Keywords epilepsy - epileptic network - resting-state functional MRI - fluorodeoxyglucose -
positron emission tomography - temporal lobe epilepsy
Introduction
Identifying the correct epileptogenic zone with various presurgical tools is vital
in patients with drug-refractory epilepsy undergoing epilepsy surgery for successful
surgical outcomes. Although positron emission tomography (PET) has limited spatiotemporal
resolution, it provides an excellent metabolic signature of the symptomatic zone in
epilepsy.[1 ] Fundamentally, the blood oxygen level–dependent (BOLD) signal and the regional glucose
metabolism represent neuronal bioenergetics and can be correlated with simultaneous
magnetic resonance imaging–PET (MRI/PET). The surgical resection can be planned based
on information from various modalities, and with the advent of neuromodulation techniques
imaging, the planning becomes crucial. Among the multimodality imaging techniques,
resting-state functional MRI (RS-fMRI) has an excellent spatial resolution for in
vivo imaging of dynamic functional networks.[1 ] Since epilepsy is a system network disorder, the degree of network recruitment is
affected by genetic predisposition, duration, age of onset, and severity.[1 ]
The RS-fMRI signals can be represented as spatial maps, power spectrum, and time series;
the time series is sluggish, spectral analysis is comparable, and spatial resolution
is better when compared with electroencephalogram (EEG). The lesion can induce an
aberrant functional network called epilepsy network (Epinet) and recruit large-scale
networks called downstream networks for the seizure propagation. These components
of the network can help predict the temporal evolution of the clinically observed
phenomenon, the kindling of the seizure onset zone (SOZ) and associated long-term
behavioral changes.[2 ]
[3 ]
[4 ]
[5 ]
[6 ]
[7 ]
[8 ]
[9 ]
The aim of the study is to determine whether the Epinets correlate or overlap with
predetermined SOZ from other modalities and the degree of RS-fMRI/PET coupling. A
subgroup RS-fMRI/PET analysis on interictal and postictal groups is also planned.
Methods
The RS-fMRI/PET acquired from January 2016 to December 2018 at a tertiary care referral
hospital is retrospectively reviewed. Written informed consent was obtained from all
the patients before starting the MRI/PET study. The study received approval from the
institutional ethical committee (No. NIMH/DO/[BS & NS DIV.]/2019–2020). Patients diagnosed
with drug-refractory temporal lobe epilepsy (TLE) due to hippocampal sclerosis who
underwent a hybrid MRI/PET study were included. Forty-three patients (25 male and
18 female patients; age range: 15–51 years; mean: 21 years) and 10 healthy controls
(4 male and 6 female participants; mean age: 25 years) were included for this study.
Baseline data including clinical and video-EEG (VEEG), EEG, structural MRI (sMRI)
findings, and the likely SOZ were identified before the scan.
Inclusion criteria were patients with drug-resistant TLE with features of unilateral or bilateral hippocampal
sclerosis on sMRI. Patients underwent simultaneous MRI/PET study with RS-fMRI in the
study protocol and during the interictal period/postictal period.
Exclusion criteria were not consenting, pregnancy, MRI-incompatible devices, ongoing seizures, self-reporting
as being unwell or not cooperative for scan, and image quality not adequate for analysis
of RS-fMRI and/or PET data.
The interictal and postictal phases are interchangeably used in the literature with
the postictal phase extending from a few minutes to days. In this study, subjects
who did not report aura, habitual seizure, or any abnormal episodes before, during
uptake, and MRI/PET scan, were on medication, and cooperative for scan were termed
interictal cases (n = 39). Study subjects who self-reported aura, seizures, and abnormal episodes approximately
3 hours before the scan and who self-reported as being normal during the scan and
were cooperative were termed postictal cases (n = 4). Ictal cases were excluded from this study.
MRI/PET Sequence
Simultaneous RS-fMRI/PET data were acquired (subjects were asked to relax and keep
their eyes open without falling asleep) using a 3T Biograph mMR scanner (Siemens Healthcare,
Erlangen, Germany). The patients rested in a quiet and warm dark room for 15 minutes
before fluorodeoxyglucose (FDG) administration and during the uptake period. Simultaneous
PET and RS-fMRI data acquisition started 30 minutes following the intravenous injection
of the recommended standard dose of 18 F-FDG (5 MBq/kg): (1) three-dimensional T1-weighted magnetization-prepared rapid acquisition
gradient-echo sequence (MPRAGE, 176 sagittal planes, 256 × 240 mm2 fields of view [FOVs], voxel size 1 × 1 × 1 mm3 , repetition time/time to echo/inversion time [TR/TE/TI] 2,300/2.96/900 ms, flip angle
9 degrees, time of acquisition (TA) = 5′14″); (2) the RS-fMRI protocol was a T2*-weighted
single-shot EPI sequence (voxel size 4 × 4 × 4 mm3 , TR/TE = 1,920/32 ms, flip angle = 90 degrees, 250 time points, FOV read = 256, distance
factor = 0, TA= 8′).
Preprocessing : RS-fMRI data were pre-processed using FMRIB's software Library (FSL) tool with following
steps: head motion and slice-timing corrections, intensity normalization, removing
nonbrain tissue by brain extraction tool (BET). Moreover, data were temporally high-pass
filtered at 0.01 Hz. Individual functional scans were registered to the patient's
high-resolution anatomical scan using linear registration and optimized using boundary-based
registration. All subjects had less than 0.5-mm head motion displacement in any direction.
Twenty independent component analyses (ICAs) were generated by ICA. These oscillating
subsignals/components can represent either brain networks or noise. The ICA was divided
as shown in the following, and the spatiotemporal and power spectrum features are
given in the algorithm ([Table 1 ]) developed from reported published literature.[2 ]
[3 ]
[4 ]
[5 ]
[6 ]
[7 ]
[8 ]
[9 ]
Table 1
Network analysis algorithm table
Noise
Large-scale neuronal network/typical RSN
Downstream epileptogenic propagation/atypical RSN
Epinet/SZ
Spatial maps
Not respecting anatomical boundary, primarily over vessels/CSF/WM, alternating slices/not
over gray matter/brain–air interface/EPI dropout in the phase encoding direction/alternate
activation and deactivation in only one slice.
Spatial distribution consistent with machine-generated artifacts such as skipping
slices
Clear known bilateral/unilateral RSN are established[25 ]
Primary sensory-motor networks, parietal networks, frontal networks, visual networks,
the default mode network, and the deep gray networks located with the bilateral putamen
and bilateral mesial thalami
RSN location but incomplete spatial coverage
RSN like spatially but overlapping with SOZ
Localized to gray matter but not RSN pattern unilateral. The alternating positive
and negative hemodynamic alteration, tail extending outward into WM toward ventricles.
Irregular border
Should overlap with area localized on EEG/PET/MRI/MEG but not with RSN
Temporal graph
Erratic time course/regular oscillation pattern
Majority of cycles have period changes < 50% of baseline
Low oscillation frequency < 0.073 Hz
The slow, regular, and smooth oscillating pattern
Cycles have no spikes and change in period < 50% period value between cycles
High oscillation frequency > 0.073 Hz
Smooth sinusoidal oscillation overlaid with regular milder frequency disruption/irregular
frequency sharp contoured bursts returning to normal
Cycles have < 50% less change in the period between cycles
High oscillation frequency > 0.073 Hz
Erratic irregular low and high frequency with sharp contours of faster frequency than
baseline with or without return to baseline
Cycles have a change in a period > 50% of baseline between cycles
Power spectrum
Powerband out of low range along with high range bands > 0.039 Hz if the cardiac or
respiratory band
Powerband in low range < 0.039HZ with high range band if near EZ
High band with low range to > 0.04 Hz
High band with low range to > 0.04 Hz
Our observation
Spatial maps
Physiological noise: CSF/vessel location
Artifactual: nonbiological behavior/unknown pattern
Some of these networks may be missing or have deactivated morphology—indicate dysfunction
of normal network
The pattern was intermediate of large scale and Epinet and may be normal networks
that have been upregulated for seizure propagation
Unknown network close to the EZ—represent disrupted local brain function, If > 1 Epinet
is noted, it represents subnetwork if all these are in the same location and when
overlaid a bull's eye pattern is noted
If they are noted in different locations, then it indicates the possibility of multifocal
EZ
Power spectrum
Peaks start with very low range power (y-axis) in lower frequency (0.01–0.08 Hz) and
intermediate band (0.08–0.16 Hz) and peaks in very high range power in the high-frequency
band (0.16–0.24) in the spectrum
Peaks show low to high range power (y-axis) in low-frequency band within < 0.04 Hz
(x-axis) and peaks fall flat in power in the intermediate frequency band (x-axis)
in the spectrum
Peaks show low to range high range power (y-axis) in low-frequency band > 0.04Hz (x-axis)
and peaks continue with very low range power in the rest of intermediate and high-frequency
bands in the spectrum
Peaks show low to high range power (y-axis) in low-frequency band > 0.04Hz (x-axis)
and peaks continue with very low range power in the rest of the high and intermediate
frequency band in the spectrum
Temporal graph y-axis normalized response
−4 to +4 range
−10 to +5 range
−4 to +4 range
−2 to +4 range
Abbreviations: CSF, cerebrospinal fluid; EEG, electroencephalogram; EPI, echo-planar
imaging; EZ, epileptogenic zone; MEG, magnetoencephalography; MRI, magnetic resonance
imaging; PET, positron emission tomography; RSN, resting-state network; SOZ, seizure
onset zone; SZ, seizure zone; WM, white matter.
Power spectrum: Conversion on the x-axis–(Hz/100) × 0.01 = (Hz) Eg: 4 Hz/100 = 0.04 Hz
Temporal graph conversion on the x-axis–TR = 1.920 s × 250 time points = 480
Analysis and Interpretation
Resting-state SOZ/Epinets were identified by spatial features if they are primarily located within gray matter
but not in resting-state network (RSN) spatial pattern and were correlated with suspected
seizure foci on MRI/PET/EEG/VEEG.[2 ]
[3 ]
[4 ]
[5 ]
[6 ]
[7 ]
[8 ]
[9 ]
Artifacts/noise : Signal arising from the motion, susceptibility, sagittal sinus vein, cerebrospinal
fluid, white matter regions, cardiac, and respiratory cycle ([Fig. 1A ]).
Fig. 1 (A ) Noise-independent component. (B ) Large-scale network. In this example, default mode network (DMN) has been represented.
(i) In the spatial image, the BOLD signals represent known bilateral DMN. (ii) The
network time course of the BOLD signal. (iii) The power spectrum of the frequency.
Large-scale network/RSN/neuronal networks : We correlated with the known large-scale networks on spatiotemporal and power spectrum
parameters and deduced the 10 large-scale networks with the highest correlation ([Fig. 1B ]).[2 ]
[3 ]
[4 ]
[5 ]
[6 ]
[7 ]
[8 ]
[9 ]
Downstream/atypical RSN : Networks with weaker correlation with large-scale networks on spatial parameters
but having temporal and power spectrum profile intermediate to large scale and Epinets
([Fig. 2A ]).[2 ]
[3 ]
[4 ]
[5 ]
[6 ]
[7 ]
[8 ]
[9 ]
Fig. 2 (A ) Downstream RSN. (B ) Epinet. These are aberrant networks that have a distinct profile compared with normal
RSN.
The first step was a correlation with the known large-scale networks of RSN networks.
We then filtered out noise/artifacts and hand-labeled downstream RSN. The unknown
networks were analyzed as potential Epinets based on spatiotemporal series and power
spectrum definition. The potential Epinet ICA was then correlated for overlap with
SOZ. If more than one Epinets were found, they were overlaid with each other. The
seizure network can fragment into subnetworks. The same processing technique was used
in healthy controls to rule out false-positive/noise mimicking Epinets.
This is an exploratory study on a large case series, and analysis was done on a per-patient
basis. Detailed results are available in [Supplementary Table S1 ] (available in the online version only).
Results
Of 43 patients, 39 patients were in the interictal phase and 4 patients were in the
postictal phase. Since 14 cases did not yield any Epinets in the interictal phase,
only 25 cases were analyzed. The likely SOZ was based on clinical EEG, VEEG, and mesial
temporal sclerosis (MTS) on sMRI ([Supplementary Table S1 ], available in the online version only). In the interictal subgroup (n = 25), 13 had left MTS, 10 had right MTS, and 2 had bilateral MTS. In the postictal
subgroup (n = 4), there were two each of left MTS and right MTS. Details of 20 ICA classification
are given in [Supplementary Table S1 ] (available in the online version only).
Algorithm and Our Observations
Our observation of spatiotemporal and power spectra profiles is given in [Table 1 ]. Noise could be either nonbiological or physiological contamination from cerebrospinal
fluid, artery, or vein ([Fig. 1A ]). Few patients had missing large-scale networks or had a deactivated morphology
([Fig. 1B ]). The downstream networks had a behavior intermediate between large scale and Epinets
([Fig. 2A ]). The Epinets were close to the SOZ, indicating a disrupted local brain network
([Fig. 2B ]), and were often more than one. When these were overlaid upon each other, they formed
a bull's eye pattern over the SOZ ([Fig. 6 ]). Figures of right MTS ([Fig. 3 ]), left MTS ([Fig. 4A ]), interictal phase, and postictal phase ([Fig. 4B ]) are provided. The power spectrum of Epinets was higher than the normal RSN. On
a time series graph when the normalized response (y-axis) was analyzed, the large
scales had values ranging from −10 to +5, whereas downstream had values ranging from
−4 to +4 and Epinets had values ranging from −2 to +4 range, indicating that Epinets
were faster in frequency.
Fig. 3 A case of drug-refractory epilepsy (DRE) in the interictal phase of left MTS (case
9). In the spatial map, the BOLD signals are localized to gray matter. Alternating
positive and negative hemodynamic alteration with irregular border and overlaps with
the temporal region.
Fig. 4 (A ) A case of DRE in the interictal phase of right MTS (case 24). (B ) A case of DRE in the postictal phase (case 1) of left MTS and left parietal gliosis.
Healthy control ICA versus epilepsy ICA : None of the healthy controls had Epinet/downstream-like network profile. An average
of 7 to 13 large-scale networks were noted in controls, and patients had 3 to 7 large-scale
networks and 2 to 7 downstream networks. Noise components were similar across both
the groups.
Bilateral versus unilateral MTS : The number of Epinets and downstream networks was similar across bilateral, right,
and left MTS. In the two cases with bilateral MTS, case 32 had left lateralization
and case 37 had right lateralization. When they were correlated with PET and RS-fMRI,
the lateralization was similar. In case 2 and case 12, the sMRI showed MTS on right
and left sides, respectively, with PET showing bilateral hypometabolism (more significant
on the side of MTS). Epinets too showed lateralization to the side of MTS.
Activations/deactivation pattern on RS-fMRI and PET metabolic correlation : There was no significant difference in the pattern of activation and deactivation
in the Epinets of both postictal and interictal groups. The RS-fMRI pattern did not
correlate with the metabolic pattern on PET. The SOZ showed positive BOLD activation
on RS-fMRI, whereas PET showed decreased metabolism (negative correlation).
Interictal versus postictal : The yield of finding Epinets in the 9-minute scan was 100% (4/4) in the postictal
phase and 61% (25/39) in the interictal phase.
When the pattern of behavior was noted between the interictal and postictal groups,
the spatiotemporal and power spectrum behavior was similar but the likelihood of finding
Epinets was more in the postictal group. PET had hypometabolism in the SOZ across
both groups. We did not have any cases with prolonged postictal states such as drowsiness,
confusion, and disorientation in this series ([Figs. 2B ], [3 ], and [4A ]).
Lateralizing value of RS-fMRI with other modalities : RS-fMRI overlapped with SOZ and BOLD activation lateralized to the side of SOZ.
Epinets in RS-fMRI and PET correlation : On RS-fMRI, Epinets were more in the postictal phase but PET was similar in both
subgroups and had hypometabolism at SOZ.
Epinets overlapping with SOZ (right or left MTS) : With MTS as the structural lesion, all the Epinets were overlapped with each other
and correlated with SOZ. The other areas activated in the network may be for seizure
propagation or potential epilepsy zones that need further study ([Figs. 3 ] and [4A ]).
Postoperative cases : There were two cases, one each in interictal and postictal groups, with persistent
drug-refractory epilepsy in the postoperative period. Residual Epinets were noted
in these cases. We did not have pre- and postoperative cases for comparison ([Fig. 5A,B ]).
Fig. 5 (A , B ) A case of DRE postoperative case of right and left MTS in interictal phase (case
37) and postictal phase (case 4), respectively.
Fig. 6 Total of three Epinets were noted in case 9 with left MTS on MRI among the 20 ICA
due to its time and power spectral behavior and activation is more in the left temporal
lobe forming a bull's eye pattern.
Discussion
The International League Against Epilepsy defines a seizure as a transient occurrence
of signs and/or symptoms due to abnormal excessive or synchronous neuronal activity
in the brain, which are easily perceived by the patient or observed by others. A seizure
stops due to a variety of potential mechanisms such as energy substrate depletion,
receptor desensitization, depolarization block, desynchronization of neuronal networks,
GABAergic and non-GABAergic inhibition, hyperpolarizing sodium–potassium pump potentials,
and inhibitory neuromodulators at subcellular levels. There is no clear-cut demarcation
between the interictal and postictal phases. The postictal phase is the period after
a seizure that can overlap with the interictal phase. Behavioral and cognitive change
in these phases may be a continuum.[10 ]
[11 ]
[12 ]
EEG–PET studies in the postictal phase have shown that although we expect a postictal
hypometabolism and normal EEG baseline, 10% had persistent hypermetabolism on PET
and only 31% showed a normal EEG baseline after 12 to 24 hours.[13 ] The post-ictal period is affected by type, duration, and number of seizures. EEG–RS-fMRI
studies have shown widespread BOLD changes before clinical onset and after clinical
resolution of the seizure[13 ] and the postictal showed either abrupt return, stuttering, or no return to baseline.
The timescale is in milliseconds, seconds, or minutes for EEG–RS-fMRI/PET, hence the
variability across these modalities. All these dynamic changes put together may result
in behavior change and is subjective. Clinically, there is a continuum of aura or
a fugue state to seizure to a normal state again, which is highly variable across
patients.
The interictal period could represent an abnormal baseline, a fact established by
interictal PET showing hypometabolism, although it may appear normal clinically on
sMRI and EEG. An EEG–RS-fMRI study has shown abnormal activations, although clinical
EEG had no interictal epileptic discharge (IED).[14 ] Our study has explored the role of RS-fMRI and clinical EEG and whether a linear
correlation exists in the interictal phase. When the subgroups were compared, Epinets
were more in number in the postictal group. In 39% of the cases, there were no identifiable
Epinets, which might have been due to a short scan interval of 9 minutes. Postictal
networks may represent pathological upregulated networks that lag to return to baseline
and doing a scan during this phase can increase the yield. Epinets represent an aberrant
network at baseline and are footprints of earlier seizure activity. PET was similar
in both the subgroups because all the transient neurohemodynamic physiological changes
in the timescale of seconds may not be captured. The pattern of deactivation and activation
on RS-fMRI did not linearly correlate with the hypo- and hypermetabolic pattern of
PET. This is expected as SOZ in the interictal period has decreased metabolism but
being a seizure focus has activated BOLD signals. In an EEG–RS-fMRI/PET study, the
fast-wave component of EEG correlated with positive BOLD and slow-wave component increased
local field potential (LFP) and decreased mean firing rate of EEG during spike period
correlated with BOLD deactivation. However, deactivation on RS-fMRI could also mean
vascular steal phenomenon, venous flow, and inhibitory neurotransmitter activity.[15 ] Lateralization by RS-fMRI/PET for SOZ correlated well. In an EEG–RS-fMRI study,
in the preictal, postictal, and interictal states, the lateralization to SOZ was maintained.[13 ] In the two cases with bilateral MTS, the lateralization with sMRI, PET, and RS-fMRI
was similar as they represented the same brain state.
The smaller timescales and dynamic perturbations caused by epilepsy transients are
well captured by RS-fMRI, whereas PET represents global change. An EEG–PET showed
EEG perturbations to be far more extensive than the metabolic changes.[16 ]
[17 ] These epilepsy transients represented by Epinets indicate baseline abnormality capable
of having a transition from interictal to ictal state.[18 ] The number of postictal Epinets is more than in the interictal phase.
Research on normal RS-fMRI has shown that the BOLD changes can be subdivided into
slow-wave 5 to slow-wave 1,[19 ] and that Epinets can be discriminated with the highest accuracy at 0.01 to 0.073 Hz.[20 ] This analogy is similar to the frequency spectral bands of EEG with a high-frequency
band corresponding to epilepsy. Noise (physiological and artifactual) can be either
regular or erratic. Epinets have a biological profile of normal RSN but with higher
frequency and magnitude. Interpreting Epinets needs correlation with spatial map,
power spectrum, and time series, and a clinically relevant inference can be derived
by a change in pattern in the post-surgery phase. A transition from higher to lower
frequency band and decrease in oscillation frequency on time series graph and lack
of BOLD signal at the resected site carries a good prognosis. The change in the RS-fMRI
profile may not be abrupt but the tendency of the RS-fMRI profile to improve or persist
can dictate the outcome.[2 ]
[3 ]
[4 ] The epileptogenic lesion from sMRI and RS-fMRI zone by overlapping Epinets at SOZ
correlate partially. Hence, when surgery is planned the strategy should be not only
resection of the epileptogenic lesion but also lesion surrounding area identified
by Epinets. RS-fMRI shows the extensive area around a lesion much similar to the ECOG
recordings showing dynamic spikes which are not created equal and are much larger
in area than the ictal zone itself. The ECOG study also concludes that surgical resection
of the current zone with failure to remove the potential zone can lead to recurrence.
Although identification of potential zone may be difficult, by doing a follow-up RS-fMRI
the tendency to normalize in the postoperative period carries a good prognosis.[2 ]
[3 ]
[4 ] PET studies have shown that areas of extensive hypometabolism in extratemporal foci
resulted in surgical failures and may indicate the persistent connection of the epileptic
zone with the rest of the involved network areas.[21 ] If an RS-fMRI–PET study is done, it acts as a guide to plan area of resection and
monitoring. In a study on RS-fMRI, if the activations were nonlateralizing and more
generalized networks, vagus nerve stimulator/responsive nerve stimulator was planned
rather than surgery.[22 ]
Compared with EEG, RS-fMRI is easier to acquire and can provide crucial functional
in vivo information similar to that obtained by EEG–MRI/PET. With PET having a behavioral
correlation and RS-fMRI giving network information, clinical translation of this modality
in epilepsy surgery is feasible. RS-fMRI–PET may also be a viable option for cognitive
studies by using a methodology of processing RS-fMRI appropriate to the hypothesis.
Many analysis tools have been used in RS-fMRI, with ICA having more accuracy and high
specificity of 92% to detect epileptogenic foci.[23 ]
EEG–PET study has shown that gamma oscillations of EEG and PET had a negative correlation
very similar to our RS-fMRI–PET study.[24 ] In trimodality imaging, an EEG cap creates an artifact on PET attenuation correction
images and EEG acquisition is labor intensive process. RS-fMRI–PET allows detection
of smaller seizure foci with more specificity. An RS-fMRI–GABA PET can help correlate
network neurotransmitter information.
In summary, this study revealed a few important findings. The yield of finding Epinets
is more during the postictal period (seizures 3–6 hours before scan) than in the interictal
period (seizure-free period). Although PET in these cases did not show visual ictal
changes and was reported similar to other interictal PET, RS-fMRI showed far more
changes. The overlap between large-scale and downstream networks was noted, indicating
that epilepsy propagation can involve cognition networks and cause cognitive dysfunction
in epilepsy. Residual Epinets analysis can act as a searchlight in surgical failures
and a pre- and postoperative RS-fMRI assessment may be helpful. Shorter and faster
acquisition of RS-fMRI very similar to EEG is possible with multiband sequence, and
with robust data-driven techniques, ICA can be classified for clinical translation.
