PyHFO: lightweight deep learning-powered end-to-end high-frequency oscillations analysis application.

OBJECTIVE: This study aims to develop and validate an end-to-end software platform, PyHFO, that streamlines the application of deep learning methodologies in detecting neurophysiological biomarkers for epileptogenic zones from EEG recordings. Methods: We introduced PyHFO, which enables time-efficient HFO detection algorithms like short-term energy (STE) and Montreal Neurological Institute and Hospital (MNI) detectors. It incorporates deep learning models for artifact and HFO with spike classification, designed to operate efficiently on standard computer hardware. Main results: The validation of PyHFO was conducted on three separate datasets: the first comprised solely of grid/strip electrodes, the second a combination of grid/strip and depth electrodes, and the third derived from rodent studies, which sampled the neocortex and hippocampus using depth electrodes. PyHFO demonstrated an ability to handle datasets efficiently, with optimization techniques enabling it to achieve speeds up to 50 times faster than traditional HFO detection applications. Users have the flexibility to employ our pre-trained deep learning model or use their EEG data for custom model training. Significance: PyHFO successfully bridges the computational challenge faced in applying deep learning techniques to EEG data analysis in epilepsy studies, presenting a feasible solution for both clinical and research settings. By offering a user-friendly and computationally efficient platform, PyHFO paves the way for broader adoption of advanced EEG data analysis tools in clinical practice and fosters potential for large-scale research collaborations.

ComBating inter-site differences in field strength: harmonizing preclinical traumatic brain injury MRI data.

Integrating datasets from multiple sites and scanners can increase statistical power for neuroimaging studies but can also introduce significant inter-site confounds. We evaluated the effectiveness of ComBat, an empirical Bayes approach, to combine longitudinal preclinical MRI data acquired at 4.7 or 9.4 T at two different sites in Australia. Male Sprague Dawley rats underwent MRI on Days 2, 9, 28, and 150 following moderate/severe traumatic brain injury (TBI) or sham injury as part of Project 1 of the NIH/NINDS-funded Centre Without Walls EpiBioS4Rx project. Diffusion-weighted and multiple-gradient-echo images were acquired, and outcomes included QSM, FA, and ADC. Acute injury measures including apnea and self-righting reflex were consistent between sites. Mixed-effect analysis of ipsilateral and contralateral corpus callosum (CC) summary values revealed a significant effect of site on FA and ADC values, which was removed following ComBat harmonization. Bland-Altman plots for each metric showed reduced variability across sites following ComBat harmonization, including for QSM, despite appearing to be largely unaffected by inter-site differences and no effect of site observed. Following harmonization, the combined inter-site data revealed significant differences in the imaging metrics consistent with previously reported outcomes. TBI resulted in significantly reduced FA and increased susceptibility in the ipsilateral CC, and significantly reduced FA in the contralateral CC compared with sham-injured rats. Additionally, TBI rats also exhibited a reversal in ipsilateral CC ADC values over time with significantly reduced ADC at Day 9, followed by increased ADC 150 days after injury. Our findings demonstrate the need for harmonizing multi-site preclinical MRI data and show that this can be successfully achieved using ComBat while preserving phenotypical changes due to TBI.

Patterns of subregional cerebellar atrophy across epilepsy syndromes: An ENIGMA-Epilepsy study.

OBJECTIVE: The intricate neuroanatomical structure of the cerebellum is of longstanding interest in epilepsy, but has been poorly characterized within the current corticocentric models of this disease. We quantified cross-sectional regional cerebellar lobule volumes using structural magnetic resonance imaging in 1602 adults with epilepsy and 1022 healthy controls across 22 sites from the global ENIGMA-Epilepsy working group. METHODS: A state-of-the-art deep learning-based approach was employed that parcellates the cerebellum into 28 neuroanatomical subregions. Linear mixed models compared total and regional cerebellar volume in (1) all epilepsies, (2) temporal lobe epilepsy with hippocampal sclerosis (TLE-HS), (3) nonlesional temporal lobe epilepsy, (4) genetic generalized epilepsy, and (5) extratemporal focal epilepsy (ETLE). Relationships were examined for cerebellar volume versus age at seizure onset, duration of epilepsy, phenytoin treatment, and cerebral cortical thickness. RESULTS: Across all epilepsies, reduced total cerebellar volume was observed (d = .42). Maximum volume loss was observed in the corpus medullare (dmax = .49) and posterior lobe gray matter regions, including bilateral lobules VIIB (dmax = .47), crus I/II (dmax = .39), VIIIA (dmax = .45), and VIIIB (dmax = .40). Earlier age at seizure onset ( η ρ max 2 = .05) and longer epilepsy duration ( η ρ max 2 = .06) correlated with reduced volume in these regions. Findings were most pronounced in TLE-HS and ETLE, with distinct neuroanatomical profiles observed in the posterior lobe. Phenytoin treatment was associated with reduced posterior lobe volume. Cerebellum volume correlated with cerebral cortical thinning more strongly in the epilepsy cohort than in controls. SIGNIFICANCE: We provide robust evidence of deep cerebellar and posterior lobe subregional gray matter volume loss in patients with chronic epilepsy. Volume loss was maximal for posterior subregions implicated in nonmotor functions, relative to motor regions of both the anterior and posterior lobe. Associations between cerebral and cerebellar changes, and variability of neuroanatomical profiles across epilepsy syndromes argue for more precise incorporation of cerebellar subregional damage into neurobiological models of epilepsy.

Pathological neurons generate ripples at the UP-DOWN transition disrupting information transfer.

OBJECTIVE: To confirm and investigate why pathological high-frequency oscillations (pHFOs), including ripples (80-200 Hz) and fast ripples (200-600 Hz), are generated during the UP-DOWN transition of the slow wave and if information transmission mediated by ripple temporal coupling is disrupted in the seizure-onset zone (SOZ). METHODS: We isolated 217 total units from 175.95 intracranial electroencephalography (iEEG) contact-hours of synchronized macro- and microelectrode recordings from 6 patients. Sleep slow oscillation (.1-2 Hz) epochs were identified in the iEEG recording. iEEG HFOs that occurred superimposed on the slow wave were transformed to phasors and adjusted by the phase of maximum firing in nearby units (i.e., maximum UP). We tested whether, in the SOZ, HFOs and associated action potentials (APs) occur more often at the UP-DOWN transition. We also examined ripple temporal correlations using cross-correlograms. RESULTS: At the group level in the SOZ, HFO and HFO-associated AP probability was highest during the UP-DOWN transition of slow wave excitability (p < < .001). In the non-SOZ, HFO and HFO-associated AP was highest during the DOWN-UP transition (p < < .001). At the unit level in the SOZ, 15.6% and 20% of units exhibited more robust firing during ripples (Cohen's d = .11-.83) and fast ripples (d = .36-.90) at the UP-DOWN transition (p < .05 f.d.r. corrected), respectively. By comparison, also in the SOZ, 6.6% (d = .14-.30) and 8.5% (d = .33-.41) of units had significantly less firing during ripples and fast ripples at the UP-DOWN transition, respectively. Additional data shows that ripple and fast ripple temporal correlations, involving global slow waves, between the hippocampus, entorhinal cortex, and parahippocampal gyrus were reduced by >50% in the SOZ compared to the non-SOZ (N = 3). SIGNIFICANCE: The UP-DOWN transition of slow wave excitability facilitates the activation of pathological neurons to generate pHFOs. Ripple temporal correlations across brain regions may be important in memory consolidation and are disrupted in the SOZ, perhaps by pHFO generation.

Successful harmonization in EpiBioS4Rx biomarker study on post-traumatic epilepsy paves the way towards powered preclinical multicenter studies.

OBJECTIVE: Project 1 of the Preclinical Multicenter Epilepsy Bioinformatics Study for Antiepileptogenic Therapy (EpiBioS4Rx) consortium aims to identify preclinical biomarkers for antiepileptogenic therapies following traumatic brain injury (TBI). The international participating centers in Finland, Australia, and the United States have made a concerted effort to ensure protocol harmonization. Here, we evaluate the success of harmonization process by assessing the timing, coverage, and performance between the study sites. METHOD: We collected data on animal housing conditions, lateral fluid-percussion injury model production, postoperative care, mortality, post-TBI physiological monitoring, timing of blood sampling and quality, MR imaging timing and protocols, and duration of video-electroencephalography (EEG) follow-up using common data elements. Learning effect in harmonization was assessed by comparing procedural accuracy between the early and late stages of the project. RESULTS: The animal housing conditions were comparable between the study sites but the postoperative care procedures varied. Impact pressure, duration of apnea, righting reflex, and acute mortality differed between the study sites (p < 0.001). The severity of TBI on D2 post TBI assessed using the composite neuroscore test was similar between the sites, but recovery of acute somato-motor deficits varied (p < 0.001). A total of 99% of rats included in the final cohort in UEF, 100% in Monash, and 79% in UCLA had blood samples taken at all time points. The timing of sampling differed on day (D)2 (p < 0.05) but not D9 (p > 0.05). Plasma quality was poor in 4% of the samples in UEF, 1% in Monash and 14% in UCLA. More than 97% of the final cohort were MR imaged at all timepoints in all study sites. The timing of imaging did not differ on D2 and D9 (p > 0.05), but varied at D30, 5 months, and ex vivo timepoints (p < 0.001). The percentage of rats that completed the monthly high-density video-EEG follow-up and the duration of video-EEG recording on the 7th post-injury month used for seizure detection for diagnosis of post-traumatic epilepsy differed between the sites (p < 0.001), yet the prevalence of PTE (UEF 21%, Monash 22%, UCLA 23%) was comparable between the sites (p > 0.05). A decrease in acute mortality and increase in plasma quality across time reflected a learning effect in the TBI production and blood sampling protocols. SIGNIFICANCE: Our study is the first demonstration of the feasibility of protocol harmonization for performing powered preclinical multi-center trials for biomarker and therapy discovery of post-traumatic epilepsy.

Patterns of subregional cerebellar atrophy across epilepsy syndromes: An ENIGMA-Epilepsy study.

Objective: The intricate neuroanatomical structure of the cerebellum is of longstanding interest in epilepsy, but has been poorly characterized within the current cortico-centric models of this disease. We quantified cross-sectional regional cerebellar lobule volumes using structural MRI in 1,602 adults with epilepsy and 1,022 healthy controls across twenty-two sites from the global ENIGMA-Epilepsy working group. Methods: A state-of-the-art deep learning-based approach was employed that parcellates the cerebellum into 28 neuroanatomical subregions. Linear mixed models compared total and regional cerebellar volume in i) all epilepsies; ii) temporal lobe epilepsy with hippocampal sclerosis (TLE-HS); iii) non-lesional temporal lobe epilepsy (TLE-NL); iv) genetic generalised epilepsy; and (v) extra-temporal focal epilepsy (ETLE). Relationships were examined for cerebellar volume versus age at seizure onset, duration of epilepsy, phenytoin treatment, and cerebral cortical thickness. Results: Across all epilepsies, reduced total cerebellar volume was observed (d=0.42). Maximum volume loss was observed in the corpus medullare (dmax=0.49) and posterior lobe grey matter regions, including bilateral lobules VIIB (dmax= 0.47), Crus I/II (dmax= 0.39), VIIIA (dmax=0.45) and VIIIB (dmax=0.40). Earlier age at seizure onset (ηρ2max=0.05) and longer epilepsy duration (ηρ2max=0.06) correlated with reduced volume in these regions. Findings were most pronounced in TLE-HS and ETLE with distinct neuroanatomical profiles observed in the posterior lobe. Phenytoin treatment was associated with reduced posterior lobe volume. Cerebellum volume correlated with cerebral cortical thinning more strongly in the epilepsy cohort than in controls. Significance: We provide robust evidence of deep cerebellar and posterior lobe subregional grey matter volume loss in patients with chronic epilepsy. Volume loss was maximal for posterior subregions implicated in non-motor functions, relative to motor regions of both the anterior and posterior lobe. Associations between cerebral and cerebellar changes, and variability of neuroanatomical profiles across epilepsy syndromes argue for more precise incorporation of cerebellum subregions into neurobiological models of epilepsy.

High-rate leading spikes in propagating spike sequences predict seizure outcome in surgical patients with temporal lobe epilepsy.

Inter-ictal spikes aid in the diagnosis of epilepsy and in planning surgery of medication-resistant epilepsy. However, the localizing information from spikes can be unreliable because spikes can propagate, and the burden of spikes, often assessed as a rate, does not always correlate with the seizure onset zone or seizure outcome. Recent work indicates identifying where spikes regularly emerge and spread could localize the seizure network. Thus, the current study sought to better understand where and how rates of single and coupled spikes, and especially brain regions with high-rate and leading spike of a propagating sequence, informs the extent of the seizure network. In 37 patients with medication-resistant temporal lobe seizures, who had surgery to treat their seizure disorder, an algorithm detected spikes in the pre-surgical depth inter-ictal EEG. A separate algorithm detected spike propagation sequences and identified the location of leading and downstream spikes in each sequence. We analysed the rate and power of single spikes on each electrode and coupled spikes between pairs of electrodes, and the proportion of sites with high-rate, leading spikes in relation to the seizure onset zone of patients seizure free (n = 19) and those with continuing seizures (n = 18). We found increased rates of single spikes in mesial temporal seizure onset zone (ANOVA, P < 0.001, η2 = 0.138), and increased rates of coupled spikes within, but not between, mesial-, lateral- and extra-temporal seizure onset zone of patients with continuing seizures (P < 0.001; η2 = 0.195, 0.113 and 0.102, respectively). In these same patients, there was a higher proportion of brain regions with high-rate leaders, and each sequence contained a greater number of spikes that propagated with a higher efficiency over a longer distance outside the seizure onset zone than patients seizure free (Wilcoxon, P = 0.0172). The proportion of high-rate leaders in and outside the seizure onset zone could predict seizure outcome with area under curve = 0.699, but not rates of single or coupled spikes (0.514 and 0.566). Rates of coupled spikes to a greater extent than single spikes localize the seizure onset zone and provide evidence for inter-ictal functional segregation, which could be an adaptation to avert seizures. Spike rates, however, have little value in predicting seizure outcome. High-rate spike sites leading propagation could represent sources of spikes that are important components of an efficient seizure network beyond the clinical seizure onset zone, and like the seizure onset zone these, too, need to be removed, disconnected or stimulated to increase the likelihood for seizure control.

Fast ripples reflect increased excitability that primes epileptiform spikes.

The neuronal circuit disturbances that drive inter-ictal and ictal epileptiform discharges remain elusive. Using a combination of extra-operative macro-electrode and micro-electrode inter-ictal recordings in six pre-surgical patients during non-rapid eye movement sleep, we found that, exclusively in the seizure onset zone, fast ripples (200-600 Hz), but not ripples (80-200 Hz), frequently occur <300 ms before an inter-ictal intra-cranial EEG spike with a probability exceeding chance (bootstrapping, P < 1e-5). Such fast ripple events are associated with higher spectral power (P < 1e-10) and correlated with more vigorous neuronal firing than solitary fast ripple (generalized linear mixed-effects model, P < 1e-9). During the intra-cranial EEG spike that follows a fast ripple, action potential firing is lower than during an intra-cranial EEG spike alone (generalized linear mixed-effects model, P < 0.05), reflecting an inhibitory restraint of intra-cranial EEG spike initiation. In contrast, ripples do not appear to prime epileptiform spikes. We next investigated the clinical significance of pre-spike fast ripple in a separate cohort of 23 patients implanted with stereo EEG electrodes, who underwent resections. In non-rapid eye movement sleep recordings, sites containing a high proportion of fast ripple preceding intra-cranial EEG spikes correlate with brain areas where seizures begin more than solitary fast ripple (P < 1e-5). Despite this correlation, removal of these sites does not guarantee seizure freedom. These results are consistent with the hypothesis that fast ripple preceding EEG spikes reflect an increase in local excitability that primes EEG spike discharges preferentially in the seizure onset zone and that epileptogenic brain regions are necessary, but not sufficient, for initiating inter-ictal epileptiform discharges.

Pathological neurons generate ripples at the UP-DOWN transition disrupting information transfer.

Objective: To confirm and investigate why pathological HFOs (pHFOs), including Ripples [80-200 Hz] and fast ripples [200-600 Hz], are generated during the UP-DOWN transition of the slow wave and if pHFOs interfere with information transmission. Methods: We isolated 217 total units from 175.95 iEEG contact-hours of synchronized macro- and microelectrode recordings from 6 patients. Sleep slow oscillation (0.1-2 Hz) epochs were identified in the iEEG recording. iEEG HFOs that occurred superimposed on the slow wave were transformed to phasors and adjusted by the phase of maximum firing in nearby units (i.e., maximum UP). We tested whether, in the seizure onset zone (SOZ), HFOs and associated action potentials (AP) occur more often at the UP-DOWN transition. We also examined ripple temporal correlations using cross correlograms. Results: At the group level in the SOZ, HFO and HFO-associated AP probability was highest during the UP-DOWN transition of slow wave excitability (p<<0.001). In the non-SOZ, HFO and HFO-associated AP was highest during the DOWN-UP transition (p<<0.001). At the unit level in the SOZ, 15.6% and 20% of units exhibited more robust firing during ripples (Cohen's d=0.11-0.83) and fast ripples (d=0.36-0.90) at the UP-DOWN transition (p<0.05 f.d.r corrected), respectively. By comparison, also in the SOZ, 6.6% (d=0.14-0.30) and 8.5% (d=0.33-0.41) of units had significantly less firing during ripples and fast ripples at the UP-DOWN transition, respectively. Additional data shows ripple temporal correlations, involving global slow waves, between the hippocampus, entorhinal cortex, and parahippocampal gyrus were reduced by ~50-80% in the SOZ compared to the non-SOZ (N=3). Significance: The UP-DOWN transition of slow wave excitability facilitates the activation of pathological neurons to generate pHFOs. The pathological neurons and pHFOs disrupt ripple temporal correlations across brain regions that transfer information and may be important in memory consolidation.

Optimizing detection and deep learning-based classification of pathological high-frequency oscillations in epilepsy.

OBJECTIVE: This study aimed to explore sensitive detection methods for pathological high-frequency oscillations (HFOs) to improve seizure outcomes in epilepsy surgery. METHODS: We analyzed interictal HFOs (80-500 Hz) in 15 children with medication-resistant focal epilepsy who underwent chronic intracranial electroencephalogram via subdural grids. The HFOs were assessed using the short-term energy (STE) and Montreal Neurological Institute (MNI) detectors and examined for spike association and time-frequency plot characteristics. A deep learning (DL)-based classification was applied to purify pathological HFOs. Postoperative seizure outcomes were correlated with HFO-resection ratios to determine the optimal HFO detection method. RESULTS: The MNI detector identified a higher percentage of pathological HFOs than the STE detector, but some pathological HFOs were detected only by the STE detector. HFOs detected by both detectors had the highest spike association rate. The Union detector, which detects HFOs identified by either the MNI or STE detector, outperformed other detectors in predicting postoperative seizure outcomes using HFO-resection ratios before and after DL-based purification. CONCLUSIONS: HFOs detected by standard automated detectors displayed different signal and morphological characteristics. DL-based classification effectively purified pathological HFOs. SIGNIFICANCE: Enhancing the detection and classification methods of HFOs will improve their utility in predicting postoperative seizure outcomes.

Optimizing Detection and Deep Learning-based Classification of Pathological High-Frequency Oscillations in Epilepsy.

Objective: This study aimed to explore sensitive detection methods and deep learning (DL)-based classification for pathological high-frequency oscillations (HFOs). Methods: We analyzed interictal HFOs (80-500 Hz) in 15 children with medication-resistant focal epilepsy who underwent resection after chronic intracranial electroencephalogram via subdural grids. The HFOs were assessed using the short-term energy (STE) and Montreal Neurological Institute (MNI) detectors and examined for pathological features based on spike association and time-frequency plot characteristics. A DL-based classification was applied to purify pathological HFOs. Postoperative seizure outcomes were correlated with HFO-resection ratios to determine the optimal HFO detection method. Results: The MNI detector identified a higher percentage of pathological HFOs than the STE detector, but some pathological HFOs were detected only by the STE detector. HFOs detected by both detectors exhibited the most pathological features. The Union detector, which detects HFOs identified by either the MNI or STE detector, outperformed other detectors in predicting postoperative seizure outcomes using HFO-resection ratios before and after DL-based purification. Conclusions: HFOs detected by standard automated detectors displayed different signal and morphological characteristics. DL-based classification effectively purified pathological HFOs. Significance: Enhancing the detection and classification methods of HFOs will improve their utility in predicting postoperative seizure outcomes. HIGHLIGHTS: HFOs detected by the MNI detector showed different traits and higher pathological bias than those detected by the STE detectorHFOs detected by both MNI and STE detectors (the Intersection HFOs) were deemed the most pathologicalA deep learning-based classification was able to distill pathological HFOs, regard-less of the initial HFO detection methods.

Fast ripples reflect increased excitability that primes epileptiform spikes.

The neuronal circuit disturbances that drive interictal and ictal epileptiform discharges remains elusive. Using a combination of extraoperative macro- and micro-electrode interictal recordings in six presurgical patients during non-rapid eye movement (REM) sleep we found that, exclusively in the seizure onset zone, fast ripples (FR; 200-600Hz), but not ripples (80-200 Hz), frequently occur <300 msec before an interictal intracranial EEG (iEEG) spike with a probability exceeding chance (bootstrapping, p<1e-5). Such FR events are associated with higher spectral power (p<1e-10) and correlated with more vigorous neuronal firing than solitary FR (generalized linear mixed-effects model, GLMM, p<1e-3) irrespective of FR power. During the iEEG spike that follows a FR, action potential firing is lower than during a iEEG spike alone (GLMM, p<1e-10), reflecting an inhibitory restraint of iEEG spike initiation. In contrast, ripples do not appear to prime epileptiform spikes. We next investigated the clinical significance of pre-spike FR in a separate cohort of 23 patients implanted with stereo EEG electrodes who underwent resections. In non-REM sleep recordings, sites containing a high proportion of FR preceding iEEG spikes correlate with brain areas where seizures begin more than solitary FR (p<1e-5). Despite this correlation, removal of these sites does not guarantee seizure freedom. These results are consistent with the hypothesis that FR preceding EEG spikes reflect an increase in local excitability that primes EEG spike discharges preferentially in the seizure onset zone and that epileptogenic brain regions are necessary, but not sufficient, for initiating interictal epileptiform discharges.

Graph theoretical measures of fast ripple networks improve the accuracy of post-operative seizure outcome prediction.

Fast ripples (FR) are a biomarker of epileptogenic brain, but when larger portions of FR generating regions are resected seizure freedom is not always achieved. To evaluate and improve the diagnostic accuracy of FR resection for predicting seizure freedom we compared the FR resection ratio (RR) with FR network graph theoretical measures. In 23 patients FR were semi-automatically detected and quantified in stereo EEG recordings during sleep. MRI normalization and co-registration localized contacts and relation to resection margins. The number of FR, and graph theoretical measures, which were spatial (i.e., FR rate-distance radius) or temporal correlational (i.e., FR mutual information), were compared with the resection margins and with seizure outcome We found that the FR RR did not correlate with seizure-outcome (p > 0.05). In contrast, the FR rate-distance radius resected difference and the FR MI mean characteristic path length RR did correlate with seizure-outcome (p < 0.05). Retesting of positive FR RR patients using either FR rate-distance radius resected difference or the FR MI mean characteristic path length RR reduced seizure-free misclassifications from 44 to 22% and 17%, respectively. These results indicate that graph theoretical measures of FR networks can improve the diagnostic accuracy of the resection of FR events for predicting seizure freedom.

Characterizing physiological high-frequency oscillations using deep learning.

Objective.Intracranially-recorded interictal high-frequency oscillations (HFOs) have been proposed as a promising spatial biomarker of the epileptogenic zone. However, HFOs can also be recorded in the healthy brain regions, which complicates the interpretation of HFOs. The present study aimed to characterize salient features of physiological HFOs using deep learning (DL).Approach.We studied children with neocortical epilepsy who underwent intracranial strip/grid evaluation. Time-series EEG data were transformed into DL training inputs. The eloquent cortex (EC) was defined by functional cortical mapping and used as a DL label. Morphological characteristics of HFOs obtained from EC (ecHFOs) were distilled and interpreted through a novel weakly supervised DL model.Main results.A total of 63 379 interictal intracranially-recorded HFOs from 18 children were analyzed. The ecHFOs had lower amplitude throughout the 80-500 Hz frequency band around the HFO onset and also had a lower signal amplitude in the low frequency band throughout a one-second time window than non-ecHFOs, resembling a bell-shaped template in the time-frequency map. A minority of ecHFOs were HFOs with spikes (22.9%). Such morphological characteristics were confirmed to influence DL model prediction via perturbation analyses. Using the resection ratio (removed HFOs/detected HFOs) of non-ecHFOs, the prediction of postoperative seizure outcomes improved compared to using uncorrected HFOs (area under the ROC curve of 0.82, increased from 0.76).Significance.We characterized salient features of physiological HFOs using a DL algorithm. Our results suggested that this DL-based HFO classification, once trained, might help separate physiological from pathological HFOs, and efficiently guide surgical resection using HFOs.

Delta oscillation coupled propagating fast ripples precede epileptiform discharges in patients with focal epilepsy.

Epileptiform spikes are used to localize epileptogenic brain tissue. The mechanisms that spontaneously trigger epileptiform discharges are not yet elucidated. Pathological fast ripple (FR, 200-600 Hz) are biomarkers of epileptogenic brain, and we postulated that FR network interactions are involved in generating epileptiform spikes. Using macroelectrode stereo intracranial EEG (iEEG) recordings from a cohort of 46 patients we found that, in the seizure onset zone (SOZ), propagating FR were more often followed by an epileptiform spike, as compared with non-propagating FR (p < 0.05). Propagating FR had a distinct frequency and larger power (p < 1e-10) and were more strongly phase coupled to the peak of iEEG delta oscillation, which likely correspond with the DOWN states during non-REM sleep (p < 1e-8), than non-propagating FR. While FR propagation was rare, all FR occurred with the highest probability within +/- 400 msec of epileptiform spikes with superimposed high-frequency oscillations (p < 0.05). Thus, a sub-population of epileptiform spikes in the SOZ, are preceded by propagating FR that are coordinated by the DOWN state during non-REM sleep.

Graph theoretical measures of fast ripples support the epileptic network hypothesis.

The epileptic network hypothesis and epileptogenic zone hypothesis are two theories of ictogenesis. The network hypothesis posits that coordinated activity among interconnected nodes produces seizures. The epileptogenic zone hypothesis posits that distinct regions are necessary and sufficient for seizure generation. High-frequency oscillations, and particularly fast ripples, are thought to be biomarkers of the epileptogenic zone. We sought to test these theories by comparing high-frequency oscillation rates and networks in surgical responders and non-responders, with no appreciable change in seizure frequency or severity, within a retrospective cohort of 48 patients implanted with stereo-EEG electrodes. We recorded inter-ictal activity during non-rapid eye movement sleep and semi-automatically detected and quantified high-frequency oscillations. Each electrode contact was localized in normalized coordinates. We found that the accuracy of seizure onset zone electrode contact classification using high-frequency oscillation rates was not significantly different in surgical responders and non-responders, suggesting that in non-responders the epileptogenic zone partially encompassed the seizure onset zone(s) (P > 0.05). We also found that in the responders, fast ripple on oscillations exhibited a higher spectral content in the seizure onset zone compared with the non-seizure onset zone (P < 1 × 10-5). By contrast, in the non-responders, fast ripple had a lower spectral content in the seizure onset zone (P < 1 × 10-5). We constructed two different networks of fast ripple with a spectral content >350 Hz. The first was a rate-distance network that multiplied the Euclidian distance between fast ripple-generating contacts by the average rate of fast ripple in the two contacts. The radius of the rate-distance network, which excluded seizure onset zone nodes, discriminated non-responders, including patients not offered resection or responsive neurostimulation due to diffuse multifocal onsets, with an accuracy of 0.77 [95% confidence interval (CI) 0.56-0.98]. The second fast ripple network was constructed using the mutual information between the timing of the events to measure functional connectivity. For most non-responders, this network had a longer characteristic path length, lower mean local efficiency in the non-seizure onset zone, and a higher nodal strength among non-seizure onset zone nodes relative to seizure onset zone nodes. The graphical theoretical measures from the rate-distance and mutual information networks of 22 non- responsive neurostimulation treated patients was used to train a support vector machine, which when tested on 13 distinct patients classified non-responders with an accuracy of 0.92 (95% CI 0.75-1). These results indicate patients who do not respond to surgery or those not selected for resection or responsive neurostimulation can be explained by the epileptic network hypothesis that is a decentralized network consisting of widely distributed, hyperexcitable fast ripple-generating nodes.

Susceptibility to epilepsy after traumatic brain injury is associated with preexistent gut microbiome profile.

OBJECTIVE: We examined whether posttraumatic epilepsy (PTE) is associated with measurable perturbations in gut microbiome. METHODS: Adult Sprague Dawley rats were subjected to lateral fluid percussion injury (LFPI). PTE was examined 7 months after LFPI, during 4-week continuous video-electroencephalographic monitoring. 16S ribosomal RNA gene sequencing was performed in fecal samples collected before LFPI/sham-LFPI and 1 week, 1 month, and 7 months thereafter. Longitudinal analyses of alpha diversity, beta diversity, and differential microbial abundance were performed. Short-chain fatty acids (SCFAs) were measured in fecal samples collected before LFPI by liquid chromatography with tandem mass spectrometry. RESULTS: Alpha diversity changed over time in both LFPI and sham-LFPI subjects; no association was observed between alpha diversity and LFPI, the severity of post-LFPI neuromotor impairments, and PTE. LFPI produced significant changes in beta diversity and selective changes in microbial abundances associated with the severity of neuromotor impairments. No association between LFPI-dependent microbial perturbations and PTE was detected. PTE was associated with beta diversity irrespective of timepoint vis-à-vis LFPI, including at baseline. Preexistent fecal microbial abundances of four amplicon sequence variants belonging to the Lachnospiraceae family (three enriched and one depleted) predicted the risk of PTE, with area under the curve (AUC) of .73. Global SCFA content was associated with the increased risk of PTE, with AUC of .722, and with 2-methylbutyric (depleted), valeric (depleted), isobutyric (enriched), and isovaleric (enriched) acids being the most important factors (AUC = .717). When the analyses of baseline microbial and SCFA compositions were combined, AUC to predict PTE increased to .78. SIGNIFICANCE: Whereas LFPI produces no perturbations in the gut microbiome that are associated with PTE, the risk of PTE can be stratified based on preexistent microbial abundances and SCFA content.

Refining epileptogenic high-frequency oscillations using deep learning: a reverse engineering approach.

Intracranially recorded interictal high-frequency oscillations have been proposed as a promising spatial biomarker of the epileptogenic zone. However, its visual verification is time-consuming and exhibits poor inter-rater reliability. Furthermore, no method is currently available to distinguish high-frequency oscillations generated from the epileptogenic zone (epileptogenic high-frequency oscillations) from those generated from other areas (non-epileptogenic high-frequency oscillations). To address these issues, we constructed a deep learning-based algorithm using chronic intracranial EEG data via subdural grids from 19 children with medication-resistant neocortical epilepsy to: (i) replicate human expert annotation of artefacts and high-frequency oscillations with or without spikes, and (ii) discover epileptogenic high-frequency oscillations by designing a novel weakly supervised model. The 'purification power' of deep learning is then used to automatically relabel the high-frequency oscillations to distill epileptogenic high-frequency oscillations. Using 12 958 annotated high-frequency oscillation events from 19 patients, the model achieved 96.3% accuracy on artefact detection (F1 score = 96.8%) and 86.5% accuracy on classifying high-frequency oscillations with or without spikes (F1 score = 80.8%) using patient-wise cross-validation. Based on the algorithm trained from 84 602 high-frequency oscillation events from nine patients who achieved seizure-freedom after resection, the majority of such discovered epileptogenic high-frequency oscillations were found to be ones with spikes (78.6%, P < 0.001). While the resection ratio of detected high-frequency oscillations (number of resected events/number of detected events) did not correlate significantly with post-operative seizure freedom (the area under the curve = 0.76, P = 0.06), the resection ratio of epileptogenic high-frequency oscillations positively correlated with post-operative seizure freedom (the area under the curve = 0.87, P = 0.01). We discovered that epileptogenic high-frequency oscillations had a higher signal intensity associated with ripple (80-250 Hz) and fast ripple (250-500 Hz) bands at the high-frequency oscillation onset and with a lower frequency band throughout the event time window (the inverted T-shaped), compared to non-epileptogenic high-frequency oscillations. We then designed perturbations on the input of the trained model for non-epileptogenic high-frequency oscillations to determine the model's decision-making logic. The model confidence significantly increased towards epileptogenic high-frequency oscillations by the artificial introduction of the inverted T-shaped signal template (mean probability increase: 0.285, P < 0.001), and by the artificial insertion of spike-like signals into the time domain (mean probability increase: 0.452, P < 0.001). With this deep learning-based framework, we reliably replicated high-frequency oscillation classification tasks by human experts. Using a reverse engineering technique, we distinguished epileptogenic high-frequency oscillations from others and identified its salient features that aligned with current knowledge.

Automated preprocessing and phase-amplitude coupling analysis of scalp EEG discriminates infantile spasms from controls during wakefulness.

OBJECTIVE: Delta-gamma phase-amplitude coupling in EEG is useful for localizing epileptic sources and to evaluate severity in children with infantile spasms. We (1) develop an automated EEG preprocessing pipeline to clean data using artifact subspace reconstruction (ASR) and independent component (IC) analysis (ICA) and (2) evaluate delta-gamma modulation index (MI) as a method to distinguish children with epileptic spasms (cases) from normal controls during sleep and awake. METHODS: Using 400 scalp EEG datasets (200 sleep, 200 awake) from 100 subjects, we calculated MI after applying high-pass and line-noise filters (Clean 0), and after ASR followed by either conservative (Clean 1) or stringent (Clean 2) artifactual IC rejection. Classification of cases and controls using MI was evaluated with Receiver Operating Characteristics (ROC) to obtain area under curve (AUC). RESULTS: The artifact rejection algorithm reduced raw signal variance by 29-45% and 38-60% for Clean 1 and Clean 2, respectively. MI derived from sleep data, with or without preprocessing, robustly classified the groups (all AUC > 0.98). In contrast, group classification using MI derived from awake data was successful only after Clean 2 (AUC = 0.85). CONCLUSIONS: We have developed an automated EEG preprocessing pipeline to perform artifact rejection and quantify delta-gamma modulation index.

Accuracy of high-frequency oscillations recorded intraoperatively for classification of epileptogenic regions.

To see whether acute intraoperative recordings using stereo EEG (SEEG) electrodes can replace prolonged interictal intracranial EEG (iEEG) recording, making the process more efficient and safer, 10 min of iEEG were recorded following electrode implantation in 16 anesthetized patients, and 1-2 days later during non-rapid eye movement (REM) sleep. Ripples on oscillations (RonO, 80-250 Hz), ripples on spikes (RonS), sharp-spikes, fast RonO (fRonO, 250-600 Hz), and fast RonS (fRonS) were semi-automatically detected. HFO power and frequency were compared between the conditions using a generalized linear mixed-effects model. HFO rates were compared using a two-way repeated measures ANOVA with anesthesia type and SOZ as factors. A receiver-operating characteristic (ROC) curve analysis quantified seizure onset zone (SOZ) classification accuracy, and the scalar product was used to assess spatial reliability. Resection of contacts with the highest rate of events was compared with outcome. During sleep, all HFOs, except fRonO, were larger in amplitude compared to intraoperatively (p < 0.01). HFO frequency was also affected (p < 0.01). Anesthesia selection affected HFO and sharp-spike rates. In both conditions combined, sharp-spikes and all HFO subtypes were increased in the SOZ (p < 0.01). However, the increases were larger during the sleep recordings (p < 0.05). The area under the ROC curves for SOZ classification were significantly smaller for intraoperative sharp-spikes, fRonO, and fRonS rates (p < 0.05). HFOs and spikes were only significantly spatially reliable for a subset of the patients (p < 0.05). A failure to resect fRonO areas in the sleep recordings trended the most sensitive and accurate for predicting failure. In summary, HFO morphology is altered by anesthesia. Intraoperative SEEG recordings exhibit increased rates of HFOs in the SOZ, but their spatial distribution can differ from sleep recordings. Recording these biomarkers during non-REM sleep offers a more accurate delineation of the SOZ and possibly the epileptogenic zone.