Worrell, Gregory A.

Many efforts have been made to image the spatiotemporal electrical activity of the brain with the purpose of mapping its function and dysfunction as well as aiding the management of brain disorders. Here, we propose a non-conventional deep learning–based source imaging framework (DeepSIF) that provides robust and precise spatiotemporal estimates of underlying brain dynamics from noninvasive high-density electroencephalography (EEG) recordings. DeepSIF employs synthetic training data generated by biophysical models capable of modeling mesoscale brain dynamics. The rich characteristics of underlying brain sources are embedded in the realistic training data and implicitly learned by DeepSIF networks, avoiding complications associated with explicitly formulating and tuning priors in an optimization problem, as often is the case in conventional source imaging approaches. The performance of DeepSIF is evaluated by 1) a series of numerical experiments, 2) imaging sensory and cognitive brain responses in a total of 20 healthy subjects from three public datasets, and 3) rigorously validating DeepSIF’s capability in identifying epileptogenic regions in a cohort of 20 drug-resistant epilepsy patients by comparing DeepSIF results with invasive measurements and surgical resection outcomes. DeepSIF demonstrates robust and excellent performance, producing results that are concordant with common neuroscience knowledge about sensory and cognitive information processing as well as clinical findings about the location and extent of the epileptogenic tissue and outperforming conventional source imaging methods. The DeepSIF method, as a data-driven imaging framework, enables efficient and effective high-resolution functional imaging of spatiotemporal brain dynamics, suggesting its wide applicability and value to neuroscience research and clinical applications.

Abstract Epilepsy is well-recognized as a disorder of brain networks. There is a growing body of research to identify critical nodes within dynamic epileptic networks with the aim to target therapies that halt the onset and propagation of seizures. In parallel, intracranial neuromodulation, including deep brain stimulation and responsive neurostimulation, are well-established and expanding as therapies to reduce seizures in adults with focal-onset epilepsy; and there is emerging evidence for their efficacy in children and generalized-onset seizure disorders. The convergence of these advancing fields is driving an era of ‘network-guided neuromodulation’ for epilepsy. In this review, we distil the current literature on network mechanisms underlying neurostimulation for epilepsy. We discuss the modulation of key ‘propagation points’ in the epileptogenic network, focusing primarily on thalamic nuclei targeted in current clinical practice. These include (i) the anterior nucleus of thalamus, now a clinically approved and targeted site for open loop stimulation, and increasingly targeted for responsive neurostimulation; and (ii) the centromedian nucleus of the thalamus, a target for both deep brain stimulation and responsive neurostimulation in generalized-onset epilepsies. We discuss briefly the networks associated with other emerging neuromodulation targets, such as the pulvinar of the thalamus, piriform cortex, septal area, subthalamic nucleus, cerebellum and others. We report synergistic findings garnered from multiple modalities of investigation that have revealed structural and functional networks associated with these propagation points — including scalp and invasive EEG, and diffusion and functional MRI. We also report on intracranial recordings from implanted devices which provide us data on the dynamic networks we are aiming to modulate. Finally, we review the continuing evolution of network-guided neuromodulation for epilepsy to accelerate progress towards two translational goals: (i) to use pre-surgical network analyses to determine patient candidacy for neurostimulation for epilepsy by providing network biomarkers that predict efficacy; and (ii) to deliver precise, personalized and effective antiepileptic stimulation to prevent and arrest seizure propagation through mapping and modulation of each patients’ individual epileptogenic networks.

AbstractObjectiveThe medial temporal lobe (MTL) encodes and recalls memories and can be a predominant site for interictal spikes (IS) in patients with focal epilepsy. It is unclear whether memory deficits are due to IS in the MTL producing a transient decline. Here, we investigated whether IS in the MTL subregions and lateral temporal cortex impact episodic memory encoding and recall.MethodsSeventy‐eight participants undergoing presurgical evaluation for medically refractory focal epilepsy with depth electrodes placed in the temporal lobe participated in a verbal free recall task. IS were manually annotated during the pre‐encoding, encoding, and recall epochs. We examined the effect of IS on word recall using mixed‐effects logistic regression.ResultsIS in the left hippocampus (odds ratio [OR] = .73, 95% confidence interval [CI] = .63–.84, p < .001) and left middle temporal gyrus (OR = .46, 95% CI = .27–.78, p < .05) during word encoding decreased subsequent recall performance. Within the left hippocampus, this effect was specific for area CA1 (OR = .76, 95% CI = .66–.88, p < .01) and dentate gyrus (OR = .74, 95% CI = .62–.89, p < .05). IS in other MTL subregions or inferior and superior temporal gyrus and IS occurring during the prestimulus window did not affect word encoding (p > .05). IS during retrieval in right hippocampal (OR = .22, 95% CI = .08–.63, p = .01) and parahippocampal regions (OR = .24, 95% CI = .07–.8, p < .05) reduced the probability of recalling a word.SignificanceIS in medial and lateral temporal cortex contribute to transient memory decline during verbal episodic memory.

Abstract In drug-resistant focal epilepsy, interictal high-frequency oscillations (HFOs) recorded from intracranial EEG (iEEG) may provide clinical information for delineating epileptogenic brain tissue. The iEEG electrode contacts that contain HFO are hypothesized to delineate the epileptogenic zone; their resection should then lead to postsurgical seizure freedom. We test whether our prospective definition of clinically relevant HFO is in agreement with postsurgical seizure outcome. The algorithm is fully automated and is equally applied to all data sets. The aim is to assess the reliability of the proposed detector and analysis approach. We use an automated data-independent prospective definition of clinically relevant HFO that has been validated in data from two independent epilepsy centres. In this study, we combine retrospectively collected data sets from nine independent epilepsy centres. The analysis is blinded to clinical outcome. We use iEEG recordings during NREM sleep with a minimum of 12 epochs of 5 min of NREM sleep. We automatically detect HFO in the ripple (80–250 Hz) and in the fast ripple (250–500 Hz) band. There is no manual rejection of events in this fully automated algorithm. The type of HFO that we consider clinically relevant is defined as the simultaneous occurrence of a fast ripple and a ripple. We calculate the temporal consistency of each patient’s HFO rates over several data epochs within and between nights. Patients with temporal consistency &lt;50% are excluded from further analysis. We determine whether all electrode contacts with high HFO rate are included in the resection volume and whether seizure freedom (ILAE 1) was achieved at ≥2 years follow-up. Applying a previously validated algorithm to a large cohort from several independent epilepsy centres may advance the clinical relevance and the generalizability of HFO analysis as essential next step for use of HFO in clinical practice.

Abstract Early implantable epilepsy therapy devices provided open-loop electrical stimulation without brain sensing, computing, or an interface for synchronized behavioural inputs from patients. Recent epilepsy stimulation devices provide brain sensing but have not yet developed analytics for accurately tracking and quantifying behaviour and seizures. Here we describe a distributed brain co-processor providing an intuitive bi-directional interface between patient, implanted neural stimulation and sensing device, and local and distributed computing resources. Automated analysis of continuous streaming electrophysiology is synchronized with patient reports using a handheld device and integrated with distributed cloud computing resources for quantifying seizures, interictal epileptiform spikes and patient symptoms during therapeutic electrical brain stimulation. The classification algorithms for interictal epileptiform spikes and seizures were developed and parameterized using long-term ambulatory data from nine humans and eight canines with epilepsy, and then implemented prospectively in out-of-sample testing in two pet canines and four humans with drug-resistant epilepsy living in their natural environments. Accurate seizure diaries are needed as the primary clinical outcome measure of epilepsy therapy and to guide brain-stimulation optimization. The brain co-processor system described here enables tracking interictal epileptiform spikes, seizures and correlation with patient behavioural reports. In the future, correlation of spikes and seizures with behaviour will allow more detailed investigation of the clinical impact of spikes and seizures on patients.

Abstract This study describes a generalized cross‐patient seizure‐forecasting approach using recurrent neural networks with ultra‐long‐term subcutaneous EEG (sqEEG) recordings. Data from six patients diagnosed with refractory epilepsy and monitored with an sqEEG device were used to develop a generalized algorithm for seizure forecasting using long short‐term memory (LSTM) deep‐learning classifiers. Electrographic seizures were identified by a board‐certified epileptologist. One‐minute data segments were labeled as preictal or interictal based on their relationship to confirmed seizures. Data were separated into training and testing data sets, and to compensate for the unbalanced data ratio in training, noise‐added copies of preictal data segments were generated to expand the training data set. The mean and standard deviation (SD) of the training data were used to normalize all data, preserving the pseudo‐prospective nature of the analysis. Different architecture classifiers were trained and tested using a leave‐one‐patient‐out cross‐validation method, and the area under the receiver‐operating characteristic (ROC) curve (AUC) was used to evaluate the performance classifiers. The importance of each input signal was evaluated using a leave‐one‐signal‐out method with repeated training and testing for each classifier. Cross‐patient classifiers achieved performance significantly better than chance in four of the six patients and an overall mean AUC of 0.602 ± 0.126 (mean ± SD). A time in warning of 37.386% ± 5.006% (mean ± std) and sensitivity of 0.691 ± 0.068 (mean ± std) were observed for patients with better than chance results. Analysis of input channels showed a significant contribution ( p < .05) by the Fourier transform of signals channels to overall classifier performance. The relative contribution of input signals varied among patients and architectures, suggesting that the inclusion of all signals contributes to robustness in a cross‐patient classifier. These early results show that it is possible to forecast seizures training with data from different patients using two‐channel ultra‐long‐term sqEEG.

Abstract 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 &gt; 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 &lt; 1 × 10−5). By contrast, in the non-responders, fast ripple had a lower spectral content in the seizure onset zone (P &lt; 1 × 10−5). We constructed two different networks of fast ripple with a spectral content &gt;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.

Abstract Objective Anterior temporal lobectomy (ATL) is a widely performed and successful intervention for drug‐resistant temporal lobe epilepsy (TLE). However, up to one third of patients experience seizure recurrence within 1 year after ATL. Despite the extensive literature on presurgical electroencephalography (EEG) and magnetic resonance imaging (MRI) abnormalities to prognosticate seizure freedom following ATL, the value of quantitative analysis of visually reviewed normal interictal EEG in such prognostication remains unclear. In this retrospective multicenter study, we investigate whether machine learning analysis of normal interictal scalp EEG studies can inform the prediction of postoperative seizure freedom outcomes in patients who have undergone ATL. Methods We analyzed normal presurgical scalp EEG recordings from 41 Mayo Clinic (MC) and 23 Cleveland Clinic (CC) patients. We used an unbiased automated algorithm to extract eyes closed awake epochs from scalp EEG studies that were free of any epileptiform activity and then extracted spectral EEG features representing (a) spectral power and (b) interhemispheric spectral coherence in frequencies between 1 and 25 Hz across several brain regions. We analyzed the differences between the seizure‐free and non–seizure‐free patients and employed a Naïve Bayes classifier using multiple spectral features to predict surgery outcomes. We trained the classifier using a leave‐one‐patient‐out cross‐validation scheme within the MC data set and then tested using the out‐of‐sample CC data set. Finally, we compared the predictive performance of normal scalp EEG‐derived features against MRI abnormalities. Results We found that several spectral power and coherence features showed significant differences correlated with surgical outcomes and that they were most pronounced in the 10–25 Hz range. The Naïve Bayes classification based on those features predicted 1‐year seizure freedom following ATL with area under the curve (AUC) values of 0.78 and 0.76 for the MC and CC data sets, respectively. Subsequent analyses revealed that (a) interhemispheric spectral coherence features in the 10–25 Hz range provided better predictability than other combinations and (b) normal scalp EEG‐derived features provided superior and potentially distinct predictive value when compared with MRI abnormalities (>10% higher F1 score). Significance These results support that quantitative analysis of even a normal presurgical scalp EEG may help prognosticate seizure freedom following ATL in patients with drug‐resistant TLE. Although the mechanism for this result is not known, the scalp EEG spectral and coherence properties predicting seizure freedom may represent activity arising from the neocortex or the networks responsible for temporal lobe seizure generation within vs outside the margins of an ATL.

The objective of this study was to characterize network-level changes in nonfluent/agrammatic Primary Progressive Aphasia (agPPA) and Primary Progressive Apraxia of Speech (PPAOS) with graph theory (GT) measures derived from scalp electroencephalography (EEG) recordings. EEGs of 15 agPPA and 7 PPAOS patients were collected during relaxed wakefulness with eyes closed (21 electrodes, 10–20 positions, 256 Hz sampling rate, 1–200 Hz bandpass filter). Eight artifact-free, non-overlapping 1024-point epochs were selected. Via Brainwave software, GT weighted connectivity and minimum spanning tree (MST) measures were calculated for theta and upper and lower alpha frequency bands. Differences in GT and MST measures between agPPA and PPAOS were assessed with Wilcoxon rank-sum tests. Of greatest interest, Spearman correlations were computed between behavioral and network measures in all frequency bands across all patients. There were no statistically significant differences in GT or MST measures between agPPA and PPAOS. There were significant correlations between several network and behavioral variables. The correlations demonstrate a relationship between reduced global efficiency and clinical symptom severity (e.g., parkinsonism, AOS). This preliminary, exploratory study demonstrates potential for EEG GT measures to quantify network changes associated with degenerative speech–language disorders.