Semiautomated classification of nocturnal seizures using video recordings.

OBJECTIVE: The objective of this study was to evaluate the accuracy of a semiautomated classification of nocturnal seizures using a hybrid system consisting of an artificial intelligence-based algorithm, which selects epochs with potential clinical relevance to be reviewed by human experts. METHODS: Consecutive patients with nocturnal motor seizures admitted for video-electroencephalographic long-term monitoring (LTM) were prospectively recruited. We determined the extent of data reduction by using the algorithm, and we evaluated the accuracy of seizure classification from the hybrid system compared with the gold standard of LTM. RESULTS: Forty consecutive patients (24 male; median age = 15 years) were analyzed. The algorithm reduced the duration of epochs to be reviewed to 14% of the total recording time (1874 h). There was a fair agreement beyond chance in seizure classification between the hybrid system and the gold standard (agreement coefficient = .33, 95% confidence interval = .20-.47). The hybrid system correctly identified all tonic-clonic and clonic seizures and 82% of focal motor seizures. However, there was low accuracy in identifying seizure types with more discrete or subtle motor phenomena. SIGNIFICANCE: Using a hybrid (algorithm-human) system for reviewing nocturnal video recordings significantly decreased the workload and provided accurate classification of major motor seizures (tonic-clonic, clonic, and focal motor seizures).

Seizure semiology, localization, and the 2017 ILAE seizure classification.

In the study of epilepsy, the term semiology is used to comprise the clinical characteristics of a seizure, both subjective symptoms and objective phenomena. It is produced by activation of the symptomagenic zone, and an accurate and comprehensive understanding of the localizing value of seizure semiology is crucial for presurgical evaluation and planning. Myriad publications in epilepsy journals detail correlations between various semiological features and activation of specific cortical regions. Traditionally these studies involved scalp EEG recorded in epilepsy monitoring units. The increasing use of invasive monitoring, and specifically the use of depth electrodes and stereo-electroencephalography, has advanced our understanding of the characteristics of seizures arising from ictal foci deep to the scalp, including the cingulate, insula and operculum. However, the distinction between seizure onset and symptomogenic zones is not always clear. In 2017 the International League Against Epilepsy (ILAE) published an operational classification of seizure types based heavily on seizure semiology. The current paper provides an updated review of the current body of knowledge relating to seizure semiology, incorporating both scalp EEG studies and more recent stereo-electroencephalography discoveries in the framework of the 2017 ILAE classification.

On seizure semiology.

The clinical expression of seizures represents the main symptomatic burden of epilepsy. Neural mechanisms of semiologic production in epilepsy, especially for complex behaviors, remain poorly known. In a framework of epilepsy as a network rather than as a focal disorder, we can think of semiology as being dynamically produced by a set of interconnected structures, in which specific rhythmic interactions, and not just anatomical localization, are likely to play an important part in clinical expression. This requires a paradigm shift in how we think about seizure organization, including from a presurgical evaluation perspective. Semiology is a key data source, albeit with significant methodological challenges for its use in research, including observer bias and choice of semiologic categories. Better understanding of semiologic categorization and pathophysiological correlates is relevant to seizure classification systems. Advances in knowledge of neural mechanisms as well as anatomic correlates of different semiologic patterns could help improve knowledge of epilepsy networks and potentially contribute to therapeutic innovations.

A propositional AI system for supporting epilepsy diagnosis based on the 2017 epilepsy classification: Illustrated by Dravet syndrome.

PURPOSE: The 2017 epilepsy and seizure diagnosis framework emphasizes epilepsy syndromes and the etiology-based approach. We developed a propositional artificial intelligence (AI) system based on the above concepts to support physicians in the diagnosis of epilepsy. METHODS: We analyzed and built ontology knowledge for the classification of seizure patterns, epilepsy, epilepsy syndrome, and etiologies. Protégé ontology tool was applied in this study. In order to enable the system to be close to the inferential thinking of clinical experts, we classified and constructed knowledge of other epilepsy-related knowledge, including comorbidities, epilepsy imitators, epilepsy descriptors, characteristic electroencephalography (EEG) findings, treatments, etc. We used the Ontology Web Language with Description Logic (OWL-DL) and Semantic Web Rule Language (SWRL) to design rules for expressing the relationship between these ontologies. RESULTS: Dravet syndrome was taken as an illustration for epilepsy syndromes implementation. We designed an interface for the physician to enter the various characteristics of the patients. Clinical data of an 18-year-old boy with epilepsy was applied to the AI system. Through SWRL and reasoning engine Drool's execution, we successfully demonstrate the process of differential diagnosis. CONCLUSION: We developed a propositional AI system by using the OWL-DL/SWRL approach to deal with the complexity of current epilepsy diagnosis. The experience of this system, centered on the clinical epilepsy syndromes, paves a path to construct an AI system for further complicated epilepsy diagnosis.

Epileptic seizure classifications using empirical mode decomposition and its derivative.

BACKGROUND: Epilepsy is one of the most common neurological disorders associated with disruption of brain activity. In the classification and detection of epileptic seizures, electroencephalography (EEG) measurements, which record the electrical activities of the brain, are frequently used. Empirical mode decomposition (EMD) and its derivative, ensemble EMD (EEMD) are recently developed methods used to decompose non-stationary and nonlinear signals such as EEG into a finite number of oscillations called intrinsic mode functions (IMFs). Our main objective in this study is to present a hybrid IMF selection method combining four different approaches (energy, correlation, power spectral distance, and statistical significance measures), and investigate the effect of selected IMFs extracted by EMD and EEMD on the classification. We have applied the proposed IMF selection approach on the classification of EEG signals recorded from epilepsy patients who are under treatment at our collaborator hospital. Multichannel EEG signals collected from epilepsy patients are decomposed into IMFs, and then IMF selection was performed. Finally, time- and spectral-domain, and nonlinear features are extracted and feature sets are created for the classification. RESULTS: The maximum classification accuracies obtained using various combinations of IMFs were 94.56%, 95.63%, 96.8%, and 96.25% for SVM, KNN, naive Bayes, and logistic regression classifiers, respectively, by using EMD analysis; whereas, the EEMD approach has provided maximum classification accuracies of 96.06%, 97%, 97%, and 96.25% for SVM, KNN, naive Bayes, and logistic regression, respectively. Classification performance with the same features obtained using direct EEG signals instead of the decomposed IMFs was worse than the aforementioned 2 approaches for every combination. CONCLUSION: Simulation results demonstrate that the proposed IMF selection approach affects the classification results. Also, EEMD provides a robust method for feature extraction from EEG signals in order to classify pre-seizure and seizure segments.

Fast ripple analysis in human mesial temporal lobe epilepsy suggests two different seizure-generating mechanisms.

OBJECTIVE: The distinction of hypersynchronous (HYP) and low-voltage fast (LVF) onset seizures in mesial temporal lobe epilepsy (MTLE) is well established, but classifying individual seizures and patients is often challenging. Experimental work indicates a strong association of HYP with fast ripples (250-500 Hz) and of LVF with ripples (80-250 Hz). We aimed to investigate whether analysis of high-frequency oscillations can be useful for characterizing the process of seizure generation in human MTLE patients. METHODS: We retrospectively compared 19 HYP and 14 LVF onset clinical seizures from ten and six consecutive MTLE patients with a predominance of the respective pattern. Five-second intervals of stereotactic EEGs from the seizure onset zone were selected, each representing the onset of HYP and LVF, the corresponding pre-ictal periods and, after the large spikes of HYP onsets, the LVF-like pattern that frequently followed. RESULTS: Pre-ictal fast ripple density and rate were higher for HYP than for LVF seizures (p < .05). This association was also found for initial ictal segments (p < .001). Furthermore, fast ripple density and rate were higher during the LVF-like pattern after HYP spikes than during LVF without preceding HYP (p < .01). Ripple density and rate in contrast did not differ significantly (p > .05). Fast ripple (p < .01) and ripple (p < .001) amplitude was higher during the LVF-like pattern after HYP spikes when compared to LVF without preceding HYP. SIGNIFICANCE: Our findings indicate a clear connection between experimental findings and human epilepsy. The association of fast ripples with HYP suggests that out-of-phase firing of different pyramidal cell clusters contributes specifically to generation of these seizures, rather than to LVF onsets. Both during and immediately before seizures, fast ripple analysis may facilitate classification.

Seizure onset predicts its type.

OBJECTIVE: Epilepsy is characterized by transient alterations in brain synchronization resulting in seizures with a wide spectrum of manifestations. Seizure severity and risks for patients depend on the evolution and spread of the hypersynchronous discharges. With standard visual inspection and pattern classification, this evolution could not be predicted early on. It is still unclear to what degree the seizure onset zone determines seizure severity. Such information would improve our understanding of ictal epileptic activity and the existing electroencephalogram (EEG)-based warning and intervention systems, providing specific reactions to upcoming seizure types. We investigate the possibility of predicting the future development of an epileptic seizure during the first seconds of recordings after their electrographic onset. METHODS: Based on intracranial EEG recordings of 493 ictal events from 26 patients with focal epilepsy, a set of 25 time and frequency domain features was computed using nonoverlapping 1-second time windows, from the first 3, 5, and 10 seconds of ictal EEG. Three random forest classifiers were trained to predict the future evolution of the seizure, distinguishing between subclinical events, focal onset aware and impaired awareness, and focal to bilateral tonic-clonic seizures. RESULTS: Results show that early seizure type prediction is possible based on a single EEG channel located in the seizure onset zone with correct prediction rates of 76.2 ± 14.5% for distinguishing subclinical electrographic events from clinically manifest seizures, 75 ± 16.8% for distinguishing focal onset seizures that are or are not bilateral tonic-clonic, and 71.4 ± 17.2% for distinguishing between focal onset seizures with or without impaired awareness. All predictions are above the chance level (P < .01). SIGNIFICANCE: These findings provide the basis for developing systems for specific early warning of patients and health care providers, and for targeting EEG-based closed-loop intervention approaches to electrographic patterns with a high inherent risk to become clinically manifest.

Differential diagnosis of a paroxysmal neurological event: Do neurologists know how to clinically recognize it?

PURPOSE: To investigate ability to recognize paroxysmal neurological events (PNE) based on video-recorded events alone in a group of physicians treating prevalent neurological conditions. METHODS: Total of 12 patients' videos (6 epileptic seizures (ES), 4 psychogenic nonepileptic seizures (PNES), 2 other nonepileptic seizures (oNES)) were selected. Videos were displayed once to physicians blind to clinical data and final diagnosis. Physicians determined their clinical choice: ES, PNES, oNES, and I don't know (IDK). When ES was chosen, subjects determined type of ES: focal ES, secondary generalized tonic-clonic seizure (GTCS), primary GTCS, and IDK. RESULTS: In total 145 physicians (62% female, mean age 46.2±9years) (neurologists 58.6%, neuropsychiatrists 25.5%, psychiatrists 5%, and neurology residents 10.3%) were enrolled. Physician's exposure to patients with epilepsy per week was diverse: ≤1 patient (43.7%); 1-7 patients (37.2%); >7 patients (14.5%). Reported frequency of observation of PNE was as follows: frequent (21.4%), sometimes (47.6%); rarely (26.9%); never (2.1%). Majority of subjects were not EEG readers (60.7%). Median percentage (Mdn%) of correct answers (CA) was 75% (range 25-100). Predictor of better PNE recognition was higher frequency of clinical exposure to PNE (OR 1.65; CI95% 1.11-2.45; p=0.013). Mdn% of ES CA was 83.3%, (range 33.3-100), and of PNES CA was 50% (range 0-100). Physicians were more accurate in ES than PNES identification (p<0,001). Mdn% of type of ES CA was 50%, (range 0-100). CONCLUSIONS: We demonstrate the need for education about clinical features of PNE across subgroups of physicians who deliver neurological service, with emphasis on PNES and ES type classification.

Generalized onset seizures with focal evolution (GOFE) - A unique seizure type in the setting of generalized epilepsy.

PURPOSE: We report clinical and electrographic features of generalized onset seizures with focal evolution (GOFE) and present arguments for the inclusion of this seizure type in the seizure classification. METHODS: The adult and pediatric Epilepsy Monitoring Unit databases at Vanderbilt Medical Center and Children's Hospital were screened to identify generalized onset seizures with focal evolution. We reviewed medical records for epilepsy characteristics, epilepsy risk factors, MRI abnormalities, neurologic examination, antiepileptic medications before and after diagnosis, and response to medications. We also reviewed ictal and interictal EEG tracings, as well as video-recorded semiology. RESULTS: Ten patients were identified, 7 males and 3 females. All of the patients developed generalized epilepsy in childhood or adolescence (ages 3-15years). Generalized onset seizures with focal evolution developed years after onset in 9 patients, with a semiology concerning for focal seizures or nonepileptic events. Ictal discharges had a generalized onset on EEG, described as either generalized spike-and-wave and/or polyspike-and-wave discharges, or generalized fast activity. This electrographic activity then evolved to focal rhythmic activity most commonly localized to one temporal or frontal region; five patients had multiple seizures evolving to focal activity in different regions of both hemispheres. The predominant interictal epileptiform activity included generalized spike-and-wave and/or polyspike-and-wave discharges in all patients. Taking into consideration all clinical and EEG data, six patients were classified with genetic (idiopathic) generalized epilepsy, and four were classified with structural/metabolic (symptomatic) generalized epilepsy. All of the patients had modifications to their medications following discharge, with three becoming seizure-free and five responding with >50% reduction in seizure frequency. CONCLUSION: Generalized onset seizures may occasionally have focal evolution with semiology suggestive of focal seizures, leading to a misdiagnosis of focal onset. This unique seizure type may occur with genetic as well as structural/metabolic forms of epilepsy. The identification of this seizure type may help clinicians choose appropriate medications, avoiding narrow spectrum agents known to aggravate generalized onset seizures.

The influence of staffing on diagnostic yield of EMU admissions: a comparison study between two institutions.

Patients with recurrent, stereotyped neurological events of unclear etiology often warrant admission for continuous video-EEG monitoring in an epilepsy monitoring unit (EMU) for diagnosis. Epilepsy monitoring unit admission duration has been reported to range from days to weeks. To date, there are limited data on the average duration of admission for patients admitted to the EMU for spell classification. Many EMUs are forced to limit the overall duration of admission for numerous reasons including limited resources. It is unclear if a time-limited EMU stay reduces the event capture rate and, therefore, diagnostic yield of event classification admissions. The goal of this study was to determine how a time-limited length of stay strategy impacted event capture in patients admitted for spell classification. A retrospective chart review was performed at two comparable adult epilepsy monitoring units, Mayo Clinic Hospital (MCH) in Phoenix, Arizona, and Banner Good Samaritan Medical Center (BGSMC) in Phoenix, Arizona. Banner Good Samaritan Medical Center is only staffed Monday through Friday, thereby limiting the total possible duration of admission to five days. The goal was to determine if the rate of event capture differed between two institutions employing a time-limited EMU admission (BGSMC) when compared with the nonlimited admission (MCH). A total of 300 patient admissions at MCH and 260 patient admissions at BGSMC were reviewed over a comparable time period. The event capture rates at MCH and BGSMC were 74% and 72%, respectively. There was a greater percentage of patients with nonepileptic events (NEEs) at MCH than at BGSMC (62.7% vs. 47.3%). The mean duration until first event was 31h at MCH and 38 h at BGSMC. The mean length of stay was greater at MCH (4.5 days) when compared with BGSMC (3.3 days). The overall diagnostic yield of a time-limited EMU admission was similar to that of a nonlimited admission for the purpose of spell classification. There was a statistically significant difference when comparing the time until first event at both institutions; however, this still fell within the 5-day duration that the time-restricted admission was limited to. These results may be important in optimizing an EMU practice in patients requiring admission for spell classification.

Generalized onset seizures with focal evolution (GOFE) - a largely unknown ictal variation in genetic generalized epilepsies.

PURPOSE: At onset of generalized seizures, focal electroclinical features are commonly seen, while generalized onset seizures with focal evolution (GOFE) are largely unknown bearing the risk of misclassification. METHODS: In two German epilepsy-centers, patients with GOFE documented by video-EEG monitoring (VEM) between 2017 and 2022 were identified retrospectively. In addition to analysis of ictal electroclinical features, detailed epilepsy and family history, response to antiseizure medication (ASM), and findings from neuroimaging were considered. RESULTS: We identified five patients with GOFE, three females, age 14 to 22 years. All patients developed genetic generalized epilepsy in childhood or adolescence, each presenting with two or three generalized seizure types. In each of the five patients, one GOFE was recorded by VEM. At onset, EEG seizure patterns were characterized by generalized spike-wave discharges at 2.5 to 3.5/sec for 9 to 16 s followed by focal evolution of the discharges. Interictally, all patients presented with generalized spike-wave discharges without focal abnormalities. Semiology at onset was behavioral arrest in two patients and generalized increase in tone in one, while two onsets were clinically inapparent. Semiological signs during focal evolution were variable, comprising head and body version, figure 4 sign, unilateral arm clonic activity, and staring with oral automatisms. In one case, focality involved both hemispheres successively. CONCLUSION: Prominent focal semiological features in GOFE carry a high risk of misclassification as focal seizures and epilepsy and thus wrong choice of ASM. This calls for low-threshold VEM if any doubts of focal genesis of seizures exist.

A generalized seizure type: Myoclonic-to-tonic seizure.

BACKGROUND AND PURPOSE: To test the hypothesis that myoclonic seizures can evolve to tonic seizures, we documented the electroclinical features of this under-recognized seizure type. METHODS: We observed a distinct seizure pattern starting with myoclonus without returning to an interictal state, which subsequently evolved into generalized tonic seizures. The detailed symptomatic and electroencephalographic characteristics of this seizure were extracted, and the clinical manifestations, drug curative responses in patients with this seizure were reviewed and analyzed. RESULTS: The onset of all seizures was characterized by a preceding period of myoclonus and bursts of generalized spike or poly-spike slow wave discharges with high amplitude. This was closely followed by the occurrence of tonic seizures, which were distinguished by bursts of generalized fast activity at 10 Hz or higher frequency. This under-recognized seizure type has been designated as myoclonic-to-tonic (MT) seizure. The number of patients identified with MT seizures in this study was 34. The prevalence rate of MT seizures was found to be higher in males. While MT seizures typically included a tonic component, it should be noted that some patients experiencing this seizure type never presented with isolated tonic seizures. Generalized Epilepsy not further defined (GE) accounted for approximately one-third of the diagnosed cases, followed by Lennox-Gastaut syndrome and Epilepsy with Myoclonic-Atonic seizures. In comparison to other types of epilepsy, GE with MT seizures demonstrated a more favorable prognosis. CONCLUSIONS: The classification of myoclonic-to-tonic seizure represents a novel approach in comprehending the ictogenesis of generalized seizures and can provide valuable assistance to clinicians in epilepsy diagnosis.

Classification of epileptic seizures in EEG data based on iterative gated graph convolution network.

Introduction: The automatic and precise classification of epilepsy types using electroencephalogram (EEG) data promises significant advancements in diagnosing patients with epilepsy. However, the intricate interplay among multiple electrode signals in EEG data poses challenges. Recently, Graph Convolutional Neural Networks (GCN) have shown strength in analyzing EEG data due to their capability to describe complex relationships among different EEG regions. Nevertheless, several challenges remain: (1) GCN typically rely on predefined or prior graph topologies, which may not accurately reflect the complex correlations between brain regions. (2) GCN struggle to capture the long-temporal dependencies inherent in EEG signals, limiting their ability to effectively extract temporal features. Methods: To address these challenges, we propose an innovative epileptic seizure classification model based on an Iterative Gated Graph Convolutional Network (IGGCN). For the epileptic seizure classification task, the original EEG graph structure is iteratively optimized using a multi-head attention mechanism during training, rather than relying on a static, predefined prior graph. We introduce Gated Graph Neural Networks (GGNN) to enhance the model's capacity to capture long-term dependencies in EEG series between brain regions. Additionally, Focal Loss is employed to alleviate the imbalance caused by the scarcity of epileptic EEG data. Results: Our model was evaluated on the Temple University Hospital EEG Seizure Corpus (TUSZ) for classifying four types of epileptic seizures. The results are outstanding, achieving an average F1 score of 91.5% and an average Recall of 91.8%, showing a substantial improvement over current state-of-the-art models. Discussion: Ablation experiments verified the efficacy of iterative graph optimization and gated graph convolution. The optimized graph structure significantly differs from the predefined EEG topology. Gated graph convolutions demonstrate superior performance in capturing the long-term dependencies in EEG series. Additionally, Focal Loss outperforms other commonly used loss functions in the TUSZ classification task.

Advancing post-traumatic seizure classification and biomarker identification: Information decomposition based multimodal fusion and explainable machine learning with missing neuroimaging data.

A late post-traumatic seizure (LPTS), a consequence of traumatic brain injury (TBI), can potentially evolve into a lifelong condition known as post-traumatic epilepsy (PTE). Presently, the mechanism that triggers epileptogenesis in TBI patients remains elusive, inspiring the epilepsy community to devise ways to predict which TBI patients will develop PTE and to identify potential biomarkers. In response to this need, our study collected comprehensive, longitudinal multimodal data from 48 TBI patients across multiple participating institutions. A supervised binary classification task was created, contrasting data from LPTS patients with those without LPTS. To accommodate missing modalities in some subjects, we took a two-pronged approach. Firstly, we extended a graphical model-based Bayesian estimator to directly classify subjects with incomplete modality. Secondly, we explored conventional imputation techniques. The imputed multimodal information was then combined, following several fusion and dimensionality reduction techniques found in the literature, and subsequently fitted to a kernel- or a tree-based classifier. For this fusion, we proposed two new algorithms: recursive elimination of correlated components (RECC) that filters information based on the correlation between the already selected features, and information decomposition and selective fusion (IDSF), which effectively recombines information from decomposed multimodal features. Our cross-validation findings showed that the proposed IDSF algorithm delivers superior performance based on the area under the curve (AUC) score. Ultimately, after rigorous statistical comparisons and interpretable machine learning examination using Shapley values of the most frequently selected features, we recommend the two following magnetic resonance imaging (MRI) abnormalities as potential biomarkers: the left anterior limb of internal capsule in diffusion MRI (dMRI), and the right middle temporal gyrus in functional MRI (fMRI).

Multiband seizure type classification based on 3D convolution with attention mechanisms.

Electroencephalogram (EEG) signal contains important information about abnormal brain activity, which has become an important basis for epilepsy diagnosis. Recently, epilepsy EEG signal classification methods mainly extract features from the perspective of a single domain, which cannot effectively utilize the spatial domain information in EEG signals. The redundant information in EEG signals will affect the learning features with the increase of convolution layer and multi-domain features, resulting in inefficient learning and a lack of distinguishing features. To tackle these issues, we propose an end-to-end 3D convolutional multiband seizure-type classification model based on attention mechanisms. Specifically, to process preprocessed electroencephalogram (EEG) data, a multilevel wavelet decomposition is applied to obtain the joint distribution information in the two-dimensional time-frequency domain across multiple frequency bands. Subsequently, this information is transformed into three-dimensional spatial data based on the electrode configuration. Discriminative joint activity features in the time, frequency, and spatial domains are then extracted by a series of parallel 3D convolutional sub-networks, where 3D channels and spatial attention mechanisms improve the ability to learn critical global and local information. A multi-layer perceptron is finally implemented to integrate the extracted features and further map them to the classification results. Experimental results on the TUSZ dataset, the world's largest publicly available seizure corpus, show that 3D-CBAMNet significantly outperforms the state-of-the-art methods, indicating effectiveness in the seizure type classification task.

Automatic classification of hyperkinetic, tonic, and tonic-clonic seizures using unsupervised clustering of video signals.

Introduction: This study evaluated the accuracy of motion signals extracted from video monitoring data to differentiate epileptic motor seizures in patients with drug-resistant epilepsy. 3D near-infrared video was recorded by the Nelli® seizure monitoring system (Tampere, Finland). Methods: 10 patients with 130 seizures were included in the training dataset, and 17 different patients with 98 seizures formed the testing dataset. Only seizures with unequivocal hyperkinetic, tonic, and tonic-clonic semiology were included. Motion features from the catch22 feature collection extracted from video were explored to transform the patients' videos into numerical time series for clustering and visualization. Results: Changes in feature generation provided incremental discrimination power to differentiate between hyperkinetic, tonic, and tonic-clonic seizures. Temporal motion features showed the best results in the unsupervised clustering analysis. Using these features, the system differentiated hyperkinetic, tonic and tonic-clonic seizures with 91, 88, and 45% accuracy after 100 cross-validation runs, respectively. F1-scores were 93, 90, and 37%, respectively. Overall accuracy and f1-score were 74%. Conclusion: The selected features of motion distinguished semiological differences within epileptic seizure types, enabling seizure classification to distinct motor seizure types. Further studies are needed with a larger dataset and additional seizure types. These results indicate the potential of video-based hybrid seizure monitoring systems to facilitate seizure classification improving the algorithmic processing and thus streamlining the clinical workflow for human annotators in hybrid (algorithmic-human) seizure monitoring systems.

Expert and deep learning model identification of iEEG seizures and seizure onset times.

Hundreds of 90-s iEEG records are typically captured from each NeuroPace RNS System patient between clinic visits. While these records provide invaluable information about the patient's electrographic seizure and interictal activity patterns, manually classifying them into electrographic seizure/non-seizure activity, and manually identifying the seizure onset channels and times is an extremely time-consuming process. A convolutional neural network based Electrographic Seizure Classifier (ESC) model was developed in an earlier study. In this study, the classification model is tested against iEEG annotations provided by three expert reviewers board certified in epilepsy. The three experts individually annotated 3,874 iEEG channels from 36, 29, and 35 patients with leads in the mesiotemporal (MTL), neocortical (NEO), and MTL + NEO regions, respectively. The ESC model's seizure/non-seizure classification scores agreed with the three reviewers at 88.7%, 89.6%, and 84.3% which was similar to how reviewers agreed with each other (92.9%-86.4%). On iEEG channels with all 3 experts in agreement (83.2%), the ESC model had an agreement score of 93.2%. Additionally, the ESC model's certainty scores reflected combined reviewer certainty scores. When 0, 1, 2 and 3 (out of 3) reviewers annotated iEEG channels as electrographic seizures, the ESC model's seizure certainty scores were in the range: [0.12-0.19], [0.32-0.42], [0.61-0.70], and [0.92-0.95] respectively. The ESC model was used as a starting-point model for training a second Seizure Onset Detection (SOD) model. For this task, seizure onset times were manually annotated on a relatively small number of iEEG channels (4,859 from 50 patients). Experiments showed that fine-tuning the ESC models with augmented data (30,768 iEEG channels) resulted in a better validation performance (on 20% of the manually annotated data) compared to training with only the original data (3.1s vs 4.4s median absolute error). Similarly, using the ESC model weights as the starting point for fine-tuning instead of other model weight initialization methods provided significant advantage in SOD model validation performance (3.1s vs 4.7s and 3.5s median absolute error). Finally, on iEEG channels where three expert annotations of seizure onset times were within 1.5 s, the SOD model's seizure onset time prediction was within 1.7 s of expert annotation.

Unsupervised domain adaptation for cross-patient seizure classification.

Objective. Epileptic seizure is a chronic neurological disease affecting millions of patients. Electroencephalogram (EEG) is the gold standard in epileptic seizure classification. However, its low signal-to-noise ratio, strong non-stationarity, and large individual difference nature make it difficult to directly extend the seizure classification model from one patient to another. This paper considers multi-source unsupervised domain adaptation for cross-patient EEG-based seizure classification, i.e. there are multiple source patients with labeled EEG data, which are used to label the EEG trials of a new patient.Approach. We propose an source domain selection (SDS)-global domain adaptation (GDA)-target agent subdomain adaptation (TASA) approach, which includes SDS to filter out dissimilar source domains, GDA to align the overall distributions of the selected source domains and the target domain, and TASA to identify the most similar source domain to the target domain so that its labels can be utilized.Main results. Experiments on two public seizure datasets demonstrated that SDS-GDA-TASA outperformed 13 existing approaches in unsupervised cross-patient seizure classification.Significance. Our approach could save clinicians plenty of time in labeling EEG data for epilepsy patients, greatly increasing the efficiency of seizure diagnostics.

Measuring multivariate phase synchronization with symbolization and permutation.

Phase synchronization is an important mechanism for the information processing of neurons in the brain. Most of the current phase synchronization measures are bivariate and focus on the synchronization between pairs of time series. However, these methods do not provide a full picture of global interactions in neural systems. Considering the prevalence and importance of multivariate neural signal analysis, there is an urgent need to quantify global phase synchronization (GPS) in neural networks. Therefore, we propose a new measure named symbolic phase difference and permutation entropy (SPDPE), which symbolizes the phase difference in multivariate neural signals and estimates GPS according to the permutation patterns of the symbolic sequences. The performance of SPDPE was evaluated using simulated data generated by Kuramoto and Rössler model. The results demonstrate that SPDPE exhibits low sensitivity to data length and outperforms existing methods in accurately characterizing GPS and effectively resisting noise. Moreover, to validate the method with real data, it was applied to classify seizures and non-seizures by calculating the GPS of stereoelectroencephalography (SEEG) data recorded from the onset zones of ten epilepsy patients. We believe that SPDPE will improve the estimation of GPS in many applications, such as EEG-based brain-computer interfaces, brain modeling, and simultaneous EEG-fMRI analysis.

Variational mode decomposition and binary grey wolf optimization-based automated epilepsy seizure classification framework.

This work proposes a variational mode decomposition (VMD) and binary grey wolf optimization (BGWO) based seizure classification framework. VMD decomposes the EEG signal into band-limited intrinsic mode function (BL-IMFs) non-recursively. The frequency domain, time domain, and information theory-based features are extracted from the BL-IMFs. Further, an optimal feature subset is selected using BGWO. Finally, the selected features were utilized for classification using six different supervised machine learning algorithms. The proposed framework has been validated experimentally by 58 test cases from the CHB-MIT scalp EEG and the Bonn University database. The proposed framework performance is quantified by average sensitivity, specificity, and accuracy. The selected features, along with Bayesian regularized shallow neural networks (BR-SNNs), resulted in maximum accuracy of 99.53 and 99.64 for 1 and 2 s epochs, respectively, for database 1. The proposed framework has achieved 99.79 and 99.84 accuracy for 1 and 2 s epochs, respectively, for database 2.