A 27-year-old, right-handed male with known medically refractory focal epilepsy that started when he was 14 years old. His epilepsy risk factor included febrile seizures with spinal meningitis during childhood (6 months old) that resolved when he was 1 year old. He was experiencing three seizure subtypes: (1) focal onset seizures without loss of awareness (previously called simple partial) (28) where he reported déjà vu, increased anxiety lasting 10–30 s with frequency 20 per day; (2) focal onset seizures with impaired awareness (previously called complex partial), lip and manual automatism lasting 2–3 min at a frequency of 1–2 times every week; and (3) focal to bilateral tonic–clonic seizures 1–2 times per year. His MRI brain was non-lesional, but 18-fluoro-2-deoxyglucose (18F-FDG) PET-imaged showed right temporal (mesial and lateral) hypometabolism. After scalp EEG monitoring, he underwent right anterior temporal lobectomy at an outside hospital when he was 21 years old. He was seizure free for first 6 months, but all his seizure subtypes gradually returned and progressively increased in frequency. He was then reevaluated for possible second epilepsy surgery at our level-IV epilepsy center (University of Alabama at Birmingham).

The patient underwent SEEG monitoring with depth electrodes implanted in the right anterior and posterior insula, right basal temporal, right anterior and posterior orbitofrontal, a remnant of right mesial temporal structures, posterior and superior temporal gyrus. The superior temporal depth electrode was extended medially to include ventrolateral thalamus. Overall, eight depth electrodes with a total of 96 contacts (i.e., 96 channels) were sampled at 1 kHz with a Natus Xltek EEG machine (see the electrode montage in Figure 1).

During his stay in the epilepsy monitoring unit (EMU), his home antiepileptic (AED) medications (clonazepam, calproic acid, oxcarbazepine) were gradually weaned off. Within days after weaning off medications, electroclinical seizures, and subsequently a prolonged (lasting over 3 h) SE was recorded. The SE was of focal onset around the remnant of previously resected mesial temporal structure, progressed gradually to involve insula, lateral temporal structures, and orbitofrontal. The SE spread to the thalamus in the second half of the SE.

Before the onset of SE, patient was awake and alert. The patient initially had no definite clinical correlates which subsequently progressed to bilateral tonic–clonic activities. Throughout the SE, patient was examined periodically by the nurse and EEG technician in the EMU. Examination included assessment of speech (by asking to read a sign board), awareness (by asking about time, place), motor response (by asking to grab a pen), and asking the patient to self-report his feelings during the event. Patient had fluctuating consciousness in the first half of the SE but was lethargic, confused, and had minimal motor response toward the end. Immediately after SE, patient was not oriented to time or place but had appropriate motor response and had no aphasia.

The SE was successfully managed by parenteral valproic acid and two doses of lorazepam. The patient eventually underwent a large resection of temporal lobe structures that included the remnant hippocampus, superior temporal gyrus, and anterior insula [these structures were clinically identified as the seizure onset zone (SOZ)]. Eighteen months postresection the patient is seizure free (Engel’s class I) and remains on AED medication.

Histopathology confirmed chronic neuronal loss, the presence of dysmorphic neurons and extensive gliosis. In particular, immunohistochemical stains for GFAP, neurofilament protein (NFP), and Neu-N neuronal marker were performed on all the tissue blocks. GFAP staining showed mild gliosis in middle temporal gyrus, moderate gliosis in superior temporal gyrus, and marked gliosis in residual hippocampus. Neu-N staining showed mild neuronal losses in hippocampus, middle and superior temporal gyrus. Staining for hyperphosphorylated NFP showed no neuronal somatic staining in middle temporal gyrus, occasional positive cortical cells in superior temporal gyrus, and also a few scattered positive neuronal cells in hippocampus. Appropriate staining was seen in the positive and negative controls. The margin of superior temporal gyrus and the remnant of right hippocampus showed dysmorphic neurons. Anterior insula and margin of middle temporal gyrus showed chronic neuronal loss and astrogliosis.

A 24-h portion of the SEEG (starting 12 h before the onset of SE) was stored and made available for retrospective mathematical analysis. This retrospective case study had institutional IRB approval from the University of Alabama at Birmingham.

We employed the Teager Energy (TE) operator (29, 30) to estimate the energy of the SEEG and detect occurring seizures. TE has been used widely for seizure detection, solely or in combination with other measures of EEG characteristics (31–33). TE was applied to band-pass (5–15 Hz) filtered 60 s, non-overlapping consecutive SEEG segments over the duration of the available recording per channel. TE was estimated by: (1)TE(i)=x2(i)−x(i−1)⋅x(i+1), where x(i) is the amplitude of the SEEG signal at time point i. With a sampling frequency of 1 kHz, 60,000 TE(i) values were produced for every 60 s SEEG segment. Then, the median (TE_med) of those TE values was estimated per 60 s segment generating a TE_med temporal profile per electrode site every 60 s. Finally, averaging the TE_med profiles across electrodes (spatial averaging) generated the TE_global energy profile. We herein employed the TE_global measure as a tool for detection of seizures and SE from the SEEG.

We employed a recently proposed measure of the directional inflow of information a node is experiencing from its communication with the rest of the nodes in the network, the measure of the total effective inflow (TEI) of information. The TEI is estimated from the generalized partial direct coherence (GPDC) that measures the directed (causal) interactions between nodes in a network based on the principle of Granger’s causality (23, 24). We have shown that GPDC outperforms many traditional measures of directed interactions in the frequency domain [e.g., partial coherence, directed coherence, partial directed coherence (PDC), normalized PDC, directed transfer function (DTF), normalized DTF] when the question of localization of the epileptogenic focus is addressed (see (23) and references therein). In particular, from such an analysis of interictal intracranial EEG recordings (23) and interictal magnetoencephalographic (MEG) recordings in respective groups of patients with focal epilepsy (24), we have found that the epileptogenic focus can be accurately localized as the region that most frequently receives the maximum effective inflow (TEI) from other brain regions in the interictal period. We briefly describe our methodology below.

A multivariate autoregressive model X(t)=∑τ=1pA(τ)X(t−τ)+ϵ(t), of order p = 7 and dimension of vector X equal to 96, was fit to each of 60 s successive, non-overlapping SEEG segments over the available 24-h SEEG recording from our patient. Components of the vectors X were the values of SEEG from each of the 96 electrodes at every time point t. The coefficients A (7, 96) of the model were estimated via the Vieira-Morf partial correlation method. The GPDC value from site j to site i was estimated by: (2)GPDCj→i(f)=|Bij(f)|/σii∑k=1n|Bkj(f)|2/σkk2, where B(f)=I−∑k=1pA(τ)e−i2πfτ, and σ_ii is the standard deviation of the generated fitting error from the model at each site i. Then, ∑j=1,j≠inGPDCj→i(f) measures the effective information inflow (EI_i(f)) to each site i from the rest (n − 1) of the recording sites at each frequency f with resolution of 1 Hz. The EI_i(f) values were then averaged over all frequencies in each of three spectral bands: 0.1–50 Hz [low frequency band (LFB)], 70–110 Hz [high frequency band (HFB)], and 130–170 Hz [ultrahigh frequency band (UFB)] producing the TEI_i values per electrode site i within specific spectral bands. TEI_i has been proven extremely valuable in localization of the epileptogenic focus from interictal periods (23).

To characterize the brain’s global dynamics through SE, the TEI_i measures of connectivity were averaged across all 96 electrode sites for each 60 s SEEG segment, and the TEI_mean (mean) and TEIσ temporal profiles were then calculated every 60 s.

To illustrate the relation over time of the effective local inflow per site versus the effective global inflow, we divided the values of each site’s TEI_i by the values of TEI_mean within each 60 s SEEG segment, thus creating a normalized TEI_i,norm profile over time per site i. In order to also gain a better insight into the relative position of each site’s TEI_i values compared to the ones at other sites, we ranked the TEI_i values of the n = 96 individual sites within each 60 s SEEG segment and plotted the rank of each site i over time (TEI_i,rank profiles).

The TE values per electrode were averaged across all electrode sites over time and the global Teager energy (TE_global) profile was estimated from the available 24 h SEEG record and is depicted in Figure 3. About 1 h before the development of SE, the patient had one clinical seizure that was well captured by the occurrence of a prominent (well above the background) peak in the global TE (black arrow in Figure 3). At the onset of SE, and during its 3 h duration, TE_global values became prominent again and stayed elevated until the end of the medical interventions (magenta arrows in the figure), after which TE_global abruptly assumed low, but still a bit greater than its pre-SE, values over hours (recovery period). Two points are worth noting here: (a) the resetting of TE_global values occurred at the end of the medical interventions, that is, about 1 h prior to the clinically assessed end of SE (green arrow). This raises the possibility for use of TE_global as an early biomarker of the effectiveness of the administered treatment of SE; (b) the post-SE values of TE_global remained elevated compared to its pre-SE values over hours, at least till the end of this recording. This raises the possibility of using the difference between pre and post-SE values of TE_global to objectively measure the duration of the recovery period for patients with SE. Finally, it is noteworthy that no pre-SE trends were captured by the TE profiles, that is, issue of warnings for an upcoming SE based on TE_global would not have been possible.

The TEI profiles and their spatial average (TEI_mean) and standard deviation (TEIσ) were also estimated over the available 24 h SEEG record and over three different frequency bands: LFB = 0.1–50 Hz, HFB = 70–110 Hz, and UFB = 130–170 Hz. The TEI_mean profiles are shown in the left and the TEIσ profiles in the right panels of Figure 4. The TEI_mean profiles from LFB and UFB showed a progressively decreasing trend in values up to the onset of SE (red vertical line) with a reverse (increasing) trend thereafter and a further abrupt increase to a higher and steady level post-SE about 1 h after the onset of clinically assessed patient’s recovery (green vertical line). The TEI_mean profile from the HFB showed a similarly progressive decrease in values pre-SE almost up to the onset of the pre-SE clinical seizure, with resetting of this trend thereafter to high values during SE and an abrupt further increase in values to a steady higher level right after the administration of the anti-SE AED regiment (magenda lines). It is noteworthy that it was not until approximately 1 h after SE’s end that TEI_mean values attained their maximum value in all frequency bands. It is also noteworthy that the variance in TEI values (TEIσ) in all frequency bands (illustrated in the right panels of Figure 4) is higher pre-SE than post-SE, implying the existence of a much more unstable pre-SE state compared to the post-SE state. Finally, it is also noteworthy, from all frequency bands, that a couple of hours after onset of patient’s recovery and assuming resetting of the pathology of dynamics: (a) the TEI_mean obtains its highest values, that is, even higher values than the ones 8 h before the onset of SE, implying that the observed downward trend of TEI_mean values and the route to SE may have started even earlier than the beginning of the recording and (b) the TEIσ obtains its lowest values, that is, even lower values than the ones 8 h before the onset of SE, again implying that instabilities and the route to SE may have started even earlier than the beginning of this recording.

The SOZ of this patient was determined in a multidisciplinary patient management conference taking in consideration input from several recording modalities including EEG. We labeled electrodes within the SOZ as focal while all the rest as non-focal and examined the relation of the TEI over time between focal and non-focal sites as well as between each individual site and the spatial average (global) flow. Toward this goal, we estimated the ranking of TE values over time (TEI_i,rank) at one site i versus the rest of the sites, and the normalized TE values over time at one site i with respect to the global TE value (TEI_i,norm). In Figure 5, we show the TEI_i,norm profiles (left panels) and the TEI_i,rank (right panels) at one focal and one non-focal site. The focal site was within the mesial temporal region, the non-focal site in the posterior orbital frontal and the TE values were estimated in the three frequency bands (LFB, HFB, and UFB). From the left panels of Figure 5, and concentrating on the trends over time of TEI at the focal site with respect to the global EI (focal TEI_norm profile), we see that the focal TEI_norm values were less than 1 (and hence less than the global TEI values) over the whole pre-SE period (7 h) with this difference increasing dramatically as the onset of SE approaches. After onset of SE, this difference starts to decrease and attains values close to 1 (i.e., focal site inflow approximating the global inflow values) in the post-SE period. The administration of AEDs may have assisted with this increase of the focal inflow. Also from Figure 5, we observe that the trend over time of TEI at the non-focal site with respect to the effective global inflow is opposite from the one at the focal site. If we compare the TEI profiles of the focal and non-focal sites, it is evident that: (a) the EI at the focal site is increasingly becoming smaller than the one at the non-focal site as SE approaches (more evident in LFB) and (b) this difference is reversed post-SE.

The corresponding rank (TEI_rank) profiles of TEI at the focal and non-focal sites are illustrated in the right panels of Figure 5 and reinforce our observations from the left panels. The rank of the TEI at the focal site drops from about 50 (out of a total of 96) 7 h pre-SE to 10 at the onset of SE and back to 50 post-SE. The rank of the TEI at the non-focal site shows the opposite behavior over time especially in HFB. It is noteworthy that the pre-SE difference between the ranks of TEI at the focal and non-focal sites is reversed post-SE. This resetting is most prominent in the UFB, in which TEI at the focal site is larger than the TEI at the non-focal site (rank of 60 versus 50, respectively) at the beginning of the recording, this trend is reversed within 1.5 h into the recording en route to SE, and reversed again post-SE.