|Home | About | Journals | Submit | Contact Us | Français|
The role of sharps and spikes, interictal epileptiform discharges (IEDs), in guiding epilepsy surgery in children remains controversial, particularly with intracranial EEG (IEEG). While ictal recording is the mainstay of localizing epileptic networks for surgical resection, current practice dictates removing regions generating frequent IEDs if they are near the ictal onset zone. Indeed, past studies suggest an inconsistent relationship between IED and seizure onset location, though these studies were based upon relatively short EEG epochs.
We employ a previously validated, computerized spike detector, to measure and localize IED activity over prolonged, representative segments of IEEG recorded from 19 children with intractable, mostly extra temporal lobe epilepsy. Approximately 8 hours of IEEG, randomly selected thirty-minute segments of continuous interictal IEEG per patient were analyzed over all intracranial electrode contacts.
When spike frequency was averaged over the 16-time segments, electrodes with the highest mean spike frequency were found to be within the seizure onset region in 11 of 19 patients. There was significant variability between individual 30-minute segments in these patients, indicating that large statistical samples of interictal activity were required for improved localization. Low voltage fast EEG at seizure onset was the only clinical factor predicting IED localization to the seizure onset region.
Our data suggest that automated IED detection over multiple representative samples of IEEG may be of utility in planning epilepsy surgery for children with intractable epilepsy. Further research is required to better determine which patients may benefit from this technique a priori.
Epilepsy surgery is performed in patients with medically refractory epilepsy and evidence of localized or regional seizure onset on non-invasive preoperative testing. Intracranial EEG monitoring or intraoperative electrocorticography (ECOG) are often performed to tailor resections, especially those patients with extra-temporal epilepsy. Both techniques have inherent advantages and disadvantages. Intracranial electrode implantations have increased morbidity, do not always capture seizures or improve localization. ECOG typically does not capture seizures, can be affected by type and depth of anesthesia, and by artifact in the operating room. The limitations of both techniques may reduce post-operative seizure freedom and new methods to improve presurgical localization of epileptic networks are greatly needed. This study analyzes interictal spike occurrence over prolonged epochs of IEEG to determine if quantitative analysis can improve localization of the seizure onset zone prior to surgery. If successful, this technique may decrease morbidity, hospital stay, and potentially improve outcome in patients undergoing epilepsy surgery and also might be extrapolated to patients undergoing ECOG during surgical resection.
Interictal epileptiform discharges have been hypothesized to play a role in seizure generation, however, their relationship to seizure onset has been difficult to define (see Staley, et al., 2005 for review). The localization value of IEDs may relate to the modality used to record this activity. Scalp EEG is the most imprecise method for localizing IEDs, due to its comparatively poor spatial resolution. Magnetoencephalography has much better spatial resolution for surface sources but signals beyond a depth of a centimeter fall off rapidly. Consequently, epileptic networks involving deep structures, such as the hippocampus, are not well “imaged” by MEG. Still, both of these non-invasive techniques can be used to tailor epilepsy surgical resections. Electrocorticography (ECOG) provides much higher spatial resolution and good quality signals, however, the technique is limited by restricted surgical time, confounding of anesthesia, and frank seizures are rarely captured, except in severe cases. Therefore it is important to understand how frequently IEDs identify the area of seizure onset to improve surgical outcome.
Several studies utilizing both long-term IEEG and ECOG report that resections incorporating a “significant” portion of regions generating interictal spikes and sharp waves, in addition to the ictal onset zone, improves seizure freedom postoperatively (Wyllie, et al., 1987, Paolicchi, et al., 2000, Krsek, et al., 2008). Regions of spiking on subdural and depth electrodes usually extend beyond the seizure onset zone, and in some reports, appear to represent multiple distinct independent spiking foci (Hufnagel, et al., 2000, Kobayashi, et al., 2001). One study found that electrodes with the highest frequency of interictal spiking were within the seizure onset zone 100% of the time, but in another study frequency of interictal spiking could only identify the seizure onset zone in ~50% of the patients (Hufnagel, et al., 2000, Asano, et al., 2003). The reason for this variability is not clear, though there were differences in methodologies and predominant pathology. Both groups analyzed relatively brief periods of IEEG, which could have under-sampled IED distributions. These studies suggest that spiking is part of the dysfunctional “epileptogenic zone,” but further studies are needed to determine if there is a quantitative relationship between IEDs, seizure onset and surgical outcome.
Prior studies have identified sleep state, seizures and medication withdrawal as variables that alter the frequency and to some extent the regional localization of IEDs (Gotman & Marciani, 1985, Marciani, et al., 1985, Gotman & Koffler, 1989, Sammaritano, et al., 1991, Spencer, et al., 2008). The extent of variability in discharge frequency and localization may vary between temporal and extra-temporal IEEG especially over longer periods of IEEG (Spencer, et al., 2008). Investigators analyzing recordings from human temporal lobe epilepsy, for the most part, have not been able to identify changes in IED activity prior to a seizure (Lieb, et al., 1978, Lange, et al., 1983). Mixed results on changes in IEDs following medication withdrawal have been reported and several studies have reported decreases in spike frequency post seizure (Gotman & Marciani, 1985, Gotman & Koffler, 1989, Spencer, et al., 2008). The variability in IED frequency and spatial extent with sleep state and seizure activity suggest that there are fluctuations in this activity over time, though the mechanism for the variability is not well understood. For example, during non-REM sleep there is not only an increase in IEDs, but also an extension of the spiking area (Sammaritano, et al., 1991, Staba, et al., 2002).
Here we have set out to map the frequency of IEDs in long-term IEEG recordings in a group of children with intractable epilepsy, with the goal of determining if the location of frequent IEDs is a surrogate marker for seizure onset.
The Children’s Hospital of Philadelphia (CHOP) institutional review board approved this study. IEEG recordings were obtained prospectively from 30 patients undergoing subdural electrode implantation for epilepsy surgery between 2003–2007. All IEEGs were recorded with Grass-Telefactor 128 Electrode CTE EEG machine using 16-bit amplifiers (Astro Med Corp., West Warwick, RI) sampled at 200 samples/second/electrode. Analog anti-aliasing bandpass filter (frequency cut-offs at 0.1- and 70-Hz) and notch filter (null at 60 Hz) were used for signal conditioning. Recordings were reviewed in a referential montage, and marked by two reviewers to identify seizure onset times, morphology and electrode locations.
All clinical data, including seizure type, pre-surgical and post-surgical data were obtained by retrospective chart review. The primary neurologist’s seizure outcome assessment was graded using a modified Engle scale at last patient contact (Engel, et al., 1993).
From 30 consecutively consented patients, 19 patient’s IEEGs were selected for the current study. Patients were chosen using the following criteria: detailed IEEG seizure markings were present, recordings were continuous and predominantly artifact free, and inter-ictal epileptiform activity was present. If any of the criteria were not met, the patient was excluded from the study. Of the patients who did not meet criteria, incomplete recordings and excessive artifact were the reasons for exclusion in 6 patients. Lack of or very rare IEDs excluded 5 patients.
Two clinical epileptologists marked seizure onset times and locations. Clinical reports were consulted, but the entire recording was reviewed for the presence of seizures. All electrographic seizures were marked for times of unequivocal electrical seizure onset and earliest electrical change (Litt, et al., 2001). Final consensus between the two IEEG readers established the electrodes and time markings for seizure onsets were used. All electrodes determined to be involved at the unequivocal electrographic onset were included in the seizure onset zone calculations. A description of the electrographic onset was recorded as either: attenuated with low voltage fast activity, repetitive spiking, or rhythmic slow wave onsets.
An automated detector developed by our group, implemented in MATLAB (Natick, MA) and previously validated against human markings was used (Brown, et al., 2007). Briefly, the detector is mimetic-based and models IEDs as deflections in the IEEG satisfying three criteria: (i) a peak amplitude greater than a threshold value based on the background iEEG, (ii) a characteristic width of the spike component lasting between 30–300 ms, and (iii) an after going slow wave with a duration of 100 – 500 ms.
For this study the detector was “tuned” for each patients IEDs by first selecting representative IEEG epochs containing electrodes with both characteristic epileptiform activity and no IEDs. The detector’s output was plotted against the IEEG and reviewed, independently, by two epileptologists for the threshold settings with the highest true positive and lowest false positive detections. Reviewers had to qualitatively agree with the spike detections (approximately 50% accuracy was used as this is the previously quantified accuracy of the detector) or the detector thresholds were recalculated. Thus with our current spike detection technology we will miss and overcall spikes. Once a threshold was selected, the detector analyzed the full data set (described below). In addition, the quantitative detection maps generated by collating detector outputs by channel were reviewed and compared to the pattern of detection with the epileptologist’s interpretation of the most active spike foci.
The full duration of each patient’s IEEG recording was divided into 30-minute segments. A MATLAB random number algorithm reordered the segments with the first 16 being reviewed. If the randomly chosen segment contained a seizure or was interrupted by file change the next segment in the list was used. After the original segment selection was completed, three segments, determined to be unusable due to artifacts, were discarded. This resulted in 3 patients having only 15, 30-minute segments for analysis.
The selected segments were analyzed using the predetermined thresholds and the output sorted by electrode and segment. Mean, Range, Variance, and Standard Deviation of spikes per 30 minutes were calculated. The mean spike density (spikes/30min/electrode) was plotted in X-Y coordinates. The entire process is shown in Figure 1. The location of each electrode was extracted from surgical photographs, schematic drawings, and post-implant MRI when available. Mean IED densities were graphed on the X-Y coordinate plots of the grids.
All graphs were qualitatively assessed for correlation between maximal IED density and seizure onset. Quantitative assessment was performed using a t-test to compare mean spike values in channels at or away from seizure onset location. All calculations were performed in MATLAB using the statistical toolbox. The mean spike density in the seizure onset electrodes versus electrodes not part of the seizure onset was analyzed. Our null hypothesis was no difference in mean spiking between seizure onset and other electrodes. Significance and “tails” were set to the default values of 0.05 and ‘both’.
Intra-patient segmental variation was tested in three ways: First, the t-test was calculated for each thirty-minute segment individually to test for variability of seizure onset with spike density across segments for a single patient. Second, a segment-versus-segment test of differences within each patient was performed using an F-test to compare the total variance of all electrodes to each segmental variance, and a null hypothesis that no difference in variance existed across segments. Third, an electrode-by-electrode F-test statistic compared the total variance against the segmental mean variance. A single sided F-test was used, as we were only interested if the electrode variance was greater than the total variance. This method allowed us to test the hypothesis that spiking in a subset of the electrodes varied more than the overall variability.
Clinical data was analyzed for differences in patients with and without spike density and seizure onset correlation using the Fisher’s exact T test for seizure outcome, pathology and seizure onset morphology.
Nineteen patients, 8 female, aged 1 to 20 years (mean 12 years) at time of epilepsy surgery were analyzed. Electrode coverage is described in Table 1, with most patients having multi-lobar coverage. Cortical dysplasia Palmini type 2a was the pathological diagnosis in 14 of the 18 patients who underwent resection (Palmini, et al., 2004). Two patients had Palmini type 2b dysplasia, one patient had evidence of hemorrhage, and one a chronic infarction. Post-operative seizure freedom occurred in 10 patients (Engel class I) while 5 others were felt to demonstrate significant seizure improvement (Engel Class III; See Table 1 for details).
Seizure onset electrodes for all patients ranged between 3 and 102 contacts with a median of 13 involved electrodes. This was 4 to 82% of the total electrodes on any given patient. The types of seizure onsets on IEEG were rhythmic spiking in 12 patients; low voltage attenuation in 5 patients, and 1 patient had a mixture of seizures with rhythmic spiking and low voltage attenuation with fast activity. Only one patient had rhythmic slowing at seizure onset.
The 15 or 16 randomly selected 30 minute IEEG sections covered wake and sleep segments as well as pre-ictal and post-ictal recording states. Seven and one half or eight hours of total IEEG was analyzed per patient, comprising between 2.5 and 16.7 percent of each patients total recording. Analysis of IED frequency per 30-minute segment per electrode was calculated and compared to the location of the seizure onset zone.
Mean spike densities (spikes/30 minutes/electrode) across all processed data epochs for each patient were compared to that patient’s seizure onset zone (Figs 2 and and3).3). Qualitative assessment by two IEEG readers using the color-coded maps revealed 10 of 19 patients had overlap between the electrode with highest spike density and seizure onset electrodes (Fig 2). A t-test was calculated to quantify overlap between spike density and seizure onset. Eleven of the 19 patients (57.89%) had mean spike densities in the seizure onset channels significantly different than all other channels (p < 0.05; Table 2). Four additional patients’ results trended toward significant correlation between spike density and location of the seizure onset zone. Of the 8 patients for whom IED density was not statistically different in seizure onset and all other channel locations, 4 of 8 appeared to have seizure onsets that surrounded or abutted the region of highest spike density, while the remaining 4 patients appeared to have no pattern of overlap or correlation between the two variables (Fig 3). Overall the majority of patients have mean spike counts that are highest in the seizure onset region.
We assessed if IED correlation with seizure onset electrode contacts remained constant across all the segments (Fig 4). Two tests were performed. First, the t-test was performed on each data segment individually. In the 11 patients describe above with significant correlation between IEDs and seizure onset zone, up to 11 (range 0–11) of the 16 individual segments p-value was greater than 0.05 (Supplementary Table 1). In patients with p-values (p < 0.001), there were fewer segments (range 0 to 6) that had p values greater than 0.05. For patients whose 16-segment average showed no significant correlation (p >0.05) there also was variability between segments, with zero to six segments having significant correlation (p < 0.05) between IED density and seizure onset zone location. Only 1 of the 19 patients had no segments in which IED density did not correlate with seizure onset zone location. Overall, 47.35% of the 302 IEEG epochs analyzed had IED densities that correlated with seizure onset zones. Therefore, increased data sampling/averaging significantly improves the likelihood that IED distribution correlates with seizure onset zone location.
To further assess the extent of variability between segments, an F-test was performed to determine if the variance of IED density in each segment differed significantly from the mean variance of the grouped data epochs in each patient (Supplementary Table 2). In almost all patients this method found significant variability between epochs with only one patient’s data epochs demonstrating little variance from the mean. The other patients had between 5 and 13 segments whose individual epoch variances differed from the mean variance. Taken together these data suggest that there are significant differences in IED density localization between segments.
Among all patients the total number of IEDs per electrode varied from 0 to 5997 per 30-minute segment (a maximum of 3.33 IEDs/second/electrode were measured). The mean IED frequency across all epochs per electrode ranged from 15.93 to 1092.12 discharges/electrode/30 minutes (Table 3). Because of this large variability, we asked if total IED number influenced the correlation between IED density and seizure onset. A linear regression analysis for the mean total IEDs per electrode and the 16-segment t test p-values did not produce a significant correlation (p=0.33, Table 3). Comparison of the 16 segment p-values and either the range or the standard deviation of IEDs also did not correlate (p= 0.83 and 0.98 respectively, Table 3). Finally, a linear regression was performed for each patient, looking for a correlation between the number of total IEDs in a segment and the t-test p value of that segment (Supplementary Table 3). In 13 of the 19 (68.42%) patients the number of IEDs in a segment had no relationship to the p value between seizure onset and IED density. For the six patients in whom there was a positive correlation, three demonstrated that a larger number of IEDs in a segment was associated with better agreement between IED generator and seizure onset zone, and the opposite held true in three.
We performed an F-test to determine if IED counts per electrode correlated with a global increase or decrease in IEDs in specific data epochs. A F-test compared the variance of each electrode to the patient’s mean variance, highlighting variability in number of IEDs for a particular electrode compared to an increase or decrease in total IEDs for that segment. Between zero and 61 electrodes differed in the number of IEDs to a greater extent than the overall IED variance for a segment (i.e. F-test P<0.01 due to larger variability in an electrode compared to the entire segment; Supplementary Table 4). Three patients had no electrodes and another 6/19 had fewer than 10% of electrodes with IED number varying more than the overall number of IEDs within epochs. However, the percentage of electrodes with IED variability greater than the overall variance did not distinguish patients with IED density correlation to seizure onset (15.7% ± 5%SEM) to those without IED density correlation (6.7% ± 5%SEM). Together these data demonstrate that IED frequency change over time in a given patient.
To determine if, patients with and without IED density correlation to seizure onset zone differed clinically, we performed a Fisher exact test comparing the two groups for surgical outcome, pathology, seizure onset morphology on IEEG. There was no significant difference in post-operative outcomes (P=0.63), or pathology IIa versus other (IIb, hemorrhage or infarction) (P=0.58) between patients with and without IED correlation to seizure onset zone. Interestingly while only 6 patients had low voltage attenuation as the first manifestation of their seizures, all 6 had correlation between frequency of IED and seizure onset electrodes. In contrast patients with spiking or slowing at seizure onset were mixed with some demonstrating a correlation between high IED frequency and seizure onset location (n=7) and others no correlation (n=6) (Fisher exact T test P<0.05).
Utilizing a computerized spike detection system to analyze long periods of IEEG we identified correlation between interictal IED location and the seizure onset zone in 11 of 19 patients. In the other 8 patients, the mean distribution of interictal IEDs was a poor marker for identifying the seizure onset region. In addition, when comparing across sixteen different 30-minute segments of IEEG for each individual patient, those with and without overall correlation, IED density varied significantly between sections for all but one patient. Thus in the majority of the patients, brief recordings of IEDs are unlikely to be useful in localizing seizure onset.
Prior studies have utilized short IEEG epochs to determine if IEDs can serve as a marker to identify seizure onset electrodes (Hufnagel, et al., 2000, Asano, et al., 2003). Asano et al. 2003 studied a patient population similar to our own, children mostly with cortical dysplasia, and found that the electrode with the highest IED frequency was contained within the seizure onset region in 13 of 13 patients. IED frequency had the best correlation, though amplitude and leading spike were almost as good at identifying seizure onset electrodes. In contrast, only 11 of 19 of our patients had the highest IED frequency within the seizure onset zone. Our findings are in keeping with a larger study in adults and children also using a computer based spike detector were the seizure onset electrode was within 2cm of the maximal spike frequency electrode in only 53% of patients (Hufnagel, et al., 2000).
Interictal IEDs are clinically used in a variety of ways to help identify the region of surgical resection. Our data suggest that electrodes with the highest frequency of IEDs over long-periods of IEEG correlate with the electrodes involved in the seizure onset in about two-thirds of pediatric patients with medically refractory epilepsy. When only a single 30-minute segment of recording is used, there is less consistent correlation between the electrodes generating IEDs and the seizure onset zone. Indeed, in all but one patient, the IED density pattern in a random 30-minute segment did not consistently correlate with seizure onset zone. These data suggest that the use of short epochs of subdural electrode recordings to identify regions of highest spike density is generally insufficient to provide an accurate localization of IED density. However, since at least a single segment in all but one patient correlated with seizure onset, picking the correct time to quantify seizure density could help delineate the seizure onset zone.
From a clinical perspective, can we use our findings to improve seizure free outcomes following surgery? Previous reports, utilizing IEEG, describe increased surgical success if both the area of “prominent interictal spiking and background abnormalities” and seizure onset are resected (Wyllie, et al., 1987, Paolicchi, et al., 2000, Krsek, et al., 2008). Their findings imply a distinct location for IEDs and seizure onset in a subset of patients. Our data quantify this relationship that IEDs had a distinct localization from seizure onset in about forty percent of our patients. A potential hypothesis from combining these studies is that IEDs have a distinct localization and a causal relationship to seizure generation. One complicating factor in this discussion is the issue of generators vs. propagators of IEDs. It may be that regions generating IEDs may be vital to seizure generation, but that those that merely conduct discharges confound efforts to localize the ictal onset zone. Further quantitative studies looking at IED timing, correlation and propagation may shed considerable light on this issue. In the end, the question of what constitutes a seizure and what are the cellular and network elements that are necessary to generate it are central to this discussion.
A prominent finding in this study was the variability of IEDs in any given electrode over time, and relative changes in IED frequencies between electrodes over time. A number of physiological factors likely play an important role in this variability. Prior studies in the temporal lobe have implicated sleep as being one of the major sources of variability in the location of IEDs over time (Sammaritano, et al., 1991, Staba, et al., 2002). The frequency of IEDs and their spatial spread were greater in slow wave sleep than in wakefulness or REM sleep. Increased number of IEDs following a seizure has been reported, which may be affected by post-ictal sleep state (Gotman & Marciani, 1985, Gotman & Koffler, 1989). Epilepsy syndrome and localization may also play a role in the relationship between IEDs, sleep, and seizures, particularly in temporal lobe epilepsy (Spencer, et al., 2008). Our data suggest that there is a great deal of variability in frequency of IEDs in extra-temporal epilepsy.
Another potential cause for IED variability, in addition to sleep state, is alterations in anti-epileptic drug levels during the phase II surgical evaluation. Studies have reported a mixed relationship between IED frequency, and AED levels (Rodin, et al., 1974, Milligan, et al., 1983, Gotman & Marciani, 1985, Gotman & Koffler, 1989, Spencer, et al., 2008). Many of the patients in our study had medication adjustments over the course of their IEEG monitoring. An additional consideration for the variability in IED number is whether data epochs are pre ictal. There is as of yet no consensus whether there is a reproducible change in the temporal distribution of IEDs as seizures approach (Lieb, et al., 1978, Gotman, et al., 1982, Lange, et al., 1983, Katz, et al., 1991). These factors were not examined in the current study.
Cortical dysplasia, Palmini grade 2A, was the most predominant finding on pathology and these patients had IEDs that both correlated and did not correlate with seizure onset. Similar to prior studies in patients with cortical dysplasia (Turkdogan, et al., 2005), we found a mixture of seizure onset morphologies including rhythmic spiking, low voltage fast activity onsets, and rhythmic slowing at onset. Of the 13 patients with a rhythmic spiking morphology at seizure onset, 7 had IED frequency correlating with seizure onset. In contrast all the patients with the low voltage seizure onset displayed a correlation between IED frequency and seizure onset, suggesting that IED frequency may be a better marker for seizure onset electrodes in patients with low voltage fast activity on IEEG at seizure onset. It will be important to replicate these findings on a second set of patients.
In this study we utilized automated, computer-based EEG analysis to quantify IEDs over long periods of IEEG and identified patients with and without a correlation between IED and seizure onset regions. The improvement in computer technology over the past decade has allowed implementation of methods capable of easily analyzing IEDs over 8 hours of IEEG in up to 140 electrodes per patient. While our results are promising for using computer based detection methods to quantify IEDs it remains an open question as to how best utilize and refine these methods to improve outcome from epilepsy surgery, shorten length of stay, and potentially maximize the utility of intra-operative electrocorticography during electrode placement and resection. Even more importantly, this study raises important questions about how seizures and interictal epileptiform discharges are generated in human brain, and how to define and map epileptic networks.
CURE and NINDS NS044869 funding for B.E.P and funding from the Woman’s auxiliary of the Children’s Hospital of Philadelphia for P.B.S. Funding from NINDS NS048598, NS041811-02, and CA084438-05, and the following foundations, Klingenstein, Dana, Epilepsy Research, Whitaker, and PA Tobacco Fund for Brian Litt.
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.
The authors have no conflict of interest to report.