Professional Documents
Culture Documents
Abstract— High frequency oscillations (HFOs) are potential 16] used short intervals of iEEG recordings (i.e., less than 15
biomarkers of epileptic areas. In patients with drug-resistant minutes of data per intervals) chosen arbitrarily, with often
epilepsy, HFO rates tend to be higher in the seizure onset zone only one interval per channel. Furthermore, the SOZ and
(SOZ) than in other brain regions and the resection of HFO- non-SOZ channels included in these studies were only a
generating areas positively correlates with seizure-free surgery subset of the implanted iEEG recording channels and were
outcome. Nonetheless, the development of robust unsupervised chosen non-contiguous in space to better capture differences
HFO-based tools for SOZ localization remains challenging. between SOZ and non-SOZ. Finally, in [4, 6-9, 14-16], HFOs
Current approaches predict the SOZ by processing small were manually labelled by trained clinicians through visual
samples of intracranial EEG (iEEG) data and applying patient- inspection of the iEEG time series.
specific thresholds on the HFO rate. The HFO rate, though,
varies largely over time with the patient’s conditions (e.g., sleep Recent studies [18, 19], instead, have considered a more
versus wakefulness) and across patients. We propose a novel clinically-relevant scenario and proposed HFO-rate-based
localization method for SOZ that uses a time-varying, HFO- criteria to automatically localize the SOZ. First, each of these
based index to estimate the epileptic susceptibility of the iEEG solutions has included a clinically-validated algorithm for the
channels. The method is insensitive to the average HFO rate unsupervised detection of the HFOs. Then, for each patient,
across channels (which is both patient- and condition-specific), they have included iEEG data collected across the entire set
tracks the channel susceptibility over time, and predicts the of recording channels, thus extending their analysis to critical
SOZ based on the temporal evolution of the HFO rate. Tested regions that are located in between the clearly separable SOZ
on a preliminary dataset of continuous multi-day multichannel and non-SOZ. However, the proposed solutions were tested
interictal iEEG recordings from two epileptic patients (117± either on small, non-consecutive samples of iEEG recordings
97.6 h/per patient, mean ± S.D.), the reported SOZ prediction [19] or on averaged HFO rates calculated over several hours
had an average 0.70±0.18 accuracy and 0.67±0.07 area under of recordings [18], thus combining interictal and peri-ictal
the ROC curve (mean ± S.D.) across patients. data. Furthermore, each of these solutions used a threshold-
I. INTRODUCTION based policy to identify the SOZ, i.e., a channel is part of the
SOZ if the HFO rate in that channel is higher than a chosen
High frequency oscillations (HFOs, 80-500 Hz) are short threshold. Since the rate of HFOs may vary largely across
(i.e., typically 150 ms or less) events detectable in high-pass patients [12] and – in the same patient – over time [5], this
filtered intracranial EEG (iEEG) recordings [1, 2]. HFOs are means that the method must be calibrated to the specific
currently investigated as potential biomarkers of epileptic patient and iEEG dataset.
areas. Epileptiform HFOs may have distinctive shapes when In this preliminary study, we propose a novel HFO-based
compared with normal physiologic HFOs [3], occur more method for automatic localization of the SOZ and we tested it
often during slow-wave sleep than in wakefulness [4, 5], and on multi-day multichannel iEEG data from two patients in
the rate of epileptiform HFOs tend to be higher in the seizure the iEEG Portal database [20]. Differently from current
onset zone (SOZ) that in non-SOZ regions, both during solutions, the method uses a series of HFO rates estimated
interictal stages and at the onset of seizures [6-10]. Moreover, over consecutive 10-min-long windows to build a time-
the surgical removal of high-rate HFO regions significantly varying normalized vector index. The vector assigns a value
correlates with seizure-free surgery outcomes while poor between 0 and 1 to each iEEG channel based on the level of
surgery outcomes have been reported when the resected epileptogenic susceptibility in that channel and allows to
regions overlap only in part with the HFO-generating regions automatically identify the candidate channels for the SOZ.
[11-17]. Overall, these results indicate that the occurrence of We tested the algorithm on continuous interictal recordings
epileptiform HFOs can be a sign of seizure susceptibility and spanning several days per patient (i.e., periods of sleep and
higher HFO rates may indicate a greater risk of seizures [7]. wakefulness are included) and we performed a receiver
Despite the interest in using HFOs as a predictor of the operating characteristic (ROC) analysis to determine the
SOZ, the development of unsupervised HFO-based methods optimal parameters across the population of patients involved
for SOZ localization remains challenging. Studies [4, 6-9, 12- in the study. Data from ictal periods was initially excluded
given the low frequency of seizure events compared to the
*Equal contribution to the research. Research supported in part by the interictal data and the specific nature of ictal HFOs [21, 22].
US National Science Foundation through grant NSF ECCS-1518672. We found that our solution can provide robust predictions of
P. M. Murphy and A. J. von Paternos are with the Biomedical the SOZ with 20 minutes of consecutive iEEG recordings,
Engineering Department, University of Connecticut, Storrs, CT 06269 USA
(e-mail: {paige.murphy, adam.von_paternos}@uconn.edu). regardless of when recordings are collected (i.e., wakefulness
S. Santaniello is with the Biomedical Engineering Department and the or sleep), and that the accuracy of the method remains
CT Institute for Brain and Cognitive Sciences, University of Connecticut, statistically significant over time (above the 95th percentile)
Storrs, CT 06269 (corresponding author; phone: +1 860-486-4701; fax: +1 with average value 53.0±4.2% across patients (mean ± S.D.).
860-486-2500; e-mail: sabato.santaniello@uconn.edu).
Age (y)/Sex 27 / F 16 / M
PS/GTC CPS
Type of Seizures
Engel Class IV I
No. iEEG channels 56 88
Seizure Focus RT LO
Duration of iEEG
186 48
recordings (h)
Legend: F= female; M= male; PS= partial seizure; GTC= generalized tonic-
clonic seizure; CPS= complex partial seizure; RT= right temporal lobe;
LO= left occipital lobe.
II. METHODS
A. Dataset and HFO Detection Figure 1. A-B) Average HFO rate (mean±S.D.) for each iEEG channel
estimated across consecutive 10-min-long windows in Study-004-2 (A) and
Continuous multichannel iEEG recordings from two Study-023 (B). C-D) Box-plot of the components of Sk corresponding to the
patients in the iEEG Portal database (https://www.ieeg.org/) SOZ and non-SOZ regions across consecutive windows in Study-004-2 (C)
[20] were included in this study and processed in common- and Study-023 (D). For each box-plot, the median (red line), the 25th and
reference mode. See Table I for details. For each patient, the 75th percentiles (blue lines), the 98.5th percentile (whiskers), and outliers
(red asterisks) are reported.
channels covering the SOZ were identified as described in
[23] by board-certified epileptologists along with the clinical non-overlapping time windows Wk, k=1, 2, 3,.., i.e., R = Rk,
seizures. For each patient, iEEG recordings collected at least and the vector index
2 hours after the last clinical seizure were included in this Sk = exp(1– Rk) ∙ exp(1– Rk–1) (1)
study (interictal data) and sampled at 500 Hz. The decision of
excluding recordings collected before the first seizure event was introduced. Intuitively, vector Sk aims to characterize the
or between seizures was made to guarantee that the analysis level of epileptic susceptibility [25] in each channel in the
is conducted on consecutive interictal data, thus representing window Wk. The components of Sk are uniquely related to the
a scenario where seizures are rare events and the SOZ must iEEG channels (i.e., the j-th component corresponds to
be localized with no prior information about the occurrence channel j, j=1, 2, …, N), they all range between 0 and 1, and
of clinical seizures. Data from ictal periods was excluded as low values correspond to discharge patters with occasional
well given the low frequency of seizure events compared to HFOs (i.e., low susceptibility) while high values correspond
the interictal data and the specific nature of ictal HFOs [21, to sequences of HFO rates that are consistently above the
22]. Inclusion of ictal data in our analysis, however, did not average trend across channels (i.e., high susceptibility). In
significantly affect results reported in section III. (1), the epileptic susceptibility of a channel is associated with
HFO were detected on each iEEG channel independently the occurrence of HFOs in the previous 20 minutes (i.e., the
by using the unsupervised algorithm proposed by Staba et al. current window and the previous window) and the relative
[24] with artifact removal, where the artifacts were detected rate of HFOs compared to the remaining iEEG channels. The
as in [18]. Briefly, each iEEG time series was divided in number of past windows included in (1) was chosen by
consecutive, non-overlapping 10-min-long intervals and, for minimizing the fluctuations of the susceptibility index on a
each interval, the signal was high-pass filtered at 100 Hz training set including 3 h/channel of interictal activity (i.e.,
(fourth order Butterworth filter). HFOs were then detected as 18 windows) per patient.
oscillatory events in the filtered signal that last 100ms or less To assess the predictive power of index Sk, we performed
and contain at least 6 peaks, each peak having amplitude of at a receiver operating characteristic (ROC) curve analysis [26]
least 10 µV (absolute value) and being at least 3 times the for each window Wk across both patients. We also repeated
standard deviation of the mean rectified signal. The algorithm the ROC curve analysis over the entire set of windows Wk for
was also run on the common average reference and HFOs each patient, thus determining the average performance over
detected on this signal were considered artifact and removed multiple days. Regions that were either surgically removed or
from the analysis. otherwise marked as belonging to the SOZ in the clinical
notes were labelled as “true positives” and all the remaining
B. Index of Epileptic Susceptibility and ROC Analysis regions were “true negatives”. For each Sk, channels were
Denoted with N the number of iEEG channels recorded in marked as part of the SOZ if the correspondent components
a generic 10-min-long interval W, we estimated the HFO rate in Sk were above a threshold α, and the ROC curve was built
fj of channel j, j=1, 2, 3, …, N, as the average number of by varying α in [0,1]. For each ROC curve, the area under the
detected HFOs per minute in the channel and we ranked all curve (AUC) and the values of sensitivity, specificity, and
the channels based on the rate values, i.e., the higher the rate accuracy at cutoff point (i.e., the point maximizing the
the lower the rank position of the channel. Let us denote with Youden index [27]) were stored.
R = [r1 r2 r3 … rN] the vector of rank positions, where rj is the The significance of the AUC values was evaluated using
rank of channel j, j=1, 2, 3, …, N, and 1 ≤ rj ≤ N. For each the nonparametric permutation test described in [28]. Briefly,
patient, the rank vector R was recalculated for consecutive, the components of each vector Sk, k=1, 2, …, were randomly
1055
Figure 2. A-B) ROC curves for the vector index Sk (black lines) and the
vector of average HFO rates Fk (blue lines), respectively, estimated on the
entire dataset for Study-004-2 (A) and Study-023 (B). Curves are reported
along with the 99% confidence intervals. In A-B), cutoff points are marked
with red circles and estimated AUC values are reported in the legend.
1056
Finally, we chose the threshold α* that maximizes the [12] J. R. Cho, D. L. Koo, E. Y. Joo, D. W. Seo, S. C. Hong, P. Jiruska,
average AUC value across the entire dataset (i.e., all intervals and S. B. Hong, "Resection of individually identified high-rate high-
frequency oscillations region is associated with favorable outcome in
and patients) and we assessed the performance of such neocortical epilepsy," Epilepsia, vol. 55, pp. 1872-83, Nov 2014.
threshold on each patient individually. We found that the [13] H. Fujiwara, H. M. Greiner, K. H. Lee, K. D. Holland-Bouley, J. H.
average prediction accuracy remained high in both patients Seo, T. Arthur, F. T. Mangano, J. L. Leach, and D. F. Rose,
(Study-004-2: 0.80±0.12; Study-023: 0.85±0.1; mean ± S.D.) "Resection of ictal high-frequency oscillations leads to favorable
along with the sensitivity (Study-004-2: 0.77±0.12; Study- surgical outcome in pediatric epilepsy," Epilepsia, vol. 53, pp. 1607-
17, Sep 2012.
023: 0.45±0.17) and specificity (Study-004-2: 0.81±0.29;
[14] J. Jacobs, M. Zijlmans, R. Zelmann, C. E. Chatillon, J. Hall, A.
Study-023: 0.91±0.26), thus suggesting that the proposed Olivier, F. Dubeau, and J. Gotman, "High-frequency
index may need minimal customization to the patient. electroencephalographic oscillations correlate with outcome of
epilepsy surgery," Ann Neurol, vol. 67, pp. 209-20, Feb 2010.
IV. CONCLUSIONS [15] P. N. Modur, S. Zhang, and T. W. Vitaz, "Ictal high-frequency
oscillations in neocortical epilepsy: implications for seizure
In this study, we propose a novel method to automatically localization and surgical resection," Epilepsia, vol. 52, pp. 1792-801,
localize the SOZ in patients with multichannel iEEG. We use Oct 2011.
a time-varying HFO-ranking-based index to measure the [16] K. Kerber, M. Dumpelmann, B. Schelter, P. Le Van, R.
epileptogenic susceptibility of each iEEG channel and we Korinthenberg, A. Schulze-Bonhage, and J. Jacobs, "Differentiation of
choose candidate SOZ channels based on the value of the specific ripple patterns helps to identify epileptogenic areas for
surgical procedures," Clin Neurophysiol, vol. 125, pp. 1339-45, Jul
index over a 20-min-long interval. Tested on a preliminary 2014.
dataset of two patients, the method was robust against [17] C. Haegelen, P. Perucca, C. E. Chatillon, L. Andrade-Valenca, R.
fluctuations in the HFO rate over time, showed consistent Zelmann, J. Jacobs, D. L. Collins, F. Dubeau, A. Olivier, and J.
results across patients, and provided predictions of the SOZ Gotman, "High-frequency oscillations, extent of surgical resection,
that were significantly more accurate than a traditional HFO- and surgical outcome in drug-resistant focal epilepsy," Epilepsia, vol.
54, pp. 848-57, May 2013.
count-based approach. [18] S. V. Gliske, Z. T. Irwin, K. A. Davis, K. Sahaya, C. Chestek, and W.
C. Stacey, "Universal automated high frequency oscillation detector
REFERENCES for real-time, long term EEG," Clin Neurophysiol, vol. 127, pp. 1057-
66, Feb 2016.
[1] J. D. Jirsch, E. Urrestarazu, P. LeVan, A. Olivier, F. Dubeau, and J.
[19] S. Liu, Z. Sha, A. Sencer, A. Aydoseli, N. Bebek, A. Abosch, T.
Gotman, "High-frequency oscillations during human focal seizures,"
Henry, C. Gurses, and N. F. Ince, "Exploring the time-frequency
Brain, vol. 129, pp. 1593-608, Jun 2006.
content of high frequency oscillations for automated identification of
[2] J. Jacobs, R. Staba, E. Asano, H. Otsubo, J. Y. Wu, M. Zijlmans, I.
seizure onset zone in epilepsy," J Neural Eng, vol. 13, p. 026026, Apr
Mohamed, P. Kahane, F. Dubeau, V. Navarro, and J. Gotman, "High-
2016.
frequency oscillations (HFOs) in clinical epilepsy," Prog Neurobiol,
[20] A. A. Azarion, J. Wu, A. Pearce, V. T. Krish, J. Wagenaar, W. Chen,
vol. 98, pp. 302-15, Sep 2012.
Y. Zheng, H. Wang, T. H. Lucas, B. Litt, J. C. Gee, and K. A. Davis,
[3] A. Matsumoto, B. H. Brinkmann, S. Matthew Stead, J. Matsumoto, M.
"An open-source automated platform for three-dimensional
T. Kucewicz, W. R. Marsh, F. Meyer, and G. Worrell, "Pathological
visualization of subdural electrodes using CT-MRI coregistration,"
and physiological high-frequency oscillations in focal human
Epilepsia, vol. 55, pp. 2028-37, Dec 2014.
epilepsy," J Neurophysiol, vol. 110, pp. 1958-64, Oct 2013.
[21] S. A. Weiss, G. P. Banks, G. M. McKhann, Jr., R. R. Goodman, R. G.
[4] A. P. Bagshaw, J. Jacobs, P. LeVan, F. Dubeau, and J. Gotman,
Emerson, A. J. Trevelyan, and C. A. Schevon, "Ictal high frequency
"Effect of sleep stage on interictal high-frequency oscillations
oscillations distinguish two types of seizure territories in humans,"
recorded from depth macroelectrodes in patients with focal epilepsy,"
Brain, vol. 136, pp. 3796-808, Dec 2013.
Epilepsia, vol. 50, pp. 617-28, Apr 2009.
[22] P. Perucca, F. Dubeau, and J. Gotman, "Intracranial
[5] M. Dumpelmann, J. Jacobs, and A. Schulze-Bonhage, "Temporal and
electroencephalographic seizure-onset patterns: effect of underlying
spatial characteristics of high frequency oscillations as a new
pathology," Brain, vol. 137, pp. 183-96, Jan 2014.
biomarker in epilepsy," Epilepsia, vol. 56, pp. 197-206, Feb 2015.
[23] F. Rosenow and H. Luders, "Presurgical evaluation of epilepsy,"
[6] E. Urrestarazu, R. Chander, F. Dubeau, and J. Gotman, "Interictal
Brain, vol. 124, pp. 1683-700, Sep 2001.
high-frequency oscillations (100-500 Hz) in the intracerebral EEG of
[24] R. J. Staba, C. L. Wilson, A. Bragin, I. Fried, and J. Engel, Jr.,
epileptic patients," Brain, vol. 130, pp. 2354-66, Sep 2007.
"Quantitative analysis of high-frequency oscillations (80-500 Hz)
[7] J. Jacobs, R. Zelmann, J. Jirsch, R. Chander, C. E. Dubeau, and J.
recorded in human epileptic hippocampus and entorhinal cortex," J
Gotman, "High frequency oscillations (80-500 Hz) in the preictal
Neurophysiol, vol. 88, pp. 1743-52, Oct 2002.
period in patients with focal seizures," Epilepsia, vol. 50, pp. 1780-92,
[25] F. Bartolomei, P. Chauvel, and F. Wendling, "Epileptogenicity of
Jul 2009.
brain structures in human temporal lobe epilepsy: a quantified study
[8] J. A. Blanco, M. Stead, A. Krieger, W. Stacey, D. Maus, E. Marsh, J.
from intracerebral EEG," Brain, vol. 131, pp. 1818-30, Jul 2008.
Viventi, K. H. Lee, R. Marsh, B. Litt, and G. A. Worrell, "Data
[26] T. Hastie, R. Tibshirani, and J. H. Friedman, The elements of
mining neocortical high-frequency oscillations in epilepsy and
statistical learning : data mining, inference, and prediction, 2nd ed.
controls," Brain, vol. 134, pp. 2948-59, Oct 2011.
New York, NY: Springer, 2009.
[9] L. Andrade-Valenca, F. Mari, J. Jacobs, M. Zijlmans, A. Olivier, J.
[27] W. J. Youden, "Index for rating diagnostic tests," Cancer, vol. 3, pp.
Gotman, and F. Dubeau, "Interictal high frequency oscillations
32-5, Jan 1950.
(HFOs) in patients with focal epilepsy and normal MRI," Clin
[28] S. P. Burns, S. Santaniello, R. B. Yaffe, C. C. Jouny, N. E. Crone, G.
Neurophysiol, vol. 123, pp. 100-5, Jan 2012.
K. Bergey, W. S. Anderson, and S. V. Sarma, "Network dynamics of
[10] U. Malinowska, G. K. Bergey, J. Harezlak, and C. C. Jouny,
the brain and influence of the epileptic seizure onset zone," Proc Natl
"Identification of seizure onset zone and preictal state based on
Acad Sci U S A, vol. 111, pp. E5321-30, Dec 09 2014.
characteristics of high frequency oscillations," Clin Neurophysiol, vol.
[29] J. Engel, Surgical treatment of the epilepsies, 2nd ed. New York:
126, pp. 1505-13, Aug 2015.
Raven Press, 1993.
[11] T. Akiyama, B. McCoy, C. Y. Go, A. Ochi, I. M. Elliott, M. Akiyama,
[30] R. Alkawadri, N. Gaspard, Goncharova, II, D. D. Spencer, J. L.
E. J. Donner, S. K. Weiss, O. C. Snead, 3rd, J. T. Rutka, J. M. Drake,
Gerrard, H. Zaveri, R. B. Duckrow, H. Blumenfeld, and L. J. Hirsch,
and H. Otsubo, "Focal resection of fast ripples on extraoperative
"The spatial and signal characteristics of physiologic high frequency
intracranial EEG improves seizure outcome in pediatric epilepsy,"
oscillations," Epilepsia, vol. 55, pp. 1986-95, Dec 2014.
Epilepsia, vol. 52, pp. 1802-11, Oct 2011.
1057