of EZ includes semiology evaluation and cEEG analysis, structural MRI, interictal FDG-PET, and dESI.
2.3. Structural MRI All MRI data were acquired at Huashan Hospital with a 1.5-T MRI (Signa HDx 1.5 T, GE). MRI scanning was performed according to a standardized epilepsy protocol (image size, 512 512; thickness, 4 mm) including coronal T2WI (TR, 5640 ms; TE, 90.2 ms;), coronal T2 FLAIR (TR, 7327 ms; TE, 153 ms), axial T1WI (TR, 760 ms; TE, 14.4 ms), axial T2WI (TR, 3820 ms; TE, 89.5 ms), axial T2 FLAIR (TR, 7427 ms; TE, 160 ms), and sagittal T2 FLAIR (TR, 7327 ms; TE, 159.8 ms). An enhanced T1WI sequence was performed when a tumor was suspected. 2.4. FDG-PET All interictal PET studies were carried out at Huashan Hospital. PET hypometabolism was identified by two experienced physicians who were blind to the patients’ diagnoses. The physicians noted the presence of unilateral, focal decreased FDG uptake in the temporal lobe by visual inspection. 2.5. 256-channel dESI A standardized workflow of 256-channel dEEG data processing was applied in our center (Fig. 1). 2.5.1. 256-channel dEEG acquisition A 256-channel Geodesic Sensor Net (Electrical Geodesics, Inc., Eugene, OR, USA) was selected according to each patient’s head circumference. Each electrode in the net contains a small salinesoaked sponge that contacts the scalp directly. The entire application process can be accomplished within 20 min by experienced technicians, and it requires no skin abrasion or a haircut. The sampling rate was set to 250 Hz. Sensor impedances were kept under 100 KO, which is an acceptable impedance level given the high-input impedance of modern EEG amplifiers (Ferree et al., 2001). Data were recorded for approximately 30–60 min. During dEEG recordings, the patients were fully relaxed and lying on a bed, most of the time with their eyes closed. Usually, patients would be awake during recordings, but if patients fell asleep, the time point would be marked by an EEG technician, who was always present. 2.5.2. IED identification, segmentation, and averaging Monomorphic and repetitive IEDs were identified and reviewed by two experienced EEG reviewers. Once identified, they were grouped according to similar pattern of spatial distribution judged by waveform distribution and topographic maps. There can be either one or several patterns of such spatial distribution for each patient. If a group of IEDs occurred fewer than three times, it was not considered for further dESI analysis. For those IEDs that were kept for analysis, they were segmented (500 ms before and 500 ms after the spike peak) and averaged to derive the average spikes with a good signal-to-noise ratio for dESI.
110
R. Feng et al. / Clinical Neurophysiology 127 (2016) 108–116
Fig. 1. The workflow of 256-channel dESI in Functional Neurosurgery Center, Department of Neurosurgery, Huashan Hospital.
2.5.3. Forward model Standard spherical head models (Berg and Scherg, 1994) account for the scalp, skull, cerebrospinal fluid (CSF), and brain, and conductivities are usually set to 0.3300 S/m (brain), 1 S/m (CSF), 0.0042 S/m (skull), and 0.3300 S/m (scalp). In this study, we employed a realistic atlas head model based on the finite difference method (FDM) (Luu et al., 2014). The FDM allows accurate characterization of cranial orifices to account for their influence on the current flow. Tissue compartments of the FDM model (e.g., scalp, skull, and CSF) were constructed from whole-head MRI and CT scans of a single subject (Colin27) whose head geometry closely matches the Montreal Neurological Institute (MNI) average MRI (MNI305) (Luu et al., 2014). The white and gray matter were derived from the probabilistic partitioning of the MNI305 average brain (http://www.bic.mni.mcgill.ca), and registered to the individual Colin27 head. Source locations (a total of 2447 sources) were evenly seeded in the gray matter volume of the MNI305 average brain. Conductivity values used in the FDM model were set to 0.25 S/m (brain), 1.8 S/m (CSF), 0.018 S/m (skull), and 0.44 S/m (scalp). According to several recent studies, the conductivity ratio of skull to brain tissue is close to 1:14, rather than the traditionally assumed 1:80 value (Ferree et al., 2000; Ryynanen et al., 2006; Zhang et al., 2006). To generate the lead field data (i.e., the description of the current propagation from a dipole position to the scalp, where the EEG is recorded), the recording location (i.e., EEG sensor positions on the scalp) must be identified. Because we did not acquire sensor locations for each individual patient, an average (based on five adult subjects) of the 256-channel location (Luu P, Ferree T. Unpublished manuscript; 2005. Determination of the HydroCel Geodesic Sensor Nets’ Average Electrode Positions and Their 10–10 International Equivalents) was used to register with the realistic atlas head model. 2.5.4. dEEG source localization dESI was performed based on the FDM model using the low-resolution electromagnetic tomography (LORETA) method (Pascual-
Marqui et al., 1994). Sources of the spikes were determined at 50% (relative to the peak) of the rising slope of the spike (Holmes, 2008), and the results are presented on the atlas MRI. The examination of source estimate results was accomplished by comparing the reasonableness of the source solution to the scalp voltage topographic maps. That is, the source solution had to be spatially coherent, and the orientation of the dominant dipoles had to be consistent with the spatial distribution of the scalp voltage data. If one pattern of spikes with the same voltage distribution is identified, it usually indicates that they emanate from one single source. If several patterns of spikes with different voltage distributions are present, it may indicate different locations of sources (multiple sources), but there are also situations when spikes with different voltage distributions emanate from the same areas (single source). 2.6. Identification of EZ 2.6.1. Semiology We only consider unilateral upper limb automatisms, unilateral blinks, head and/or body version, language alterations, and unilateral dystonic posturing as localizing/lateralizing signs. When the signs contradict, we consider the semiology to represent ‘‘incorrect localization.’’ 2.6.2. cEEG Anatomical location was estimated based on a combination of ictal and interictal EEG findings, with the maximal location of ictal and interictal events defining the EZ location. If locations of ictal and interictal events contradicted, we accepted the results of ictal EEG findings. 2.6.3. MRI The location of the structural lesion that is potentially epileptogenic (such lesions include low-grade glioma, cavernous malformation, and encephalomalacia) was identified in the MRIs by radiologists and epilepsy surgeons. Arachnoid cysts were not taken
R. Feng et al. / Clinical Neurophysiology 127 (2016) 108–116
as epileptogenic lesions in this study. Cases in which MRI data did not show obvious abnormalities were classified as ‘‘incorrect localization’’ by MRI data. 2.6.4. PET The location with minimum of hypometabolism was identified in PET. We did not consider the whole hypometabolic region to be the EZ, but rather only the area with minimal metabolism in the cerebral cortex as well as the mesial temporal structures. 2.6.5. 256-channel dESI The location displaying the maximal current density at 50% (relative to the peak) of the rising slope time point was identified as the EZ. 2.7. Surgical treatment Surgical strategy was made based on the comprehensive analysis of semiology, MRI, cEEG, PET, and 256-channel dESI by three experienced neurosurgeons in our center. Surgical strategy decisions are mainly made before surgery, although intraoperative electrocorticography (EcoG) is usually applied during surgery for validation. 2.8. Evaluation of localization accuracy The neurosurgeons who performed the resections confirmed the resected zones based on postoperative MRI data, and they
111
compared the resected locations to the dESI, PET, and MRI data. Follow-ups were conducted when patients made return visits to Huashan Hospital for consultation or by telephonic interview. Prognosis evaluation was completed according to Engel grading scale (Wieser et al., 2001): Engel class I, seizure-free or free of disabling seizures; class II, rare seizures (less than three seizure days per year); class III, effective (seizure decreased by at least 80%); and class IV, no improvement. Seizures that occurred within the first week after operation were excluded. In patients who achieved ‘‘excellent prognosis’’ (Engel I and II), localizations obtained with different neuroimaging methods that were located within the resection zone were considered correct, and those localizations obtained outside the resection or negative results were considered incorrect. Consistent with prior research (Brodbeck et al., 2011), we define sensitivity as the percentage of patients (from the entire sample) with Engle I and II scores whose localization (based on the different neuroimaging modalities) was within the resected zone. Specificity was defined as the percentage (from the entire sample) of patients with Engle III and IV scores whose localization (again, based on the different neuroimaging modalities) was outside the resected zone. We also evaluated localization accuracy of different localizing methods (dESI, PET, MRI, semiology, and cEEG) on a lobule level. This was accomplished using the following criterion: when the localization results were within the ipsilateral (to the resection) temporal lobe, it was classified as consistent on a lobule level.
Fig. 2. (A) Diagnostic value comparison among multiple tools (criterion 1, dESI, PET, MRI). (B) Diagnostic value comparison among multiple tools (criterion 2, all tools). (C) Kaplan–Meier survival analysis. Survival plot of the probability of postoperative seizure freedom in patients with ‘‘single source’’ dESI results versus patients with ‘‘multiple sources’’ dESI results, with survival as time to first seizure recurrence (months). Statistically, they were significantly different (p < 0.05, log rank). (D) Kaplan–Meier survival analysis. Survival plot of the probability of postoperative seizure freedom in patients whose dESI sources were resected versus patients whose sources were not resected, with survival as time to first seizure recurrence (months). Statistically, they were significantly different (p < 0.05, log rank).
112
R. Feng et al. / Clinical Neurophysiology 127 (2016) 108–116
2.9. Relationship between patterns of 256-channel dESI results and surgical prognosis We used Fisher’s exact test and Kaplan–Meier analysis to analyze the relationship between patterns of 256-channel dESI results and prognosis. Statistical Product and Service Solutions (SPSS) was used for statistical analysis.
freedom in the ‘‘single source’’ group (36 cases) is significantly better than the ‘‘multiple sources’’ group (seven cases) (88.9% vs. 42.9%, log rank, p < 0.05, Fig. 2C). Moreover, the probability of postoperative seizure freedom in ‘‘source(s) within resection’’ group is significantly better than that in ‘‘source(s) outside resection’’ group (94.1% vs. 33.3%, log rank, p < 0.05, Fig. 2D). 4. Discussion
3. Results Altogether, 43 patients with TLE (male = 27) with a median age of 26 (range 9–48) were included in the final analysis. Postoperative follow-up was at least 7 months after surgery (median = 14 months, average = 13.6 months). According to the Engel scale, there were 31 patients classified as Engel I, four patients as Engel II, and eight patients as Engel III or IV. According to both sub-lobule (1) and lobule (2) criteria (Fig. 2A and B), 256-channel dESI showed the best sensitivity (criterion 1, 91.4%; criterion 2, 97.1%) and specificity (criterion 1, 75%; criterion 2, 75%). PET showed poor specificity according to both criteria. However, with criterion 2, PET showed excellent sensitivity (93.5%), which was very close to 256-channel dESI (97.1%) and better than MRI (77.1%), indicating that PET is also a sensitive tool for the presurgical workup of TLE. Prognosis comparison between patients with different dESI source patterns showed significant differences. The Kaplan–Meier survival analysis defining postoperative seizure-free time as ‘‘survival time’’ showed that the probability of postoperative seizure
The 256-channel Geodesic Sensor Net is constructed to include improved electrode coverage over the entire head, including the cheek and the neck area (Fig. 3), with the goal of detecting activity from the basal brain regions (Holmes, 2008; Sparkes et al., 2009). Our study demonstrated that dEEG, even when using an atlas head model, is very sensitive for the presurgical workup of TLE. The dEEG is recorded with a common vertex reference, and re-referenced digitally to various montages for inspection, including the average reference. Because of the improved coverage of the inferior head, the average reference enables the potential at each electrode to be examined in a reference-free manner (Bertrand et al., 1985; Junghofer et al., 1999; Holmes, 2008). The average-referenced dEEG data are examined with the topographic waveform plots, a technique that allows the inspection of spatial distribution of the potential fields, making it easier for EEG technicians to identify epileptic discharges and exclude artifacts. Without appropriate automated technology, spike selection from dEEG can be time-consuming and inefficient. Generally, the selection is based on personal experience, with the confirmation
Fig. 3. (A) A 256-channel Geodesic Sensor Net. Each net is adjusted so that the electrodes over the nasion, bilateral preauricular points, and Cz (priorly marked) were correctly placed. The tension structure of the net assured that the remaining electrodes were evenly distributed on the head and at approximately the same location across different patients. Improved coverage of the cheek and the neck elevated detecting ability of epileptic discharges from the basal brain. (B) Segmentation and averaging of multiple spikes. Topographic map should be used to help judge the morphology of multiple spikes labeled under one Arabic number. The left four topographic maps refer to the segmented spikes before averaging. The right smooth topographic map refers to the averaged spike with the maximum signal-to-noise ratio.
R. Feng et al. / Clinical Neurophysiology 127 (2016) 108–116
113
Fig. 4. An mTLE case. This 30-year-old male suffered from seizure attacks for 8 years. Semiology indicated oral and manual automatism without aura. (A) MRI indicated a lesion at the tail of the left hippocampus; (B) PET showed broad hypometabolism of the lateral and mesial temporal lobes; (C and D) the 256-channel dEEG showed epileptic discharges located in the left temporal region; then, 256-channel dESI showed that sources were at the tail of the hippocampus, which is highly concordant with the lesion itself (source image: the white lines are source orientation vectors, pointing in the positive direction from each voxel center. The crosshairs show the maximal source point); (E and F) viewing the traditional view of the electrodes, one can tell that epileptic discharges from the left cheek electrodes (customized electrodes of Nos. 249, 253, 255, and 244) are more significant than the traditional 10–20 temporal channels (F7, T3, and T5). Follow-up at 14.5 months, the patient is seizure-free. (Chart view and topographic view: 1–30 Hz, 3 cm/s, 10 lV/mm, average reference.)
of IEDs in chart view (traditional view) and topographic view. One excellent characteristic of 256-channel dEEG is that the chart view can enhance the traditional 10–20 view by adding facial electrodes of interest, which makes it more sensitive to detect spikes from the basal temporal region (Fig. 4). In other words, electrodes on the face can sample activities that are generated from the inferior surfaces of the brain. It has been shown that without inferior sampling of the potential field, there is mislocalization of epileptogenic activity from brain regions (Lantz et al., 2003). Prior to source localization, topographic maps are created to examine the voltage distribution of individual spikes to ensure that similar spikes are grouped together such that the average spike represents a homogeneous set (Holmes, 2008). Results of dESI must be comprehensively examined over the entire source space, which can be done efficiently with a flat-map view, and compared against the topographic distribution of the scalp data to ensure that a given source solution reasonably describes the actual spike data. It has been broadly demonstrated that the sources obtained during the rising phase rather than the peak of the spike are more likely to be representative of EZ (Alarcon et al., 1997; Scherg et al.,
1999; Huppertz et al., 2001). We chose the time point of 50% rising phase of spikes for dESI (Holmes, 2008). Source results with a nonsmooth (i.e., irregular) distribution should be questioned due to the ‘‘smoothness’’ constraint imposed by the LORETA method. Nonsmooth solutions may reflect single or multiple bad channels. In such situations, these channels should be excluded from source localization. Additionally, they may also reflect poor signal-tonoise data, in which case regularization of the source estimate should be increased. Compared to conventional tools with respect to localization accuracy, dESI has obvious advantages. Although cEEG and semiology evaluation are fundamental for epilepsy workup, it is usually difficult to define a precise EZ using these data alone. Among nuclear medicine tools, FDG-PET is commonly deemed as one of the most sensitive presurgical workup tools for TLE (Willmann et al., 2007; Bien et al., 2009; Immonen et al., 2010). However, its value is mainly based on the prior diagnosis of epilepsy, and the EZ indicated by FDG-PET is usually beyond the boundaries of the true EZ. Thus, the specificity of FDG-PET is poor but its ability to determine the lateralization of epileptic activity is quite high.
114
R. Feng et al. / Clinical Neurophysiology 127 (2016) 108–116
Fig. 5. A TLE case in which MRI was not localizing. This 37-year-old male suffered from repeated seizure attacks for 5 years. Semiology indicates oral and pharyngeal automatism, and during GTC attack the head turns toward the right side. (A) MRI showed a cystic lesion located in the right hippocampus, while the left temporal lobe presented with no obvious abnormality; (B) PET showed that the left anterior temporal lobe is hypometabolic; (C) the anterior temporal lobectomy was performed; (D) 256channel dEEG demonstrated epileptic discharges mainly located at the left anterior temporal region with the most significant spikes at the cheek electrodes (left upper part of the figure). Sources were located at the left anterior and basal temporal lobes (source image: the white lines are source orientation vectors, pointing in the positive direction from each voxel center. The crosshairs show the maximal source point). Follow-up at 14 months postoperatively showed that the patient was seizure-free. (Topographic view of the 256-channel dEEG: 1–70 Hz, notch 50 Hz, 3 cm/s, 10 lV/mm, average reference montage.)
Structural imaging tools such as MRI are very important for the presurgical workup of epilepsy, although abnormal findings are not always related to epilepsy focus. High-resolution MRI can improve the detection rate of epileptogenic structural lesions, but in fact, as many as 20–30% of epilepsy cases are still classified as MRI-negative (Bien et al., 2009; LoPinto-Khoury et al., 2012). dESI is a promising noninvasive tool with high temporal and spatial resolution. Studies have shown that its clinical yield is high in epilepsy presurgical workup (Michel et al., 2004a; Brodbeck et al., 2011; Yamazaki et al., 2012). Our study also demonstrates that once stable and repeated epileptic spikes are captured following a standardized data processing workflow, dESI is quite accurate, even with an atlas head model, regardless of the presence of potential epileptogenic lesions identified from structural MRI (Fig. 5). Few studies have focused on the 256-channel dEEG waveform features of TLE. In our study, we only used dominant monomorphic IEDs for dESI, despite the fact that multimorphic IEDs are commonly seen in some patients. As expected, multiform spikes frequently produce source results with multiple cortical foci, suggesting the possibility of multifocal epilepsy. According to patterns of dESI results, we divided our cases into ‘‘single source’’ group and ‘‘multiple sources’’ group. The Kaplan–Meier survival analysis showed that the probability of postoperative seizure freedom (88.9%) in the ‘‘single source’’ group is better than in the ‘‘multiple sources’’ group (42.9%, p < 0.05, log rank). Although it is true that a similar viewpoint has long been described, previous studies (Chung et al., 1991; Blume et al., 2001) did not use high-density EEG source imaging to generate precise locations of the ‘‘sources.’’
Because surgical prognosis of multiple-source patients is usually poor, one should be cautious of one-stage resective surgeries for such patients who may be candidates for invasive evaluation, and the sources indicated by dESI could be used as guides for that procedure. On the other hand, cases with monomorphic IEDs may be more likely to fit resective surgeries without invasive evaluation. Presurgical evaluations of epilepsy can divide EZ into symptomatogenic zone, irritative zone, ictal onset zone, epileptic lesion, and functional deficit zone. Each one is, more or less, an index of EZ. IEDs that represent the irritative zone are highly related to EZ, especially when IEDs frequently recur in one focal area (Blume et al., 1993; Blume and Kaibara, 1993; Ebner and Hoppe, 1995; Cascino et al., 1996; Holmes et al., 1996; Javidan, 2012). Evidence from icEEG has shown that when the resection merely includes ictal onset zone while preserving the irritative zone, confirmed by repeated IEDs, seizures tend to be not completely remitted (Bautista et al., 1999; Widdess-Walsh et al., 2007). We divided our cases into ‘‘source(s) within resection’’ group and ‘‘source(s) outside resection’’ group. The Kaplan–Meier survival analysis showed that the possibility of postoperative seizure freedom in the ‘‘source(s) within resection’’ group is better than in the ‘‘source(s) outside resection’’ group (94.1% vs. 33.3%, p < 0.05, log rank), which means the resection of the irritative zone identified by IEDs is positively correlated with good surgical prognosis in TLE. Sufficient spatial sampling of the scalp voltage field (i.e., EEG) is one of the most important factors for accurate ESI (Lantz et al., 2003; Michel et al., 2004b). In fact, it has been shown that
Contents lists available at ScienceDirect
Clinical Neurophysiology journal homepage: www.elsevier.com/locate/clinph
Application of 256-channel dense array electroencephalographic source imaging in presurgical workup of temporal lobe epilepsy Rui Feng a, Jie Hu a,b,⇑, Li Pan a,⇑, Jinsong Wu a, Liqin Lang a, Shize Jiang a, Xin Gu b, Jun Guo b, Liangfu Zhou a a b
Department of Neurosurgery, Huashan Hospital, Fudan University, Shanghai, China Department of Neurosurgery, Jing’an Branch of Huashan Hospital, Fudan University, Shanghai, China
a r t i c l e
i n f o
Article history: Accepted 13 March 2015 Available online 23 March 2015 Keywords: 256-channel dense array electroencephalographic source imaging Temporal lobe epilepsy Positron emission tomography Conventional electroencephalography Magnetic resonance imaging Epileptogenic zone Finite difference model
h i g h l i g h t s 256-channel dense array electroencephalographic source imaging (dESI) showed the best sensitivity
(>90%) and specificity (75%) with comparison to conventional tools. Analysis of the relationship between the 256-channel dESI source patterns and surgical outcome
showed that cases with ‘‘single source’’ had better prognosis than with ‘‘multiple sources.’’ The resection of dESI sources representative of the irritative zone is related to good surgical prognosis
in temporal lobe epilepsy (TLE).
a b s t r a c t Objective: This study evaluated the localization precision of 256-channel dense array electroencephalographic source imaging (dESI) in comparison to conventional noninvasive tools. In addition, the study was designed to analyze the relationship between the 256-channel dESI source patterns and surgical outcome. Methods: Forty-three patients with temporal lobe epilepsy (TLE) who underwent one-stage resective surgeries were recruited in this study. We compared dESI with other noninvasive evaluation methods by comparing results with resections that eliminate or significantly reduced seizures according to sublobule and lobule criteria. Sensitivity and specificity of multiple evaluation methods were calculated. Kaplan–Meier analysis was performed to evaluate the relationship between source patterns and surgical outcome. Results: dESI showed the best sensitivity (sub-lobule, 91.4%; lobule, 97.1%) and specificity (75%) for both sub-lobule and lobule criteria. The Kaplan–Meier survival analysis showed that cases with ‘‘single source’’ had better prognosis than with ‘‘multiple sources’’ (p < 0.05); cases of ‘‘sources within resection’’ showed better surgical prognosis than cases of ‘‘sources outside resection’’ (p < 0.05). Conclusion: In this study, 256-channel dESI provided a higher clinical yield than the other most broadly used noninvasive presurgical workup tools. dESI results with ‘‘single source’’ correlated strongly to good prognosis, while cases with ‘‘multiple sources’’ may cautiously be considered as candidates for one-stage resective surgeries. The resection of the irritative zone identified by interictal epileptiform discharges (IEDs) was related to good surgical prognosis in TLE. Significance: In the presurgical workup of TLE, the clinical yield of 256-channel dESI is high. Patterns of dESI results are related to surgical prognosis, and they can be instructive for presurgical planning. The resection of the irritative zone can be related to good surgical prognosis in TLE. Ó 2015 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
⇑ Corresponding authors at: Department of Neurosurgery, Huashan Hospital, Fudan University, 12 Wulumuqi Zhong Road, Shanghai 200040, China. Tel.: +86 13361913896 (J. Hu), +86 13601683163 (L. Pan). E-mail addresses: [email protected] (J. Hu), [email protected] (L. Pan). http://dx.doi.org/10.1016/j.clinph.2015.03.009 1388-2457/Ó 2015 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
R. Feng et al. / Clinical Neurophysiology 127 (2016) 108–116
109
1. Introduction
2.2. Semiology evaluation and cEEG analysis
Temporal lobe epilepsy (TLE) is the most common type of intractable epilepsy, and it can be divided into three subgroups: hippocampal sclerosis, which constitutes at least 60–70% of TLE (Cascino et al., 1991; Carne et al., 2004), lesional TLE with various kinds of epileptogenic structural lesions (Carne et al., 2004), and TLE without observable abnormalities on structural magnetic resonance imaging (MRI) (MRI-negative TLE, Bien et al., 2009; LoPinto-Khoury et al., 2012). Invasive intracranial electroencephalographic (icEEG) recordings are currently required in clinical practice for the determination of the epileptogenic zone (EZ). Unfortunately, localization accuracy with icEEG is still far from 100%, due to the fact that icEEG electrodes practically cannot be placed to cover the entire brain. Moreover, the invasive procedure is expensive and of high risk. Because of these limitations, noninvasive tools including conventional electroencephalography (cEEG), semiology evaluation, structural MRI, and interictal fluorodeoxyglucose positron emission tomography (FDG-PET) are usually considered for a comprehensive evaluation. Recent studies have shown that dense array EEG source imaging (dESI) can significantly contribute to epilepsy presurgical evaluation (Brodbeck et al., 2011). The present study extends prior dESI for epilepsy evaluation research by comparing EZ localization accuracy, with respect to surgical outcome, of 256-channel dESI performed with an atlas head model against PET, MRI, cEEG, and semiology.
Semiology details were analyzed based on high-definition videos recorded simultaneously with EEG. cEEG examinations always include both waking and sleeping EEG, and they were performed using a 16-channel EEG system (Bio-logic, US). Electrode placement followed the international 10–20 system (Fp1/2, F3/4, F7/8, C3/4, T3/4, T5/6, P3/P4, O1/O2, Fz, Cz, and Oz), and the data were sampled at 512 Hz with a 0.1–100-Hz band-pass filter. The off-line cEEG analysis was performed by two experienced EEG experts.
2. Methods Forty-three patients with TLE were included in this study, which was approved by the ethics committee of Huashan Hospital (2013 Lin Shen No. 286). Informed consent was obtained from each patient. 2.1. Patient selection We selected patients with refractory epilepsy from a pool of patients who were candidates for one-stage resective surgeries. Our comprehensive presurgical workup for the determination of EZ includes semiology evaluation, cEEG analysis, structural MRI, interictal FDG-PET, and dESI. 2.1.1. Inclusion criteria The following were the inclusion criteria: 1. All patients were treated with multiple antiepilepsy drugs (AEDs) for at least 1 year, which yielded poor control of epilepsy. 2. TLE should fundamentally be supported by semiology indicating epilepsy and scalp cEEG, which at least showed interictal epileptic discharges (IEDs). 2.1.2. Exclusion criteria The following were the exclusion criteria: 1. Patients with malignancy-related TLE, TLE with huge epileptogenic lesions, clearly identified bilateral TLE, or multifocal epilepsy. 2. Patients whose noninvasive evaluation did not permit the determination of EZ focus, and who underwent invasive evaluation. A comprehensive presurgical workup in this study for the determination of EZ includes semiology evaluation and cEEG analysis, structural MRI, interictal FDG-PET, and dESI.
2.3. Structural MRI All MRI data were acquired at Huashan Hospital with a 1.5-T MRI (Signa HDx 1.5 T, GE). MRI scanning was performed according to a standardized epilepsy protocol (image size, 512 512; thickness, 4 mm) including coronal T2WI (TR, 5640 ms; TE, 90.2 ms;), coronal T2 FLAIR (TR, 7327 ms; TE, 153 ms), axial T1WI (TR, 760 ms; TE, 14.4 ms), axial T2WI (TR, 3820 ms; TE, 89.5 ms), axial T2 FLAIR (TR, 7427 ms; TE, 160 ms), and sagittal T2 FLAIR (TR, 7327 ms; TE, 159.8 ms). An enhanced T1WI sequence was performed when a tumor was suspected. 2.4. FDG-PET All interictal PET studies were carried out at Huashan Hospital. PET hypometabolism was identified by two experienced physicians who were blind to the patients’ diagnoses. The physicians noted the presence of unilateral, focal decreased FDG uptake in the temporal lobe by visual inspection. 2.5. 256-channel dESI A standardized workflow of 256-channel dEEG data processing was applied in our center (Fig. 1). 2.5.1. 256-channel dEEG acquisition A 256-channel Geodesic Sensor Net (Electrical Geodesics, Inc., Eugene, OR, USA) was selected according to each patient’s head circumference. Each electrode in the net contains a small salinesoaked sponge that contacts the scalp directly. The entire application process can be accomplished within 20 min by experienced technicians, and it requires no skin abrasion or a haircut. The sampling rate was set to 250 Hz. Sensor impedances were kept under 100 KO, which is an acceptable impedance level given the high-input impedance of modern EEG amplifiers (Ferree et al., 2001). Data were recorded for approximately 30–60 min. During dEEG recordings, the patients were fully relaxed and lying on a bed, most of the time with their eyes closed. Usually, patients would be awake during recordings, but if patients fell asleep, the time point would be marked by an EEG technician, who was always present. 2.5.2. IED identification, segmentation, and averaging Monomorphic and repetitive IEDs were identified and reviewed by two experienced EEG reviewers. Once identified, they were grouped according to similar pattern of spatial distribution judged by waveform distribution and topographic maps. There can be either one or several patterns of such spatial distribution for each patient. If a group of IEDs occurred fewer than three times, it was not considered for further dESI analysis. For those IEDs that were kept for analysis, they were segmented (500 ms before and 500 ms after the spike peak) and averaged to derive the average spikes with a good signal-to-noise ratio for dESI.
110
R. Feng et al. / Clinical Neurophysiology 127 (2016) 108–116
Fig. 1. The workflow of 256-channel dESI in Functional Neurosurgery Center, Department of Neurosurgery, Huashan Hospital.
2.5.3. Forward model Standard spherical head models (Berg and Scherg, 1994) account for the scalp, skull, cerebrospinal fluid (CSF), and brain, and conductivities are usually set to 0.3300 S/m (brain), 1 S/m (CSF), 0.0042 S/m (skull), and 0.3300 S/m (scalp). In this study, we employed a realistic atlas head model based on the finite difference method (FDM) (Luu et al., 2014). The FDM allows accurate characterization of cranial orifices to account for their influence on the current flow. Tissue compartments of the FDM model (e.g., scalp, skull, and CSF) were constructed from whole-head MRI and CT scans of a single subject (Colin27) whose head geometry closely matches the Montreal Neurological Institute (MNI) average MRI (MNI305) (Luu et al., 2014). The white and gray matter were derived from the probabilistic partitioning of the MNI305 average brain (http://www.bic.mni.mcgill.ca), and registered to the individual Colin27 head. Source locations (a total of 2447 sources) were evenly seeded in the gray matter volume of the MNI305 average brain. Conductivity values used in the FDM model were set to 0.25 S/m (brain), 1.8 S/m (CSF), 0.018 S/m (skull), and 0.44 S/m (scalp). According to several recent studies, the conductivity ratio of skull to brain tissue is close to 1:14, rather than the traditionally assumed 1:80 value (Ferree et al., 2000; Ryynanen et al., 2006; Zhang et al., 2006). To generate the lead field data (i.e., the description of the current propagation from a dipole position to the scalp, where the EEG is recorded), the recording location (i.e., EEG sensor positions on the scalp) must be identified. Because we did not acquire sensor locations for each individual patient, an average (based on five adult subjects) of the 256-channel location (Luu P, Ferree T. Unpublished manuscript; 2005. Determination of the HydroCel Geodesic Sensor Nets’ Average Electrode Positions and Their 10–10 International Equivalents) was used to register with the realistic atlas head model. 2.5.4. dEEG source localization dESI was performed based on the FDM model using the low-resolution electromagnetic tomography (LORETA) method (Pascual-
Marqui et al., 1994). Sources of the spikes were determined at 50% (relative to the peak) of the rising slope of the spike (Holmes, 2008), and the results are presented on the atlas MRI. The examination of source estimate results was accomplished by comparing the reasonableness of the source solution to the scalp voltage topographic maps. That is, the source solution had to be spatially coherent, and the orientation of the dominant dipoles had to be consistent with the spatial distribution of the scalp voltage data. If one pattern of spikes with the same voltage distribution is identified, it usually indicates that they emanate from one single source. If several patterns of spikes with different voltage distributions are present, it may indicate different locations of sources (multiple sources), but there are also situations when spikes with different voltage distributions emanate from the same areas (single source). 2.6. Identification of EZ 2.6.1. Semiology We only consider unilateral upper limb automatisms, unilateral blinks, head and/or body version, language alterations, and unilateral dystonic posturing as localizing/lateralizing signs. When the signs contradict, we consider the semiology to represent ‘‘incorrect localization.’’ 2.6.2. cEEG Anatomical location was estimated based on a combination of ictal and interictal EEG findings, with the maximal location of ictal and interictal events defining the EZ location. If locations of ictal and interictal events contradicted, we accepted the results of ictal EEG findings. 2.6.3. MRI The location of the structural lesion that is potentially epileptogenic (such lesions include low-grade glioma, cavernous malformation, and encephalomalacia) was identified in the MRIs by radiologists and epilepsy surgeons. Arachnoid cysts were not taken
R. Feng et al. / Clinical Neurophysiology 127 (2016) 108–116
as epileptogenic lesions in this study. Cases in which MRI data did not show obvious abnormalities were classified as ‘‘incorrect localization’’ by MRI data. 2.6.4. PET The location with minimum of hypometabolism was identified in PET. We did not consider the whole hypometabolic region to be the EZ, but rather only the area with minimal metabolism in the cerebral cortex as well as the mesial temporal structures. 2.6.5. 256-channel dESI The location displaying the maximal current density at 50% (relative to the peak) of the rising slope time point was identified as the EZ. 2.7. Surgical treatment Surgical strategy was made based on the comprehensive analysis of semiology, MRI, cEEG, PET, and 256-channel dESI by three experienced neurosurgeons in our center. Surgical strategy decisions are mainly made before surgery, although intraoperative electrocorticography (EcoG) is usually applied during surgery for validation. 2.8. Evaluation of localization accuracy The neurosurgeons who performed the resections confirmed the resected zones based on postoperative MRI data, and they
111
compared the resected locations to the dESI, PET, and MRI data. Follow-ups were conducted when patients made return visits to Huashan Hospital for consultation or by telephonic interview. Prognosis evaluation was completed according to Engel grading scale (Wieser et al., 2001): Engel class I, seizure-free or free of disabling seizures; class II, rare seizures (less than three seizure days per year); class III, effective (seizure decreased by at least 80%); and class IV, no improvement. Seizures that occurred within the first week after operation were excluded. In patients who achieved ‘‘excellent prognosis’’ (Engel I and II), localizations obtained with different neuroimaging methods that were located within the resection zone were considered correct, and those localizations obtained outside the resection or negative results were considered incorrect. Consistent with prior research (Brodbeck et al., 2011), we define sensitivity as the percentage of patients (from the entire sample) with Engle I and II scores whose localization (based on the different neuroimaging modalities) was within the resected zone. Specificity was defined as the percentage (from the entire sample) of patients with Engle III and IV scores whose localization (again, based on the different neuroimaging modalities) was outside the resected zone. We also evaluated localization accuracy of different localizing methods (dESI, PET, MRI, semiology, and cEEG) on a lobule level. This was accomplished using the following criterion: when the localization results were within the ipsilateral (to the resection) temporal lobe, it was classified as consistent on a lobule level.
Fig. 2. (A) Diagnostic value comparison among multiple tools (criterion 1, dESI, PET, MRI). (B) Diagnostic value comparison among multiple tools (criterion 2, all tools). (C) Kaplan–Meier survival analysis. Survival plot of the probability of postoperative seizure freedom in patients with ‘‘single source’’ dESI results versus patients with ‘‘multiple sources’’ dESI results, with survival as time to first seizure recurrence (months). Statistically, they were significantly different (p < 0.05, log rank). (D) Kaplan–Meier survival analysis. Survival plot of the probability of postoperative seizure freedom in patients whose dESI sources were resected versus patients whose sources were not resected, with survival as time to first seizure recurrence (months). Statistically, they were significantly different (p < 0.05, log rank).
112
R. Feng et al. / Clinical Neurophysiology 127 (2016) 108–116
2.9. Relationship between patterns of 256-channel dESI results and surgical prognosis We used Fisher’s exact test and Kaplan–Meier analysis to analyze the relationship between patterns of 256-channel dESI results and prognosis. Statistical Product and Service Solutions (SPSS) was used for statistical analysis.
freedom in the ‘‘single source’’ group (36 cases) is significantly better than the ‘‘multiple sources’’ group (seven cases) (88.9% vs. 42.9%, log rank, p < 0.05, Fig. 2C). Moreover, the probability of postoperative seizure freedom in ‘‘source(s) within resection’’ group is significantly better than that in ‘‘source(s) outside resection’’ group (94.1% vs. 33.3%, log rank, p < 0.05, Fig. 2D). 4. Discussion
3. Results Altogether, 43 patients with TLE (male = 27) with a median age of 26 (range 9–48) were included in the final analysis. Postoperative follow-up was at least 7 months after surgery (median = 14 months, average = 13.6 months). According to the Engel scale, there were 31 patients classified as Engel I, four patients as Engel II, and eight patients as Engel III or IV. According to both sub-lobule (1) and lobule (2) criteria (Fig. 2A and B), 256-channel dESI showed the best sensitivity (criterion 1, 91.4%; criterion 2, 97.1%) and specificity (criterion 1, 75%; criterion 2, 75%). PET showed poor specificity according to both criteria. However, with criterion 2, PET showed excellent sensitivity (93.5%), which was very close to 256-channel dESI (97.1%) and better than MRI (77.1%), indicating that PET is also a sensitive tool for the presurgical workup of TLE. Prognosis comparison between patients with different dESI source patterns showed significant differences. The Kaplan–Meier survival analysis defining postoperative seizure-free time as ‘‘survival time’’ showed that the probability of postoperative seizure
The 256-channel Geodesic Sensor Net is constructed to include improved electrode coverage over the entire head, including the cheek and the neck area (Fig. 3), with the goal of detecting activity from the basal brain regions (Holmes, 2008; Sparkes et al., 2009). Our study demonstrated that dEEG, even when using an atlas head model, is very sensitive for the presurgical workup of TLE. The dEEG is recorded with a common vertex reference, and re-referenced digitally to various montages for inspection, including the average reference. Because of the improved coverage of the inferior head, the average reference enables the potential at each electrode to be examined in a reference-free manner (Bertrand et al., 1985; Junghofer et al., 1999; Holmes, 2008). The average-referenced dEEG data are examined with the topographic waveform plots, a technique that allows the inspection of spatial distribution of the potential fields, making it easier for EEG technicians to identify epileptic discharges and exclude artifacts. Without appropriate automated technology, spike selection from dEEG can be time-consuming and inefficient. Generally, the selection is based on personal experience, with the confirmation
Fig. 3. (A) A 256-channel Geodesic Sensor Net. Each net is adjusted so that the electrodes over the nasion, bilateral preauricular points, and Cz (priorly marked) were correctly placed. The tension structure of the net assured that the remaining electrodes were evenly distributed on the head and at approximately the same location across different patients. Improved coverage of the cheek and the neck elevated detecting ability of epileptic discharges from the basal brain. (B) Segmentation and averaging of multiple spikes. Topographic map should be used to help judge the morphology of multiple spikes labeled under one Arabic number. The left four topographic maps refer to the segmented spikes before averaging. The right smooth topographic map refers to the averaged spike with the maximum signal-to-noise ratio.
R. Feng et al. / Clinical Neurophysiology 127 (2016) 108–116
113
Fig. 4. An mTLE case. This 30-year-old male suffered from seizure attacks for 8 years. Semiology indicated oral and manual automatism without aura. (A) MRI indicated a lesion at the tail of the left hippocampus; (B) PET showed broad hypometabolism of the lateral and mesial temporal lobes; (C and D) the 256-channel dEEG showed epileptic discharges located in the left temporal region; then, 256-channel dESI showed that sources were at the tail of the hippocampus, which is highly concordant with the lesion itself (source image: the white lines are source orientation vectors, pointing in the positive direction from each voxel center. The crosshairs show the maximal source point); (E and F) viewing the traditional view of the electrodes, one can tell that epileptic discharges from the left cheek electrodes (customized electrodes of Nos. 249, 253, 255, and 244) are more significant than the traditional 10–20 temporal channels (F7, T3, and T5). Follow-up at 14.5 months, the patient is seizure-free. (Chart view and topographic view: 1–30 Hz, 3 cm/s, 10 lV/mm, average reference.)
of IEDs in chart view (traditional view) and topographic view. One excellent characteristic of 256-channel dEEG is that the chart view can enhance the traditional 10–20 view by adding facial electrodes of interest, which makes it more sensitive to detect spikes from the basal temporal region (Fig. 4). In other words, electrodes on the face can sample activities that are generated from the inferior surfaces of the brain. It has been shown that without inferior sampling of the potential field, there is mislocalization of epileptogenic activity from brain regions (Lantz et al., 2003). Prior to source localization, topographic maps are created to examine the voltage distribution of individual spikes to ensure that similar spikes are grouped together such that the average spike represents a homogeneous set (Holmes, 2008). Results of dESI must be comprehensively examined over the entire source space, which can be done efficiently with a flat-map view, and compared against the topographic distribution of the scalp data to ensure that a given source solution reasonably describes the actual spike data. It has been broadly demonstrated that the sources obtained during the rising phase rather than the peak of the spike are more likely to be representative of EZ (Alarcon et al., 1997; Scherg et al.,
1999; Huppertz et al., 2001). We chose the time point of 50% rising phase of spikes for dESI (Holmes, 2008). Source results with a nonsmooth (i.e., irregular) distribution should be questioned due to the ‘‘smoothness’’ constraint imposed by the LORETA method. Nonsmooth solutions may reflect single or multiple bad channels. In such situations, these channels should be excluded from source localization. Additionally, they may also reflect poor signal-tonoise data, in which case regularization of the source estimate should be increased. Compared to conventional tools with respect to localization accuracy, dESI has obvious advantages. Although cEEG and semiology evaluation are fundamental for epilepsy workup, it is usually difficult to define a precise EZ using these data alone. Among nuclear medicine tools, FDG-PET is commonly deemed as one of the most sensitive presurgical workup tools for TLE (Willmann et al., 2007; Bien et al., 2009; Immonen et al., 2010). However, its value is mainly based on the prior diagnosis of epilepsy, and the EZ indicated by FDG-PET is usually beyond the boundaries of the true EZ. Thus, the specificity of FDG-PET is poor but its ability to determine the lateralization of epileptic activity is quite high.
114
R. Feng et al. / Clinical Neurophysiology 127 (2016) 108–116
Fig. 5. A TLE case in which MRI was not localizing. This 37-year-old male suffered from repeated seizure attacks for 5 years. Semiology indicates oral and pharyngeal automatism, and during GTC attack the head turns toward the right side. (A) MRI showed a cystic lesion located in the right hippocampus, while the left temporal lobe presented with no obvious abnormality; (B) PET showed that the left anterior temporal lobe is hypometabolic; (C) the anterior temporal lobectomy was performed; (D) 256channel dEEG demonstrated epileptic discharges mainly located at the left anterior temporal region with the most significant spikes at the cheek electrodes (left upper part of the figure). Sources were located at the left anterior and basal temporal lobes (source image: the white lines are source orientation vectors, pointing in the positive direction from each voxel center. The crosshairs show the maximal source point). Follow-up at 14 months postoperatively showed that the patient was seizure-free. (Topographic view of the 256-channel dEEG: 1–70 Hz, notch 50 Hz, 3 cm/s, 10 lV/mm, average reference montage.)
Structural imaging tools such as MRI are very important for the presurgical workup of epilepsy, although abnormal findings are not always related to epilepsy focus. High-resolution MRI can improve the detection rate of epileptogenic structural lesions, but in fact, as many as 20–30% of epilepsy cases are still classified as MRI-negative (Bien et al., 2009; LoPinto-Khoury et al., 2012). dESI is a promising noninvasive tool with high temporal and spatial resolution. Studies have shown that its clinical yield is high in epilepsy presurgical workup (Michel et al., 2004a; Brodbeck et al., 2011; Yamazaki et al., 2012). Our study also demonstrates that once stable and repeated epileptic spikes are captured following a standardized data processing workflow, dESI is quite accurate, even with an atlas head model, regardless of the presence of potential epileptogenic lesions identified from structural MRI (Fig. 5). Few studies have focused on the 256-channel dEEG waveform features of TLE. In our study, we only used dominant monomorphic IEDs for dESI, despite the fact that multimorphic IEDs are commonly seen in some patients. As expected, multiform spikes frequently produce source results with multiple cortical foci, suggesting the possibility of multifocal epilepsy. According to patterns of dESI results, we divided our cases into ‘‘single source’’ group and ‘‘multiple sources’’ group. The Kaplan–Meier survival analysis showed that the probability of postoperative seizure freedom (88.9%) in the ‘‘single source’’ group is better than in the ‘‘multiple sources’’ group (42.9%, p < 0.05, log rank). Although it is true that a similar viewpoint has long been described, previous studies (Chung et al., 1991; Blume et al., 2001) did not use high-density EEG source imaging to generate precise locations of the ‘‘sources.’’
Because surgical prognosis of multiple-source patients is usually poor, one should be cautious of one-stage resective surgeries for such patients who may be candidates for invasive evaluation, and the sources indicated by dESI could be used as guides for that procedure. On the other hand, cases with monomorphic IEDs may be more likely to fit resective surgeries without invasive evaluation. Presurgical evaluations of epilepsy can divide EZ into symptomatogenic zone, irritative zone, ictal onset zone, epileptic lesion, and functional deficit zone. Each one is, more or less, an index of EZ. IEDs that represent the irritative zone are highly related to EZ, especially when IEDs frequently recur in one focal area (Blume et al., 1993; Blume and Kaibara, 1993; Ebner and Hoppe, 1995; Cascino et al., 1996; Holmes et al., 1996; Javidan, 2012). Evidence from icEEG has shown that when the resection merely includes ictal onset zone while preserving the irritative zone, confirmed by repeated IEDs, seizures tend to be not completely remitted (Bautista et al., 1999; Widdess-Walsh et al., 2007). We divided our cases into ‘‘source(s) within resection’’ group and ‘‘source(s) outside resection’’ group. The Kaplan–Meier survival analysis showed that the possibility of postoperative seizure freedom in the ‘‘source(s) within resection’’ group is better than in the ‘‘source(s) outside resection’’ group (94.1% vs. 33.3%, p < 0.05, log rank), which means the resection of the irritative zone identified by IEDs is positively correlated with good surgical prognosis in TLE. Sufficient spatial sampling of the scalp voltage field (i.e., EEG) is one of the most important factors for accurate ESI (Lantz et al., 2003; Michel et al., 2004b). In fact, it has been shown that
