Diffusion tractography predicts propagated high-frequency activity during epileptic spasms

Summary

Objective:

Determine the structural networks that constrain propagation of ictal oscillations during epileptic spasm events, and compare observed propagation patterns across patients with successful or unsuccessful surgical outcomes.

Methods:

Subdural electrode recordings of 18 young patients (age 1–11 years) were analyzed during epileptic spasm events to determine ictal networks and quantify the amplitude and onset time of ictal oscillations across the cortical surface. Corresponding structural networks were generated with diffusion MRI tractography by seeding the cortical region associated with the earliest average oscillation onset time, and white matter pathways connecting active electrode regions within the ictal network were isolated. Properties of this structural network were used to predict oscillation onset times and amplitudes, and this relationship was compared across patients who did and did not achieve seizure freedom following resective surgery.

Results:

Onset propagation patterns were relatively consistent across each patients’ spasm events. An electrode’s average ictal oscillation onset latency was most significantly associated with the length of direct corticocortical tracts connecting to the area with the earliest average oscillation onset (p < .001, model R² = 0.54). Moreover, patients demonstrating a faster propagation of ictal oscillation signals within the corticocortical network were more likely to have seizure recurrence following resective surgery (p = .039). Ictal oscillation amplitude was also associated with connecting tractography length and weighted fractional anisotropy (FA) measures along these pathways (p = .002/.030, model R² = 0.31/0.25). Characteristics of analogous corticothalamic pathways did not show significant associations with ictal oscillation onset latency or amplitude.

Significance:

Spatiotemporal propagation patterns of high-frequency activity in epileptic spasms align with length and FA measures from onset-originating corticocortical pathways. Considering data in this individualized framework may help inform surgical decision making and expectations of surgical outcomes.

1 |. Introduction

Epileptic spasms are a brief and repetitive seizure semiology that represent one of the most prevalent forms of infantile seizures^1,2: single spasm events may manifest as a momentary flexion-extension movement, and spasms will often cluster together, occurring every few seconds over the course of several minutes. Patients with recurrent spasms may subsequently develop more severe epileptic phenotypes and significant cognitive regression,^3,4 and therefore the rapid identification and treatment of their condition bears particular urgency. In cases where medical management fails to control spasms, invasive surgical approaches may be successful,⁵ and past work has stressed the importance of early surgical intervention for minimizing epilepsy’s cognitive consequences.⁶

Unfortunately, it can be challenging to localize surgically targetable foci in the brains of children with epileptic spasms. The majority of cases show normal or nonspecific neuroimaging findings,⁵ and the rapid spread of motor and electroencephalographic activity during each brief spasm leaves little time to confidently distinguish sites of early epileptogenic vs. later propagated activity. Moreover, the condition is seen in a population of etiological and developmental heterogeneity, as spasms are associated with broad dysplastic, genetic, and encephalopathic causes,⁷ and may persist well into childhood for some patients. Despite these difficulties, the repetitive nature of spasms may afford increased opportunities for localization. If the characteristics of ictal electroencephalography are better understood and contextualized across different patients, ictal recordings of spasms could provide much leverageable information towards identifying surgical targets.

Past research has revealed distinctive frequency components evident on subdural recordings of epileptic spasms, characteristically involving activity in the gamma or high-gamma range superimposed on lower-frequency sub-delta activity^8,9 that spreads non-contiguously across cortical regions.¹⁰ The anatomical onset of these signals varies across patients, with spread to motor cortices corresponding to onset of the shared semiology typically seen in the condition.⁸ The pathological mechanism generating these signals and the connectivity networks facilitating their propagation remain unknown, but previous detections of localized metabolic abnormalities¹¹ and acknowledgment of semiologic bilaterality in epileptic spasms have driven hypotheses of brainstem¹² and thalamic¹³ involvement.

Related research on ictal spread in other seizure subtypes and etiologies has utilized tractography derived from diffusion-weighted imaging (DWI), revealing the white matter pathways that enable transmission of pathologically-active seizure activity across the brain. High-frequency activity during temporal lobe epilepsy has been shown to spread from the epileptogenic zone in a manner proportional to the streamline count¹⁴ and fractional anisotropy (FA)¹⁵ of pathways extending out of that region. Similarly, larger-scale efforts to model whole-brain activity and seizure propagation have successfully utilized measures of tractography streamline length and count connectivity between regions.^16–18 The present study considers ictal epileptic spasm signals on subdural recordings in the context of whole-hemisphere white matter connectivity, and aims to reconcile its agreement with the possible brain networks facilitating its propagation.

2 |. Methods

2.1: Patient population

2.1.1: Patient demographics and medical course

Patient demographics are summarized in Table 1. Present analyses consider retrospective data from children who underwent extraoperative electrocorticography (ECoG) monitoring for resective epilepsy surgery between 2009 and 2019 at the Children’s Hospital of Michigan, Detroit MI, USA. 18 patients (12 males, 6 females, age range 1–11 years) met inclusion criteria, having a past history of epileptic spasms, at least one episode of epileptic spasms during ECoG monitoring, and presurgical collection of suitable diffusion-weighted MRI as described in section 2.3.1. Any antiepileptic drugs in use at monitoring initiation were weaned until sufficient observations of ictal events allowed decision making for the subsequent resective stage of surgery. After surgery, each patient was followed up for at least 2 years and classified by International League of Epilepsy (ILAE) standards¹⁹ as having either achieved seizure freedom (ILAE class 1) or experienced seizure remission (ILAE class ≥ 2). Nine patients were categorized as achieving ILAE class 1 outcome, while the nine remaining patients did not achieve this level of seizure freedom. Mean patient ages for each outcome group were 5.44 and 4.38 years, respectively, and this age difference was not statistically significant on ANOVA (p=.54).

Table 1:

Patient characteristics across surgical outcome groups

	ILAE outcome = 1	ILAE outcome ≥ 2	Group comparison
Patient characteristics:
Number of patients	n = 9	n = 9
Extent of resection	2UL, 3ML, 4SH	2UL, 3ML, 4SH
Sex distribution	4F, 5M	2F, 7M
Mean age (s.d., range)	5.44 years (4.32, 1–11)	4.38 years (2.69, 2–8)	p = .54
ECoG characteristics:
Mean center of ictal frequency range (s.d.)	108 Hz (14)	96 Hz (23)	p = .24
Mean width of ictal frequency range (s.d.)	72 Hz (16)	73 Hz (28)	p = .92
Mean oscillation duration (s.d.)	454ms (147)	455ms (102)	p = .99
Mean oscillation onset window (s.d.)	576ms (333)	639ms (255)	p = .66
Mean oscillation onset stability (s.d.)	r = 0.75 (0.07)	r = 0.77 (0.10)	p = .58
DWI characteristics:
Mean tractography length (s.d.)	59.8mm (7.5)	64.4mm (13.3)	p = .38
Mean tractography weighted FA (s.d.)	0.34 (0.03)	0.35 (0.06)	p = .82

Abbreviations: ILAE = International League Against Epilepsy, s.d. = standard deviation, F = female, M = male, UL = unilobar resection, ML = multilobar resection, SH = subtotal hemispherectomy, FA = fractional anisotropy

2.1.2: Ethics statement

All described analyses were carried out retrospectively and had no influence on surgical decision making. The Institutional Review Board at Wayne State University approved this study, and written informed consent was obtained from the guardians of all children involved.

2.2: Spasm-associated electrocorticography data

2.2.1: Subdural electrode recordings

Platinum strip and grid electrodes (10-mm center-to-center distance) were surgically placed on each patient’s pial surface based on suspected seizure involvement, as determined by presurgical evaluation including but not limited to scalp EEG, imaging, and consideration of seizure semiology. Extraoperative ECoG data were sampled at 1000 Hz with amplifier bandpass of 0.016 to 300 Hz. Intraoperative photographs showing visible electrode locations were taken prior to craniotomy closing, and electrode location was further verified with x-ray and/or CT scan.^20,21 The mean number of analyzed, non-artifactual electrodes per patient was 118 (range 99–134). Patients were monitored for epileptic activity for 2 to 4 days following taper of antiepileptic medications. Overhead patient video was recorded for the duration of the monitoring period to corroborate clinical periods of ictal activity in the ECoG recording.

2.2.2: Offline processing of spasm-associated intracranial EEG data

Past studies of ECoG ictal activity during epileptic spasms reveal a multifaceted waveform containing both high- and low-frequency components.⁸ In the interest of maximizing the temporal accuracy of our detected signals, present analyses focused on the propagation of high-frequency ictal components. Epileptic spasm events that occurred during monitoring were identified by a clinical electrophysiologist (E.A.) reviewing the video record and ECoG tracing. Data from any electrodes exhibiting excessive noise were removed, and all data were re-referenced to a common average of the remaining electrodes.^8,9 Tracings from all electrodes were notch-filtered at 60 Hz and viewed simultaneously with a 30 Hz high-pass filter, and time points preceding individual spasm events were tagged approximately 1 second prior to the visually-perceived onset of the first high-frequency activity in any electrode. Six-second epochs centered on this event-preceding time point were then extracted. These single-event epochs were visually inspected, and any epochs containing large contaminating artifacts attributable to patient or equipment movement were removed from the analysis.

2.2.3: Automated detection of electrode-specific spasm onset

Past research has identified patient-specific frequency peaks and bandwidths during ictal oscillatory activity,²² and has stressed the importance of patient-appropriate parameter selection.^23,24 Therefore, we first sought to characterize the patient-specific frequency activity of the events under study. For each spasm event, time-frequency dynamics were extracted from each electrode by Morlet wavelet convolution at frequencies 1 Hz through 200 Hz.²⁵ A 2-second window of each electrode’s convolution, centered on its amplitude peak timepoint across the full extracted range, was taken and averaged across electrodes to estimate the dominant spectral frequencies within each event. These electrode-averaged spectrums were then averaged across each patient’s spasm events, and the top 2% of values in the resulting time-by-frequency matrix were localized to reveal a patient-specific ictal fast oscillation frequency range suitable for subsequent automatic detection of ictal onset (Figure 1C, Table S1).

Figure 1:

ECoG data processing pipeline. Example data shown from patient 03 (top) and patient 16 (bottom). A: Patient locations of subdural electrodes during pre-resection monitoring. B: Example of all electrode tracings during a single spasm event. C: Spectral density of spasm activity, averaged across all events and all electrode tracings after centering these data on their individual broadband peaks. Purple area represents the top 2% of values, and shown limits of this area were used to generate patient-specific bandpass filters for ictal fast oscillation analysis. D: Single-spasm, single-electrode examples of spasm signal detection using bandpassed data. Example tracings are shown and colored for a late onset electrode (red, top) and an early onset electrode (green, bottom). E: Same spasm event as in B, shown with signals colored by gradient of their detected onset time, green-to-red for earlier-to-later detected onsets. Extreme outlier onset detections and electrodes with insufficient involvement across all spasm events are removed as described in Section 2.

Each patient’s 6-second ECoG epochs were then band-passed within their specific frequency range, and the instantaneous amplitudes of these data were extracted by Hilbert transform. To account for variability in electrode frequency and amplitude characteristics, each electrode’s amplitude data were z-scored to a period of non-spasm baseline activity. These baseline periods were comprised of concatenated instantaneous amplitude values from the one-second window beginning 2 seconds prior to the marked pre-spasm timepoint in each non-artifactual epoch. Consistent with past research,¹⁸ ictal fast oscillation onset times for each electrode were defined as the first moment that signal reliably exceeded an arbitrary threshold, specifically an instantaneous amplitude equaling or exceeding a z-score of 5 that stayed above-threshold for over 150ms (Figure 1D). Similarly, ictal fast oscillation offset times were defined as the first moment following onset where the instantaneous amplitude fell below a z-score of 5 and remained sub-threshold for over 150ms.

2.2.4: Quantification of ictal fast oscillation characteristics across spasm events

Differences in ictal oscillation onset latency were calculated for any two electrodes meeting detection criteria within a spasm event. These differences were averaged across spasm events to approximate the typical latency between any given electrode pair. The electrode with the earliest average onset latency relative to all other electrodes was designated as the “onset electrode” for that patient. The integral of the instantaneous amplitude between detected onset and offset points was also calculated and averaged across events to determine an electrode’s mean amplitude measurement.

Given that the repetitive nature of epileptic spasms allows multiple chances to accurately detect oscillation instances, and acknowledging the broad influence that early type-1 errors (i.e. “false detections”) could have on our latency measures of interest, we applied further measures to clean detection data at the expense of possibly introducing type-2 errors (i.e. “missed detections”). Spasm events that involved few (<25%) electrodes compared to the event with maximum electrode involvement were removed (2.3% of spasms per patient on average), and electrodes that were involved in few (<25%) spasm events compared to the electrode with maximum spasm event involvement were removed (28.3% of electrodes per patient on average), ensuring a minimum level of network density across different spasm events. Ictal oscillation onset and offset times relative to their means within each spasm were also pooled across a patient’s total spasm events, and detections that exceeded 4 standard deviations outside this patient-specific set of event-normalized times were removed (0.8% of oscillations per patient on average). This conservative cutoff value was chosen to best preserve true ictal-associated oscillations while removing inconsistent extreme values (Figure S1), and removals at this threshold did not broadly impact the anatomical location of any patient’s determined “onset electrode.”

2.3: Presurgical diffusion MRI data

2.3.1: MRI acquisition and preprocessing

All patients were scanned in a GE Signa 3T scanner with an eight-channel head coil. DWI images were acquired at repetition time (TR)=12500ms and echo time (TE)=88.7ms, with a field of view (FOV)=24cm, a 128×128 acquisition matrix, and contiguous 3mm-thickness axial slices covering the whole brain, using 55 isotropic gradient directions with b=1000 s/mm², number of excitations=1, and a single b=0 image. T1-weighted (T1w) images were also acquired using a fast spoiled gradient echo sequence (FSPGR) at repetition time (TR)=9.12ms, echo time (TE)=3.66ms, and inversion time (TI)=400ms, with slice thickness of 1.2mm and a planar resolution of 0.94×0.94 mm². DWI was first denoised,²⁶ then corrected for inhomogeneity, distortions, and bias using the FMRIB software library (FSL) eddy tool²⁷ and automated segmentation tool (FAST)²⁸ package. Anatomical brain images were extracted from T1w scans using FreeSurfer software reconstruction²⁹ and co-registered to the corrected DWI averaged over all gradient directions. To facilitate the study of white matter pathways, the approximate grey matter white matter boundary was also identified on the T1w images using the FMRIB software library (FSL) automated segmentation tool (FAST) package.

2.3.2: Region of interest generation

Subdural electrodes were localized to corresponding cortical regions on preoperative T1w and DWI scans using a combination of post-implantation CT or x-ray as previously described,^30,31 and locations were validated using photographs taken prior to craniotomy closure.³² From each localized 3D coordinate, a vector oriented inward to the brain was determined based on a reconstruction of the cortex generated with FreeSurfer software: cortical reconstructions were smoothed using a 10mm³ box kernel, and from this smoothed reconstruction, surface normal vectors from the 10 nearest vertices to each electrode location were averaged. Starting at each electrode location, 4mm-radius spherical regions were generated at 10 equidistant 1mm steps along each electrode’s vector and summed to generate cylindrical regions, with any overlapping voxels assigned to the nearest electrode’s region (Figure 2C). Thalamic regions were also extracted from FreeSurfer parcellations of each patient’s T1w scan.

Figure 2:

Electrode-associated region of interest (ROI) generation and examples. A: Example of electrode locations from Patient 16. B: Example of the smoothed brain surface with inward vectors used to identify electrode-associated grey-matter/white matter boundary areas. C: Resulting electrode-associated ROI map generated for Patient 16, as seen in example coronal and axial slices. The ROI corresponding to the earliest onset electrode is shown in a lighter green shade. D: Spatial distribution of earliest-onset electrode from all patients with left hemisphere sampling. Electrodes have been mapped to their approximate location on the FreeSurfer template brain. Onset locations from patients who achieved postsurgical seizure freedom are shown in shades of green, while patients who did not are shown in shades of red. E: Spatial distribution of earliest-onset electrodes from all patients with right hemisphere sampling, presented identically to D.

2.3.3: Processing and measurement of metrics from diffusion data

Streamline tractography data approximating white matter pathways were generated using MRTrix3 software.³³ A fiber response function³⁴ was first derived from the DWI data, and fiber orientation distributions were estimated with spherical deconvolution.³⁵ Candidate tractography seeding points were determined by the intersection of the cylindrical region of interest corresponding to the “onset electrode” and the grey matter white matter boundary (Figure 2C). 1×10⁸ seeding points were randomly placed within this electrode-associated cortical boundary region and streamlines were generated based on a second-order integration over fiber orientation distributions (iFOD2) algorithm.³⁶ In efforts to maximize biological plausibility and mitigate the crossing-fiber problem, we applied anatomically constrained tractography³⁷ when generating streamlines and restricted their maximum angle to a threshold of 35 degrees across half-voxel-width steps. Streamlines originating from the seeded region were then sorted into electrode-specific corticocortical pathways if their endpoint fell within the ROI associated with another electrode. To reduce the risk of false-positive short U-fibers tractography in our single-shell DWI data, electrodes located less than 15mm by Euclidian distance from the “onset electrode” were excluded from this sorting process.

As a theory-driven alternative model of structural propagation, a comparative corticothalamic network linking cortical electrodes was also generated by analogous methods. Tractography was seeded from cortical boundaries associated with all electrode ROIs, and streamlines were selected based on terminations within the thalamus. These streamlines were then combined into cortico-thalamo-cortical pathways by summating measures along the onset-electrode-to-thalamus pathway and each paired electrode-to-thalamus pathway.

Two metrics were extracted from all generated tractography: pathway length and an average representative FA value underlying the pathway. Length measurements of each pathway were calculated by reducing each involved streamline to 12 equidistant points, averaging corresponding values to generate a centroid streamline, and determining its length. To further limit the risk of false-positive streamlines influencing volumetric representations of each pathway, pathways were reduced to their central regions by removing outlier streamlines whose average pointwise distance to this centroid exceeded 1 standard deviation above the mean pathway streamline distance-to-centroid.³⁸ For pathway FA measurement, FA values were mapped to each voxel in the DWI data, and values underlying each central pathway were weighted based on the total percentage of pathway streamlines that intersected any given voxel, thus deriving an average metric representing the most commonly-tracked white matter areas between two electrodes.

2.4: Statistical Approach

2.4.1. Within-patient characterization of ictal fast oscillation activity

Basic descriptive statistics, including mean duration of fast oscillations and mean range of electrode onset times spanning individual spasms, were calculated for each patient. To measure each patient’s onset pattern similarity, Pearson correlations between each electrode’s single-spasm onset time and the mean electrode onset time across all other patient events were calculated. The median values of these spasm-specific autocorrelations were taken as a metric of propagation consistency for each patient, and 25^th and 75^th percentile boundary values were also taken as indicators of data spread in each patient.

2.4.2: Modeling structure-propagation relationships across all patient data

Overall relationships between ictal ECoG measurements and DWI-derived structural metrics were assessed using linear mixed-effects models fitted with the MATLAB Statistics and Machine Learning Toolbox. ECoG data metrics (latency of ictal spasm oscillation onset, apparent speed of signal propagation, and amplitude of ictal spasm oscillations) were used as response variables, and DWI tractography metrics (length of centroid connecting electrode pairs and weighted FA of volumetric pathway connecting electrode pairs) were set as fixed effects. Fixed effect beta values are reported in the text below as their value ±95% confidence interval boundaries. Expected variation resulting from different ictal onset sites, sexes, and ages was accounted for by including a random effect term for each patient.

2.4.3: Across-patient comparisons

To assess individual patient differences with respect to generated structure-propagation models, random effect terms were extracted to generate patient-specific estimates of each relationship. Group descriptive statistics and beta weights extracted from linear mixed models in this way were compared across surgical outcome ILAE=1 and ILAE≥2 groups by ANOVA.

3 |. Results

3.1: Characterization of ECoG activity during epileptic spasm events

Across 18 patients, a total of 886 included spasm events, and an average of 84 ictal-active electrodes per patient, 70,188 oscillatory waveforms were analyzed. These signals lasted for an average duration of 519ms (S.D.=242ms) based on onset and offset calculations detailed above. Within spasm events, the average difference between earliest and latest onset across all active electrodes was 638ms (S.D.=237ms). Across spasm events, a given patient’s temporal pattern of electrode onset occurred in a relatively consistent manner: pooling patients’ individual-spasm onset latencies and spasm-averaged latency values revealed an overall correlation of r=.75 (p<.001), with individual patient median correlations ranging from r=.64 to r=.92 (Figure 3). This temporal consistency did not differ between patients with single vs. multiple clusters of spasm data (Table S1, F(1,17)=0.42, p=.42), speaking to the relative stability of spasm propagation in our sample.

Figure 3:

Similar consistency seen in ictal fast oscillation onset patterns regardless of postsurgical outcome. A: Examples of onset latencies (relative to each event’s mean detected onset time) are depicted for Patient 02 (left) and Patient 16 (right) in the two topmost panels. Each grid row represents onset times from a single subdural electrode, and each column represents a single spasm event, with earlier-detected onsets corresponding to darker squares. The broadly horizontal orientation of shading suggests a consistency in an electrode’s onset times across events. Below these panels, correlations between an electrode’s signal onset time in each spasm event and the mean onset time across all events are shown, with each plotted r value corresponding to the spasm event column directly above. B: Example ranges of each electrode’s onset time across spasm events. Electrode median latency values are shown as a dark blue dot, surrounded by a shaded rectangle that spans the electrode’s interquartile range. C: Stability of onset pattern across different patients. Patient median r values are shown as a vertical line, surrounded by a shaded rectangle that spans the patient’s interquartile range. Patients who achieved postsurgical seizure freedom (ILAE classification = 1) are depicted in green, while those who did not (ILAE classification ≥ 2) are depicted in red.

3.2: Individual patient spasm-associated activity propagates proportional to cortico-cortical tractography length and FA

Linear mixed models identified corticocortical pathway length as the most significant predictor of ictal oscillation onset latency (Figure 4B, β=2.20 [±0.49], p<.001, R²=0.54), where each 1mm increase in white matter pathway distance prolonged ictal oscillation onset latency by an average of 2.2ms. Pathway length also significantly predicted signal amplitude (Figure 4C, β=−5.85e-3 [±3.74e-3], p=.002, R²=0.31), with amplitude values tending to decrease as an electrode’s corticocortical distance from the onset area increased. To further explain variations within these significant relationships, weighted FA values were observed to be a significant predictor of both signal propagation speed (Figure 4D, β=2.54 [±1.66], p=.003, R²=0.41) and amplitude (Figure 4E, β=−1.87 [±1.69], p=.030, R²=0.25), where higher-FA pathways were associated with a faster apparent signal propagation speed, as well as a weakly decreased amplitude, when compared to lower-FA pathways. Corresponding structural models that connected electrodes via corticothalamic pathways did not show similar predictive relationships between tractography length and onset latency (β=0.25 [±1.44], p=.728) or amplitude (β=−8.31e-3 [±9.89e-3], p=.099).

Figure 4:

Corticocortical tractography from onset electrodes predict latencies and amplitudes of propagated ictal fast oscillations. A: Example subset of structural data from Patient 06, showing surface locations of the onset electrode (green), three propagated electrodes (red, pink, purple), and associated tractography. Example single-spasm tracings from these electrodes are shown below in corresponding colors. B: Centroid length of onset-originating corticocortical tractography predicts average latency of signal onset in propagated electrodes. Electrode pairs from each subject are plotted as dots corresponding to the key in the figure’s upper left, patient-specific random-effect model lines are plotted in the same color, and fixed effects of the model are plotted with 95% confidence intervals in blue. This color scheme was also applied to remaining figure sections. C: Length of onset-originating corticocortical tractography pathways predicts the average total amplitude of signal waveform. D: Weighted FA along pathway predicts corresponding modeled speed of spasm signal propagation. E: Weighted FA along pathway predicts the corresponding signal amplitude.

3.3: Comparing measures across surgical outcomes

Average metrics across patients who achieved postsurgical seizure freedom and those who did not are compared in Table 1. No significant differences in the average duration of ictal oscillation waveforms (F(1,17)<0.01, p=.99), the average window of onset propagation (F(1,17)=0.20, p=.66), or the median correlation measures of propagation consistency (F(1,17)=0.314, p=.58) were observed. Random effect terms from mixed models predicting onset latency from pathway length were significantly higher (i.e., showing faster corticocortical-modelled propagation speed) in patients with ILAE class-2 outcomes or worse, compared to patients who achieved class-1 outcome following surgery (F(1,17)=5.07, p=.039). Similar group differences were not seen in the prediction of onset-propagation speed based on pathway FA (F(1,17)=2.72, p=.118), or in the prediction of signal amplitude based on either pathway length (F(1,17)=0.02, p=.881) or pathway FA (F(1,17)=1.81, p=.197).

4 |. Discussion

4.1: Implications of the relationship between epileptic spasm propagation and direct corticocortical connections

Our results suggest that ictal oscillations observed during epileptic spasms reproducibly propagate along direct corticocortical white matter pathways. This observed consistency in propagation broadly parallels other research characterizing epilepsy as a network disorder,³⁹ including demonstrations of reproducibility⁴⁰ and corticocortical pathway dependance⁴¹ in spatiotemporal dynamics of interictal spikes and evoked potentials.⁴² However, the timescale of our observed signal propagation, on the order of hundreds of milliseconds, is not commensurate with expectations of rapid myelinated conduction, complicating neurophysiological interpretations of our results. We are limited to conclude that there is likely some mechanistic involvement of corticocortical pathways, which may reflect an accumulation of signals conducted along those fibers that prolongs signal propagation over longer distances, and may be better able to coordinate neural populations⁴³ to cause higher signal amplitude over shorter distances. As structure-propagation relationships in epileptic spasms are further defined, surgical teams may constrain their interpretation of ECoG recordings within this relationship to corroborate that apparent early-propagated activity aligns with individualized patient brain structure. Future prospective research will then be required to better understand the relationship between a brain region’s surgical targetability and its context in ictal signal propagation.

4.2. Possibility of additional and concurrent ictal propagation dynamics

The described relationship between corticocortical diffusion tractography metrics and ictal signal propagation does not contradict the possibility of other meaningful epileptic activity propagating along subcortical and other multisynaptic pathways linking our recorded cortical regions. Indeed, past identification of subcortical abnormalities in metabolic imaging suggests a pathological involvement of these areas in the generalization of cortically-generated spasm activity,^12,44 and it is likely that further markers of both spasm electrophysiology and disease progression may be identified subcortically. Nevertheless, our results suggest that such propagation may have less influence on the specific ECoG-recorded fast oscillations observed during epileptic spasm events. Additional data utilizing depth electrodes in subcortical regions may be able to directly identify electrophysiological markers of subcortical spasm propagation, at which point their usefulness toward guiding therapeutic efforts may be assessed.

Despite efforts to objectively approach the individualized signals present in the ictal ECoG tracing, our studied high-frequency activity and its associated propagation also represent only a subset of the information contained in our data. It is likely that other reliable characteristics of epileptic spasms can be extracted from ictal ECoG recordings – namely, signals related to cross-frequency coupling⁴⁵ that are harder to accurately assess across the time domain. Future research will determine the potential utility of this additional information, but the broad focus on high-frequency activity in this work still represents a highly individualized and objective analysis that sheds light on major components of ictal propagation dynamics.

4.3: Interpreting the relationship between structural-propagation ictal dynamics and surgical outcome

Our finding that structural-propagation modeling reveals a faster signal propagation in patients with unfavorable surgical outcome is consistent with other recent literature investigating seizure propagation in adults. The rapid ictal spread of fast beta oscillations out of the anterior temporal lobe toward posterior temporal and inferior frontal regions in TLE has been associated with surgical failure in these patients⁴⁶, and observations of phase-locked high gamma activity that quickly spreads outside of the brain lesions ultimately resected are associated with inferior surgical outcomes⁴⁷, warranting increased attention to resection margins in these types of cases. For our epileptic spasm patients, whose extensive surgical resections often already involve areas of early-propagated activity, we cannot clearly attribute surgical failure to unresected foci. Nevertheless, we hypothesize that rapid ictal oscillation propagation may be able to identify a pathological burden or condition more prone to surgical failure.

In this way, the type of analysis conducted in this study may help generate individualized patient markers that can guide expectations for surgery. Moreover, it is noteworthy that modeled corticocortical propagation speed distinguished surgical failure patients while the average spread window of signal onset did not, again stressing the importance of structurally individualized analysis. With further understanding of the involved pathophysiology, an anatomically-contextualized picture of spasm propagation may eventually help guide decisions related to surgery (e.g. smaller vs. larger resections).

4.4: Other limitations and future directions

The nature of epileptic spasms affords unique advantages and disadvantages toward understanding how seizure activity propagates through the brain. Patients may demonstrate insightful variability through repeated ictal events in a brief surgical monitoring period, but the relative scarcity of patients presenting for surgical monitoring limits our understanding of how ictal events vary across different spasm clusters and longer timescales. The broad range of patient age and etiology in epileptic spasms helps support the generalizability of findings across these populations, but also makes it difficult to account for the specific influence of confounding variables like age, sex, and development, which can strongly influence the diffusion imaging measures discussed in this study.

Ultimately, we hope that future research will be able to leverage the principles observed with our invasive ECoG recordings and identify signals from non-invasive modalities that similarly reflect this anatomically-oriented understanding of spasm signal propagation. Not only would this allow development of novel, non-invasive markers for surgical candidacy in patients without focal lesions, but it could allow study of larger datasets that better reflect typical patients with epileptic spasms. More advanced methods of assessing white matter development⁴⁸ may further characterize brain structures towards a developed mechanistic understanding, and the ability to study both hemispheres in non-surgical patients could ultimately contribute to a better understanding of structural spasm propagation that can be used as an individual marker.

5 |. Conclusions

We present a novel multimodal analysis of unique surgical data in patients with epileptic spasms, revealing a relationship between ictal fast oscillation propagation and measures of direct corticocortical structural pathways between cortical sites that significantly differed between patients with successful and unsuccessful surgical outcomes.

Key Points Box:

• Spatiotemporal propagation dynamics of high-frequency activity in epileptic spasms were studied in a patient-individualized structural framework.

• Length and FA measures along direct corticocortical pathways were associated with latency and amplitude of propagated ictal signals.

• Patients who did not achieve seizure freedom following resection showed faster ictal signal propagation along corticocortical pathways.