Focal resection of fast ripples on extraoperative intracranial EEG improves seizure outcome in pediatric epilepsy


Address correspondence to Hiroshi Otsubo, M.D., Division of Neurology, The Hospital for Sick Children, 555 University Avenue, Toronto, ON M5G 1X8 Canada. E-mail:;


Purpose: High-frequency oscillations (HFOs), termed ripples at 80–200 Hz and fast ripples (FRs) at >200/250 Hz, recorded by intracranial electroencephalography (EEG), may be a valuable surrogate marker for the localization of the epileptogenic zone. We evaluated the relationship of the resection of focal brain regions containing high-rate interictal HFOs and the seizure-onset zone (SOZ) determined by visual EEG analysis with the postsurgical seizure outcome, using extraoperative intracranial EEG monitoring in pediatric patients and automated HFO detection.

Methods: We retrospectively analyzed 28 pediatric epilepsy patients who underwent extraoperative intracranial video-EEG monitoring prior to focal resection. Utilizing the automated analysis, we identified interictal HFOs during 20 min of sleep EEG and determined the brain regions containing high-rate HFOs. We investigated spatial relationships between regions with high-rate HFOs and SOZs. We compared the size of these regions, the surgical resection, and the amount of the regions with high-rate HFOs/SOZs within the resection area with seizure outcome.

Key Findings: Ten patients were completely seizure-free and 18 were not at 2 years after surgery. The brain regions with high-rate ripples were larger than those with high-rate FRs (p = 0.0011) with partial overlap. More complete resection of the regions with high-rate FRs significantly correlated with a better seizure outcome (p = 0.046). More complete resection of the regions with high-rate ripples tended to improve seizure outcome (p = 0.091); however, the resection of SOZ did not influence seizure outcome (p = 0.18). The size of surgical resection was not associated with seizure outcome (p = 0.22–0.39).

Significance: The interictal high-rate FRs are a possible surrogate marker of the epileptogenic zone. Interictal ripples are not as specific a marker of the epileptogenic zone as interictal FRs. Resection of the brain regions with high-rate interictal FRs in addition to the SOZ may achieve a better seizure outcome.

High-frequency oscillations (HFOs), termed ripples (80–200 Hz) and fast ripples (FRs, >200/250 Hz), were recorded initially in mesial temporal lobe epilepsy by intracranial electroencephalography (EEG) using microelectrodes (Bragin et al., 1999, 2002). FRs are thought to be a more specific surrogate marker of the seizure-onset zone (SOZ) than ripples (Bragin et al., 1999; Staba et al., 2002). Subsequently HFOs were also recorded by macroelectrodes (Jirsch et al., 2006; Urrestarazu et al., 2006), and they were seen in the neocortex (Worrell et al., 2004; Jacobs et al., 2009a). The resection of the brain regions containing HFOs, especially FRs, correlated with good seizure outcome using depth macroelectrode recording in adult patients (Jacobs et al., 2009b) and intraoperative electrocorticography (ECoG) in pediatric patients (Wu et al., 2010). However, these studies used relatively small numbers of contacts with limited coverage of the brain, which may not be optimal for pediatric patients with more extensive epileptic network than adults. In addition, extensive surgical resection, as is often required for pediatric patients, may overestimate the link between HFOs and seizure outcome; it may increase the chance of success by inclusion of both brain regions with HFOs and the epileptogenic zone in the resection, even when they are independent.

Utilizing automated HFO detection and multivariate analysis to remove confounding effects, we tested the following hypotheses with chronic extraoperative intracranial EEG monitoring with relatively extensive brain coverage in pediatric epilepsy patients: (1) the more complete resection of the brain regions containing high-rate HFOs is associated with the better seizure outcome regardless of the size of surgical resection, and (2) FRs are more predictive of seizure outcome than ripples. We also investigated the relationship between the brain regions with ripples, FRs, and the SOZ.



We retrospectively identified consecutive pediatric patients (≤18 year of age) with intractable epilepsy who had chronic extraoperative intracranial video-EEG (VEEG) monitoring to localize the seizure focus prior to focal resective surgery between July 2004 and June 2008. The exclusion criteria were: (1) resective surgery was declined after the intracranial VEEG monitoring; (2) patients with known bilateral seizure foci, who underwent surgery to control the most disabling seizures from one hemisphere; (3) intracranial VEEG performed in urgent situations to treat refractory focal status epilepticus; (4) postsurgical seizure outcome at 2 years was not available; and (5) archived data from the original EEG did not fulfill our interictal EEG selection criteria described later. This study was approved by the Research Ethics Board at the Hospital for Sick Children.

Chronic extraoperative intracranial VEEG recording

The technique of implantation of intracranial electrodes, mapping of the possible epileptogenic zone, and intraoperative and extraoperative functional mapping of eloquent cortex was done as described previously (Pang et al., 2009; Benifla et al., 2009). Center to center spacing of the contacts was 6.5–10.5 mm in subdural electrodes and was 7 mm in depth electrodes. The diameter of subdural electrodes was 4 mm with exposure of 2.3 mm (effective surface area: 4.2 mm2) and the surface area of depth electrodes was 8.3 mm2 (Ad-Tech Medical Instrument, Racine, WI, U.S.A.). Intracranial VEEG recordings were recorded using the Harmonie system (Stellate, Montreal, PQ, Canada) with two electrodes over the brain regions without active spikes used as linked references. The EEG signals were sampled at 1 kHz for 24–216 h after band-pass filtering at 0.016–300 Hz.

Resective surgery and seizure outcome

The resection margin was prospectively determined based on visual analysis of ictal EEG discharges, spectral analysis of EEG at ictal onset and during early spread (Ochi et al., 2007), interictal spikes on intracranial VEEG, equivalent current dipoles for the interictal spikes on magnetoencephalography (MEG), and the location of eloquent cortices revealed by functional mapping using direct cortical stimulation and sensory evoked potentials. The SOZ was prospectively determined by clinical neurophysiologists during the intracranial VEEG. Because the evaluation of HFOs in this study was retrospectively performed, the surgical decision was not related to the results of this study.

Seizure outcome was assessed annually for 2 years postsurgery according to the International League Against Epilepsy (ILAE) classification (Wieser et al., 2001). For statistical analysis, we simplified the seizure outcome into two groups: group I, completely seizure-free without auras for 2 years following surgery (class 1a in the ILAE classification); group II, otherwise.

Evaluation of HFOs

For each subject, we detected interictal ripples (80–200 Hz) and FRs (200–300 Hz) during non–rapid-eye-movement (NREM) sleep using bipolar montage. Because the analysis of 5- to 10-min EEG provides the same information of HFOs as that of longer sections in most cases (Zelmann et al., 2009), we selected 10 epochs of 2-min interictal EEG during sleep so that they were remote from each other and from seizures by at least 1 h. We chose to collect multiple brief epochs, rather than a smaller number of epochs of longer duration (e.g., one 20-min epoch), in order to reflect the variability of the EEG during the recording period. In general, interictal HFOs occur most frequently during NREM sleep than in other states (Staba et al., 2004; Bagshaw et al., 2009). We selected EEG epochs with increased slow waves while the patients were asleep. We visually inspected each EEG epoch with a high-pass filter at 0.5 Hz and at 200 Hz to ensure that they were not contaminated by significant artifacts, such as environmental artifacts and muscle artifacts (Otsubo et al., 2008).

We used bipolar montage with pairs of two adjacent contacts in successive numbers connected, excluding the contacts for reference and those with significant artifacts. The automated detection of HFOs was performed in a way similar to previous descriptions (Staba et al., 2002; Gardner et al., 2007; Schevon et al., 2009; Crépon et al., 2010), with some modifications using MATLAB (The MathWorks, Natick, MA, U.S.A.). We validated the performance of this automated detection with two clinical neurophysiologists (BM, AO). The details are described in Fig. S1.

We calculated the rate (per minute) of ripples and FRs on each EEG channel, and determined the brain regions with high-rate ripples and FRs (Fig. 1). The analysis of 10 epochs of EEG recording yielded 10 values of the rate (per minute) of ripples and FRs for each channel; from this we calculated the median values of the rate of ripples and FRs for each channel. Subsequently, we determined the channels with high-rate HFOs (ripples/FRs) for each patient by thresholding. The thresholds for ripples and FRs were calculated separately due to the different criteria used to detect them. They were calculated for each patient because of significant interindividual variability of the rate of HFOs. Assuming roughly uniform distribution of the EEG contacts, we used Kittler’s method to determine the threshold based on the histogram of the rate of HFOs from all channels (Kittler & Illingworth, 1986). Because of small samples, we applied bootstrapping to determine the distribution of the threshold and used its mean as the final threshold. When the rate of HFOs of a channel was equal to or above the threshold, the channel was declared to have high-rate HFOs. All channels with a rate of <1/min were considered not to have high-rate HFOs to avoid declaring the channels with very low rates as having high-rate HFOs.

Figure 1.

Overview of the evaluation of HFOs and the SOZ. (A) Unfiltered interictal EEG in bipolar montage displayed at 3 s/page. (B) Results of automated detection of ripples at the top and FRs at the bottom. Detected HFOs are highlighted in pink. (C) Histograms of the rates of ripples at the top and FRs at the bottom. The threshold for high-rate ripples is 14.0/min and that for high-rate FRs is 5.0/min (see main text). (D) Topographic maps showing the rate of HFOs. The yellow line indicates the resection margin and the light blue line outlines the visually determined SOZ. The channels with high-rate HFOs are shown in red circles (filled circles: inside the resection margin, half-filled circles: at the edge of the margin, open circles: outside the margin). In this sample, Resection size = 50, Ripple region size = 44, Ripple resection ratio = 38/44 = 86.4%, FR region size = 46, FR resection ratio = 39.5/46 = 85.9%, SOZ size = 6.5, SOZ resection ratio = 6.5/6.5 = 100%, Overlap size (ripples and FRs) = 38, Overlap size (ripples and SOZ) = 6.5, Overlap size (FRs and SOZ) = 5.5.

Quantitative measures

We calculated the sizes of the regions with high-rate ripples/FRs, SOZ, and completeness of resection of these regions by the following formulas.

  • Resection size = No. of channels within resection margin.

  • Ripple region size = No. of channels with high-rate ripples.

  • FR region size = No. of channels with high-rate FRs.

  • SOZ size = No. of channels within SOZ.

  • Ripple resection ratio = No. of channels with high-rate ripples within resection margin/No. of channels with high-rate ripples × 100 (%).

  • FR resection ratio = No. of channels with high-rate FRs within resection margin/No. of channels with high-rate FRs × 100 (%).

  • SOZ resection ratio = No. of channels within SOZ within resection margin/No. of channels within SOZ × 100 (%).

The number of channels within the resection margin/SOZ was counted as follows: (1) if both contacts in a bipolar channel were within the resection margin/SOZ, the channel was counted as “1”; (2) if only one of two contacts in a bipolar channel was within the resection margin/SOZ, the channel was counted as “0.5.”

To investigate the spatial relationship between the regions with ripples, FRs, and SOZ, we calculated the sizes of overlap between them as follows.

  • Overlap size (ripples and FRs) = No. of channels with high-rate ripples and FRs.

  • Overlap size (ripples and SOZ) = No. of channels with high-rate ripples and within SOZ.

  • Overlap size (FRs and SOZ) = No. of channels with high-rate FRs and within SOZ.

Statistical analysis

We compared the Ripple region size with FR region size by paired Wilcoxon test. We compared the Ripple/FR region size and SOZ size between groups with and without (1) focal seizures, (2) epileptic spasms, (3) magnetic resonance imaging (MRI) lesion, (4) multiple lobe resection, and (5) focal cortical dysplasia (FCD) in pathology by unpaired Wilcoxon test. We tested the effect of the quantitative measures on seizure outcome by univariate and multivariate logistic regression analyses. These analyses were done by JMP 4.0.5 (SAS Institute, Cary, NC, U.S.A.) and R 2.11.1 (available at The patients who had no channel with high-rate ripples/FRs were excluded for the analyses of Ripple/FR resection ratio. The significance level was set to p < 0.05 and the trend (weak significance) level was set to p < 0.1.


Subject characteristics

Forty-two patients underwent intracranial VEEG studies. We excluded 14 patients: one with known bilateral epileptic foci, but had surgery to control the most disabling seizures from one hemisphere, one who had urgent surgery due to refractory focal status epilepticus, 6 with postsurgical seizure outcome at 2 years not available, and 6 with EEG data not fulfilling our criteria to select interictal epochs. We analyzed the remaining 28 patients (Table 1). The age of surgery was 1–18 years (median 11 years). There were 23 patients (82.1%) with focal seizures, 10 (35.7%) with epileptic spasms, and 5 (17.9%) with both at the time of surgery. There were 21 patients (75.0%) with lesion(s) detected on MRI. The intracranial VEEG was recorded using 48–124 (median 109.5) contacts for 24–216 h. There were 10 patients (35.8%) in the seizure outcome group I and 18 (64.3%) in the seizure outcome group II at 2 years after surgery.

Table 1.   Characteristics of subjects
CaseAge at surgery (years)Seizure typeMRI findingsMEG spike source clustersTotal no. of EEG contacts/no. of depth contactsSurgeryPathologySeizure outcome (1st year/2nd year/simplified)Threshold for high-rate ripples (/min)Threshold for high-rate FRs (/min)No. of channels with high-rate ripples/resected ripplesNo. of channels with high-rate FRs/resected FRsNo. of channels within SOZ/resected SOZ
  1. 2nd GTC, secondarily generalized tonic–clonic seizures; MRI, magnetic resonance imaging; MEG, magnetoencephalography; R, right; L, left; F, frontal; T, temporal; P, parietal; O, occipital; inf., inferior; sup., superior; ant., anterior; post., posterior; mid., middle; FCD, focal cortical dysplasia; CE, cortical excision; CSE, cortical and subcortical excision.

115FocalHyperintensity in R posterior hippocampusR F, mesial T124/8R T lobectomy; CE - R inf. F and inf. PGliosis4/1/II6.5153/3241/281/1
2 4FocalAbnormal signal in subcortical white matter in R OR P-O112/8R O lobectomy; R P CEFCD (IIB)1/3/II314/44/45/5
314Focal, 2nd GTCR P FCDR inf. rolandic, perisylvian109/2R P lesionectomy and CEFCD (IIB)2/3/II1110/21/0.515/11
4 7FocalNormalR post. T, inf. P52/4R T lobectomy R inf. P CEMicrodysgenesis1/1/I1.91.913/12.510/9.54/4
5 5FocalR P FCDR post. T117/18R P and post. T lesionectomy; R F CEFCD (IIB)1/1/I1116/70/019/14.5
6 4SpasmsTuberous sclerosisL OL 106/12, R 12L T and O lobectomy; CE and tuberectomy - L inf. F and inf. PTuberous sclerosis1/1/I4511.56/642/3548.5/44.5
7 6FocalTuberous sclerosisR inf. rolandic, perisylvian120/12CSE - R inf. rolandic and sup. THeterotopic neurons of unknown significance5/5/II2.216/4.52/1.54/1
8 1SpasmsNormalR P-O, L P (small)104/16R P-O CEMicrodysgenesis4/4/II11.215/43/1.516/16
911Focal, 2nd GTCR T-P FCDR T-P-O124/16R T lobectomy; R P CSEFCD (IIB)5/5/II6.9181/5149/31.513/13
1015Focal, 2nd GTCR posterior T-O FCD or hamartomaR T-O114/16R T lobectomy; CE - post. T and inf. rolandicFCD (IIB)5/5/II3.3129/1716/131.5/1.5
11 8Focal, 2nd GTCTuberous sclerosisL inf. F, TL 90/8 R 12L T lobectomy CE - L F and post. TTuberous sclerosis1/3/II4142/33.529/2322.5/22.5
12 4Focal, spasmsL F FCDL F-T-P72/8CE - L post., sup. and mesial FNo pathologic diagnosis4/1/II25.516/63/0.528.5/23
1318FocalR mesial F FCDR F-rolandic, T101/12R T lobectomy; R F CE; R subF lesionectomyFCD (IIA)1/3/II117/5.523/12.57/7
1416FocalL hemispheric atrophyL inf.-mid. F, T110/8L T lobectomy; CE - L mid.-inf. F, L mid.-post. T and L inf. PGliosis, hippocampal sclerosis3/1/II2.71.129/25.510/61.5/1.5
1517Focal, 2nd GTCAbnormal signal in R rolandicR P (sensory cortex)120/0R postcentral gyrectomyFCD (IIB)2/2/II1.713/14/34/3
167FocalDeep sulcus in L F lobeL F118/4L F lobectomyFCD (IIB)1/1/I3.514/23/31/1
173Focal, SpasmsR F-T FCDR F-T120/8R T lobectomy (not including mesial structures); CE – R inf., post. F and post. TMicrodysgenesis1/5/II2.3122/17.512/126/6
1810FocalR P FCDR P96/8R P lesionectomy and CEFCD (IIB)1/1/I111/10/03/3
1912FocalMild abnormal signal in L inf. FL F101/8L F lobectomy and CEMicrodysgenesis4/4/II1.6144/128/417.5/16.5
2011Focal, spasmsTuberous sclerosisR F90/14R F CE and tuberectomyTuberous sclerosis4/4/II4.1127/16.515/818.5/17
2110Focal,2nd GTC, spasmsNormalL F113/8CE- L inf. rolandic and ant. FGliosis1/1/I41.343.22/115/1015/11
2213Focal, spasmsNormalL rolandic, T101/4L rolandic CEGliosis4/4/II1.8118/85/58/8
2314SpasmsNormalR inf. F114/8R F CENo pathologic diagnosis2/2/II1.1135/1710/334/20
2414FocalT2 hyperintensity and atrophy in R hippocampusR mesial T, inf. F103/8R T lobectomy; R T-P CEMild hippocampal gliosis1/1/I1134/298/85/5
2513SpasmsNormalL F106/0L F lobectomy; CE – sup. F, inf. F and middle TMicrodysgenesis1/1/I1.3129/2413/1120/20
2612Focal, 2nd GTCNormalL perisylvian104/0L T lobectomy and CEFCD (IIA)1/1/I1127/23.51/19/9
2717Focal, 2nd GTCL pre-C small cystL rolandic (mainly sensory cortex)48/0L post-C gyrectomyGliosis4/1/II110/01/17/2
288SpasmsR perirolandic and post. T FCDR rolandic, perisylvian118/8CE – R C, P, sup. TPolymicrogyria1/1/I5.21.163/4169/4217/17

The size of brain regions with HFOs and SOZ, and their relationship

The intervals between the selected EEG epochs for analysis ranged from 1 to 20.7 h (mean 2.5 h, median 1 h). There were 27 patients (96.4%) with high-rate ripples in ≥1 channel, 26 (92.9%) with high-rate FRs in ≥1 channel, and 25 (89.3%) with high-rate ripples in ≥1 channel and high-rate FRs in ≥1 channel.

We investigated the spatial extent and relationship between the regions with high-rate ripples/FRs and the SOZ in the 25 patients with both high-rate ripples and FRs present (Fig. 2). In these 25 patients, Ripple region size was 2–81 channels (median 22), FR region size was 1–69 channels (median 10), and SOZ size was 1–48.5 channels (median 9). Ripple region size was significantly larger than FR region size (p = 0.011). Overlap size (ripples and FRs) ranged from 0–59 channels (median 5), which constituted 0–100% (median 28.6%) of the regions with high-rate ripples and 0–100% (median 85.5%) of those with high-rate FRs, with one patient showing no overlap. Overlap size (ripples and SOZ) ranged from 0–26 channels (median 4) and constituted 0–83.3% (median 25.0%) of the regions with high-rate ripples and 0–100% (median 46.2%) of the SOZ, with two patients showing no overlap. Overlap size (FRs and SOZ) ranged from 0–34 channels (median 2) and constituted 0–85.0% (median 19.0%) of the regions with high-rate FRs and 0–100% (median 30.8%) of the SOZ, with four patients showing no overlap.

Figure 2.

Relationship between the regions with high-rate ripples/FRs and the SOZ. This demonstrates the size of the brain regions with high-rate ripples/FRs, the SOZ, and their spatial relationship in 25 patients with high-rate ripples and FRs in ≥1 channels. The size of these regions is shown in the median number of channels and range in parentheses. The size of overlap between two regions is shown in the median percentages of each region the overlap constitutes. The details are described in the text.

Comparison of sizes of HFO-containing regions and SOZ with clinical characteristics

The 18 patients who underwent multiple lobe resection had significantly larger FR region size than the 10 patients with single lobe resection [median 12.5 (0–69) vs. 3 (0–15), p = 0.0088]. The FR region size did not differ between groups with and without focal seizures, epileptic spasms, MRI lesion, and FCD. There was no difference in Ripple region size with any comparisons.

Twenty-three patients with focal seizures had smaller SOZ size than five patients without them [median 6.5 (1–28.5) vs. 20 (16–48.5), p = 0.0072]. Ten patients with epileptic spasms had larger SOZ size than 18 patients without them [17.8 (6–48.5) vs. 5 (1–22.5), p = 0.0022]. The SOZ size did not differ between groups with and without MRI lesion, FCD, and multiple lobe resection.

Effects of HFOs and SOZ on seizure outcome

Table 2 describes effects of ripples/FRs and SOZ on seizure outcome by logistic regression analyses. In univariate analyses, the larger FR resection ratio tended to improve seizure outcome [odds ratio 1.05 (95% confidence interval, 1.00–1.10), p = 0.065]. The Resection size, Ripple region size, Ripple resection ratio, FR region size, SOZ size, and SOZ resection ratio were not associated with seizure outcome.

Table 2.   Effect of ripples, fast ripples, and the seizure-onset zone on seizure outcome
 Odds ratio (95% CI)p-value
  1. CI, confidence interval; FR, fast ripples; SOZ, seizure onset zone.

  2. *p < 0.05; p < 0.1.

Univariate analyses
 Resection size1.00 (0.95–1.05)0.97
 Ripple region size0.99 (0.95–1.03)0.58
 Ripple resection ratio (n = 27)1.03 (0.99–1.07)0.11
 FR region size1.01 (0.97–1.06)0.65
 FR resection ratio (n = 26)1.05 (1.00–1.10)0.065
 SOZ size1.02 (0.95–1.09)0.56
 SOZ resection ratio1.03 (0.98–1.09)0.28
Multivariate analyses
 Ripples (n = 27)
  Resection size0.97 (0.89–1.04)0.39
  Ripple region size1.00 (0.95–1.06)0.97
  Ripple resection ratio1.04 (0.99–1.09)0.091
 FRs (n = 26)
  Resection size0.94 (0.86–1.03)0.22
  FR region size1.10 (0.99–1.21)0.072
  FR resection ratio1.10 (1.002–1.21)0.046*
  Resection size0.96 (0.90–1.03)0.26
  SOZ size1.05 (0.96–1.15)0.25
  SOZ resection ratio1.07 (0.97–1.18)0.17

Using multivariate analysis, we incorporated the Resection size, Ripple region size, and Ripple resection ratio to see the effect of ripples on seizure outcome. The larger Ripple resection ratio tended to improve seizure outcome [odds ratio 1.04 (0.99–1.09), p = 0.091]. Similarly, we analyzed the effect of FRs using multivariate analysis. The larger FR resection ratio significantly correlated with improved seizure outcome [odds ratio 1.10 (1.002–1.21), p = 0.046]. The larger FR size tended to improve seizure outcome [odds ratio 1.10 (0.99–1.21), p = 0.072]. Resection size, Ripple region size, SOZ size, and SOZ resection ratio were not associated with seizure outcome.


HFO-containing regions correlating with epileptogenesis

The significant correlation between resection of the regions with high-rate FRs and seizure outcome in pediatric epilepsy supports that the interictal FRs could be a surrogate marker of the epileptogenic zone. Prolonged extraoperative VEEG monitoring with extensive coverage of the brain can localize the focal regions with high-rate FRs correlating with seizure outcome. In the patients without seizure freedom during the postoperative 2 years, the unresected regions containing high-rate FRs may have become ictogenic.

Compared with FRs, larger resection of the regions with high-rate ripples showed only a trend toward better seizure outcome. Regions with high-rate ripples were larger than and only partially overlapped with those with high-rate FRs. Therefore, ripples are not as specific a marker of the epileptogenic zone as FRs.

We found that resection of the SOZ did not correlate with seizure outcome. Jacobs et al., (2009b) reported no correlation between resection of SOZ and surgical outcomes in adult patients who presented with the significant correlation between FRs and seizure outcome. Although the SOZ and the regions with high-rate ripples/FRs overlapped in most patients, the size of overlap was small. The discrepancy between resection of the regions with high-rate FRs and that of the SOZ with regard to correlation with seizure outcome indicated that the regions with high-rate FRs outside the SOZ had a potential to cause seizure recurrence after surgery. We speculated the relations among epileptogenesis, epileptic zones, HFOs, and postsurgical courses in Fig. 3. It should be noted that FRs were also observed in the nonepileptogenic cortex in previous reports (Curio et al., 1994; Fukuda et al., 2008; Nagasawa et al., 2011). Therefore, epileptologists should determine the resection margin, taking into account the spatial relationship among the regions with high-rate HFOs, SOZ, irritative zone, lesion, and the eloquent cortex.

Figure 3.

Speculative model of HFOs during epileptogenesis. Epileptogenesis is the process that transforms the normal brain into the epileptic brain. During the epileptogenesis, subsets of the normal neuronal network reach the first critical point (in yellow) where epileptogenicity is gained and the process becomes irreversible. As the epileptogenesis progresses, parts of the epileptogenic zone reach the second critical point (in red) where ictogenicity is acquired to generate epileptic seizures. The irritative zone may involve not only pathologic but also normal neuronal network. If epilepsy surgery fails to resect the entire epileptogenic zone, similar process will continue in the residual brain regions (small arrows in the bottom) and seizure will recur. Although both ripples and FRs can be seen in the normal neuronal network, FRs are probably more specific to the epileptogenic zone than ripples.

Extensive coverage of the epileptic network in intracranial EEG

Jacobs et al., (2009b) demonstrated correlation between the resection of HFO-containing regions and good surgical outcome by visual analysis of HFOs recorded by depth macroelectrodes. However, the brain coverage was limited, with relatively small numbers of contacts due to stereotactic depth electrode placement for adult patients. Wu et al., (2010) reported that complete resection of the cortex containing FRs using intraoperative ECoG correlated with seizure freedom in pediatric patients with lesions. The intraoperative ECoG recording was limited to cover small areas, short duration, and environmental artifacts. The effect of anesthesia on interictal HFOs is not well known. The limited coverage of the brain by EEG contacts in these studies may not be optimal for pediatric patients, who generally have a more extensive epileptogenic zone than adults. We demonstrated that the SOZ was large in epileptic spasms and small in focal seizures. Pediatric patients with focal-onset epileptic spasms have an extensive epileptic network (Ramachandrannair et al., 2008; Nariai et al., 2011). For patients in whom a large epileptic network is suspected, especially those with epileptic spasms, extensive coverage by intracranial EEG contacts and prolonged monitoring to capture artifact-free recording is recommended to analyze interictal HFOs.

The resection size is a confounder in determining the value of FRs on seizure outcome. Extensive resections including hemispherectomy/hemispherotomy, as often reported in pediatric cases, could increase inclusion of both the epileptogenic zone and the FR-containing brain regions within the resection. Therefore, even though they are independent, they may appear linked. To minimize this confounding effect, we analyzed only patients who underwent focal resections in this study and performed multivariate logistic regression analysis. We demonstrated that larger resection of the regions with high-rate FRs correlated with improved seizure outcome independent of the resection size. When performing HFO analysis in human surgical subjects, the coverage of the brain by EEG contacts, the resection size, and the localization of HFOs needs to be cautiously reviewed.

Automated HFO analysis

Several reports on HFOs have been published where interictal HFOs were visually identified using high-pass filters (Urrestarazu et al., 2007; Jacobs et al., 2008; Bagshaw et al., 2009; Jacobs et al., 2009a; Zijlmans et al., 2009; Zelmann et al., 2009). The problem with visual analysis is subjectivity due to interobserver variability of detection given the lack of solid definition of the onset and offset of the HFOs. Moreover, intraobserver variability in detection can occur depending on the reader’s level of experience. In addition, visual analysis of HFOs is highly time-consuming. It can take 10 h to visually mark HFOs in a 10-channel 10-min recording (Zelmann et al., 2009). Because visual analysis was not practical for over 100-channel recordings in this study, we developed an automated HFO-detection program. The automated detection used consistent criteria for the detection of HFOs, although the software required configuration to ensure acceptable performance.

Criteria for high-rate HFOs

We objectively calculated thresholds to define high-rate ripples and FRs for each patient to examine spatial overlap among the brain regions with high-rate ripples, FRs, and the SOZ. Visual determination of the regions with high-rate HFOs on a topographic map showing the rate of HFOs is subjective and dependent on color scale setting. We assumed that thresholds would be different among patients and between ripples and FRs. In this study we used Kittler’s method that was originally developed for image binarization (Kittler & Illingworth, 1986), which fits a histogram by two normal distributions with unequal means and variances to calculate the threshold. Limited coverage of the brain by EEG contacts is, however, always an issue for the intracranial VEEG, as larger coverage could change the threshold.


Because of the 2-year follow-up period, this study did not completely answer whether the patients who achieved seizure freedom would remain seizure-free beyond 2 years postsurgery, even after tapering medications. Our results with trend levels (p < 0.1) would need further tests using larger samples. Even though extensive intracranial EEG electrodes were used for recording, HFOs might have occurred outside the recording area if we erroneously localized the possible epileptogenic zone by noninvasive presurgical evaluations. It would be interesting to examine spatial overlap between the brain regions with high-rate HFOs and interictal spikes to compare their values on seizure outcome. The sampling rate of 1 kHz was not ideal to record FRs that may extend up to 500 Hz or higher, because of antialiasing filtering. The use of filters to detect HFOs can produce “false” HFOs from steep interictal spikes, which may be misleading (Bénar et al., 2010). Therefore, complete differentiation of real HFOs from steep spikes without accompanying HFOs was difficult by our analysis method, although the HFOs visualized by filtering have been reported to be a marker of the seizure-onset zone independent of interictal spikes (Jacobs et al., 2008). Distinguishing of pathologic HFOs from physiologic HFOs presents a technical challenge (Engel et al., 2009). Performance of HFO detection software needs to be cautiously validated.


Interictal high-rate FRs are a possible surrogate marker of the epileptogenic zone in pediatric epilepsy patients. Interictal ripples had more extensive distribution than FRs and partially reflect epileptogenesis and irritability. The visually determined SOZ partially overlapped with high-rate FR-containing regions. The larger SOZ in epileptic spasms in comparison to focal seizures reflects the extensive epileptic network involved in epileptic spasms. The entire epileptic network containing FRs, ripples, and SOZs must be thoroughly analyzed to ensure complete resection of the epileptogenic zone with the smallest resection possible. Although this result is preliminary and requires validation with larger cohorts with longer follow-up, the analysis of interictal HFOs provides clinicians with a new approach for epilepsy surgery to potentially improve the seizure outcome.


None of the authors has any conflict of interest to disclose. 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.