EEG interpretation reliability and interpreter confidence: a large single-center study

Unveiling variability: A systematic review of reproducibility in visual EEG analysis, with focus on seizures

OBJECTIVE

Reproducibility is key for diagnostic tests involving subjective evaluation by experts. Our aim was to systematically review the reproducibility of visual analysis in clinical electroencephalogram (EEG). In this paper, we give data on the scope of EEG features found, and detailed reproducibility data for the most studied feature.

METHODS

We searched four databases for articles reporting reproducibility in clinical EEG, until June 2023. Two raters screened 24 553 citations, and then 2736 full texts. Quality was assessed according to the GRRAS guidelines.

RESULTS

We found 275 studies (268 interrater and 20 intrarater), addressing 606 different EEG features. Only 38 EEG features had been studied in >2 studies. Most studies had <50 patients and EEGs. The most often addressed feature was seizure detection (62 papers). Interrater reproducibility of seizure detection was substantial-to-almost-perfect with experienced raters and raw EEG (kappa .62-.88). With experienced raters and transformed EEG, reproducibility was substantial (kappa .63-.70). Inexperienced raters had lower reproducibility. Seizure lateralization reproducibility was moderate to substantial (kappa .58-.77) but lower than for seizure detection.

SIGNIFICANCE

Most EEG reproducibility studies are done only once. Intrarater studies are rare. The reproducibility of visual EEG analysis is variable. Interrater reproducibility for seizure detection is substantial-to-perfect with experienced raters and raw EEG, less with inexperienced raters or transformed EEG. The results of visual EEG analysis vary within the same rater, and between raters. There is a need for larger collaborative studies, using improved methodology, as well as more intrarater studies of EEG interpretation.

Machine learning algorithm for predicting seizure control after temporal lobe resection using peri-ictal electroencephalography

Brain resection is curative for a subset of patients with drug resistant epilepsy but up to half will fail to achieve sustained seizure freedom in the long term. There is a critical need for accurate prediction tools to identify patients likely to have recurrent postoperative seizures. Results from preclinical models and intracranial EEG in humans suggest that the window of time immediately before and after a seizure ("peri-ictal") represents a unique brain state with implications for clinical outcome prediction. Using a dataset of 294 patients who underwent temporal lobe resection for seizures, we show that machine learning classifiers can make accurate predictions of postoperative seizure outcome using 5 min of peri-ictal scalp EEG data that is part of universal presurgical evaluation (AUC 0.98, out-of-group testing accuracy > 90%). This is the first approach to seizure outcome prediction that employs a routine non-invasive preoperative study (scalp EEG) with accuracy range likely to translate into a clinical tool. Decision curve analysis (DCA) shows that compared to the prevalent clinical-variable based nomogram, use of the EEG-augmented approach could decrease the rate of unsuccessful brain resections by 20%.

Differentiating loss of consciousness causes through artificial intelligence-enabled decoding of functional connectivity

Differential diagnosis of acute loss of consciousness (LOC) is crucial due to the need for different therapeutic strategies despite similar clinical presentations among etiologies such as nonconvulsive status epilepticus, metabolic encephalopathy, and benzodiazepine intoxication. While altered functional connectivity (FC) plays a pivotal role in the pathophysiology of LOC, there has been a lack of efforts to develop differential diagnosis artificial intelligence (AI) models that feature the distinctive FC change patterns specific to each LOC cause. Three approaches were applied for extracting features for the AI models: three-dimensional FC adjacency matrices, vectorized FC values, and graph theoretical measurements. Deep learning using convolutional neural networks (CNN) and various machine learning algorithms were implemented to compare classification accuracy using electroencephalography (EEG) data with different epoch sizes. The CNN model using FC adjacency matrices achieved the highest accuracy with an AUC of 0.905, with 20-s epoch data being optimal for classifying the different LOC causes. The high accuracy of the CNN model was maintained in a prospective cohort. Key distinguishing features among the LOC causes were found in the delta and theta brain wave bands. This research advances the understanding of LOC's underlying mechanisms and shows promise for enhancing diagnosis and treatment selection. Moreover, the AI models can provide accurate LOC differentiation with a relatively small amount of EEG data in 20-s epochs, which may be clinically useful.

Iteratively Calibratable Network for Reliable EEG-Based Robotic Arm Control

Robotic arms are increasingly being utilized in shared workspaces, which necessitates the accurate interpretation of human intentions for both efficiency and safety. Electroencephalogram (EEG) signals, commonly employed to measure brain activity, offer a direct communication channel between humans and robotic arms. However, the ambiguous and unstable characteristics of EEG signals, coupled with their widespread distribution, make it challenging to collect sufficient data and hinder the calibration performance for new signals, thereby reducing the reliability of EEG-based applications. To address these issues, this study proposes an iteratively calibratable network aimed at enhancing the reliability and efficiency of EEG-based robotic arm control systems. The proposed method integrates feature inputs with network expansion techniques. This integration allows a network trained on an extensive initial dataset to adapt effectively to new users during calibration. Additionally, our approach combines motor imagery and speech imagery datasets to increase not only its intuitiveness but also the number of command classes. The evaluation is conducted in a pseudo-online manner, with a robotic arm operating in real-time to collect data, which is then analyzed offline. The evaluation results demonstrated that the proposed method outperformed the comparison group in 10 sessions and demonstrated competitive results when the two paradigms were combined. Therefore, it was confirmed that the network can be calibrated and personalized using only the new data from new users.

Reaching the ceiling? Empirical scaling behaviour for deep EEG pathology classification

Machine learning techniques, particularly deep convolutional neural networks (ConvNets), are increasingly being used to automate clinical EEG analysis, with the potential to reduce the clinical burden and improve patient care. However, further research is required before they can be used in clinical settings, particularly regarding the impact of the number of training samples and model parameters on their testing error. To address this, we present a comprehensive study of the empirical scaling behaviour of ConvNets for EEG pathology classification. We analysed the testing error with increasing the training samples and model size for four different ConvNet architectures. The focus of our experiments is width scaling, and we have increased the number of parameters to up to 1.8 million. Our evaluation was based on two publicly available datasets: the Temple University Hospital (TUH) Abnormal EEG Corpus and the TUH Abnormal Expansion Balanced EEG Corpus, which together contain 10,707 training samples. The results show that the testing error follows a saturating power-law with both model and dataset size. This pattern is consistent across different datasets and ConvNet architectures. Furthermore, empirically observed accuracies saturate at 85%-87%, which may be due to an imperfect inter-rater agreement on the clinical labels. The empirical scaling behaviour of the test performance with dataset and model size has significant implications for deep EEG pathology classification research and practice. Our findings highlight the potential of deep ConvNets for high-performance EEG pathology classification, and the identified scaling relationships provide valuable recommendations for the advancement of automated EEG diagnostics.

Assessing expert reliability in determining intracranial EEG channel quality and introducing the automated bad channel detection algorithm

Objective.To evaluate the inter- and intra-rater reliability for the identification of bad channels among neurologists, EEG Technologists, and naïve research personnel, and to compare their performance with the automated bad channel detection (ABCD) algorithm for detecting bad channels.Approach.Six Neurologists, ten EEG Technologists, and six naïve research personnel (22 raters in total) were asked to rate 1440 real intracranial EEG channels as good or bad. Intra- and interrater kappa statistics were calculated for each group. We then compared each group to the ABCD algorithm which uses spectral and temporal domain features to classify channels as good or bad.Main results.Analysis of channel ratings from our participants revealed variable intra-rater reliability within each group, with no significant differences across groups. Inter-rater reliability was moderate among neurologists and EEG Technologists but minimal among naïve participants. Neurologists demonstrated a slightly higher consistency in ratings than EEG Technologists. Both groups occasionally misclassified flat channels as good, and participants generally focused on low-frequency content for their assessments. The ABCD algorithm, in contrast, relied more on high-frequency content. A logistic regression model showed a linear relationship between the algorithm's ratings and user responses for predominantly good channels, but less so for channels rated as bad. Sensitivity and specificity analyses further highlighted differences in rating patterns among the groups, with neurologists showing higher sensitivity and naïve personnel higher specificity.Significance.Our study reveals the bias in human assessments of intracranial electroencephalography (iEEG) data quality and the tendency of even experienced professionals to overlook certain bad channels, highlighting the need for standardized, unbiased methods. The ABCD algorithm, outperforming human raters, suggests the potential of automated solutions for more reliable iEEG interpretation and seizure characterization, offering a reliable approach free from human biases.

Artificial intelligence/machine learning for epilepsy and seizure diagnosis

Accurate seizure and epilepsy diagnosis remains a challenging task due to the complexity and variability of manifestations, which can lead to delayed or missed diagnosis. Machine learning (ML) and artificial intelligence (AI) is a rapidly developing field, with growing interest in integrating and applying these tools to aid clinicians facing diagnostic uncertainties. ML algorithms, particularly deep neural networks, are increasingly employed in interpreting electroencephalograms (EEG), neuroimaging, wearable data, and seizure videos. This review discusses the development and testing phases of AI/ML tools, emphasizing the importance of generalizability and interpretability in medical applications, and highlights recent publications that demonstrate the current and potential utility of AI to aid clinicians in diagnosing epilepsy. Current barriers of AI integration in patient care include dataset availability and heterogeneity, which limit studies' quality, interpretability, comparability, and generalizability. ML and AI offer substantial promise in improving the accuracy and efficiency of epilepsy diagnosis. The growing availability of diverse datasets, enhanced processing speed, and ongoing efforts to standardize reporting contribute to the evolving landscape of AI applications in clinical care.

An Amplitude-Integrated EEG Evaluation of Neonatal Opioid Withdrawal Syndrome

OBJECTIVE

Infants with neonatal opioid withdrawal syndrome (NOWS) have disrupted neurobehavior that requires hospitalization and treatment. This article aimed to evaluate electroencephalography (EEG) abnormalities using amplitude-integrated EEG (aEEG) in NOWS.

STUDY DESIGN

Eighteen term born infants with NOWS were recruited prospectively for an observational pilot study. aEEG monitoring was started within 24 hours of recruitment and twice weekly through discharge. aEEG data were analyzed for background and seizures. Severity of withdrawal was monitored using the modified Finnegan scoring (MFS) system.

RESULTS

Fifteen neonates had complete datasets. Thirteen (87%) had continuous aEEG background in all recordings. None had sleep-wake cyclicity (SWC) at initial recording. Brief seizures were noted in 9 of 15 (60%) infants. Lack of SWC was associated with higher MFS scores. At discharge, 8 of 15 (53%) had absent or emerging SWC.

CONCLUSION

aEEG abnormalities (absent SWC) are frequent and persist despite treatment at the time of discharge in the majority of patients with NOWS. Brief electrographic seizures are common. Neonates with persistent aEEG abnormalities at discharge warrant close follow-up.

KEY POINTS

· EEG abnormalities are common and persist after clinical signs resolve in patients with NOWS.. · Short subclinical seizures may be seen.. · aEEG may identify neonates who need follow-up..

EEG abnormalities are not associated with poor antidepressant treatment outcome - A NeuroPharm study

EEG brain abnormalities, such as slowing and isolated epileptiform discharges (IEDs), has previously been associated with non-response to antidepressant treatment with escitalopram and venlafaxine, suggesting a potential need for treatment with anticonvulsant property in some patients. The current study aims to replicate the reported association of EEG abnormality and treatment outcomes in an open-label trial of escitalopram for major depressive disorder (MDD) and explore its relationship to mood and cognition. Pretreatment, 6 min eyes-closed resting-state 256-channel EEG was recorded in 91 patients with MDD (age 18-57) who were treated with 10-20 mg escitalopram for 12 weeks; patients could switch to duloxetine after four weeks. A certified clinical neurophysiologist rated the EEGs. IED and EEG slowing was seen in 13.2%, and in 6.6% there were findings with unclear significance (i.e., Wicket spikes and theta activity). We saw no group-difference in remission or response rates after 8 and 12 weeks of treatment or switching to duloxetine. Patients with EEG abnormalities had higher pretreatment mood disturbances driven by greater anger (p=.039) and poorer verbal memory (p=.012). However, EEG abnormality was not associated with improved mood or verbal memory after treatment. Our findings should be interpreted in light of the rarity of EEG abnormalities and the sample size. While we cannot confirm that EEG abnormalities are associated with non-response to treatment, including escitalopram, abnormal EEG activity is associated with poor mood and verbal memory. The clinical utility of EEG abnormality in antidepressant treatment selection needs careful evaluation before deciding if useful for clinical implementation.

Subclinical epileptiform activity in the Alzheimer continuum: association with disease, cognition and detection method

BACKGROUND

Epileptic seizures are an established comorbidity of Alzheimer's disease (AD). Subclinical epileptiform activity (SEA) as detected by 24-h electroencephalography (EEG) or magneto-encephalography (MEG) has been reported in temporal regions of clinically diagnosed AD patients. Although epileptic activity in AD probably arises in the mesial temporal lobe, electrical activity within this region might not propagate to EEG scalp electrodes and could remain undetected by standard EEG. However, SEA might lead to faster cognitive decline in AD.

AIMS

1. To estimate the prevalence of SEA and interictal epileptic discharges (IEDs) in a well-defined cohort of participants belonging to the AD continuum, including preclinical AD subjects, as compared with cognitively healthy controls. 2. To evaluate whether long-term-EEG (LTM-EEG), high-density-EEG (hd-EEG) or MEG is superior to detect SEA in AD. 3. To characterise AD patients with SEA based on clinical, neuropsychological and neuroimaging parameters.

METHODS

Subjects (n = 49) belonging to the AD continuum were diagnosed according to the 2011 NIA-AA research criteria, with a high likelihood of underlying AD pathophysiology. Healthy volunteers (n = 24) scored normal on neuropsychological testing and were amyloid negative. None of the participants experienced a seizure before. Subjects underwent LTM-EEG and/or 50-min MEG and/or 50-min hd-EEG to detect IEDs.

RESULTS

We found an increased prevalence of SEA in AD subjects (31%) as compared to controls (8%) (p = 0.041; Fisher's exact test), with increasing prevalence over the disease course (50% in dementia, 27% in MCI and 25% in preclinical AD). Although MEG (25%) did not withhold a higher prevalence of SEA in AD as compared to LTM-EEG (19%) and hd-EEG (19%), MEG was significantly superior to detect spikes per 50 min (p = 0.002; Kruskall-Wallis test). AD patients with SEA scored worse on the RBANS visuospatial and attention subset (p = 0.009 and p = 0.05, respectively; Mann-Whitney U test) and had higher left frontal, (left) temporal and (left and right) entorhinal cortex volumes than those without.

CONCLUSION

We confirmed that SEA is increased in the AD continuum as compared to controls, with increasing prevalence with AD disease stage. In AD patients, SEA is associated with more severe visuospatial and attention deficits and with increased left frontal, (left) temporal and entorhinal cortex volumes.

TRIAL REGISTRATION

Clinicaltrials.gov, NCT04131491. 12/02/2020.

Ictal EEG Source Imaging

SUMMARY

Ictal EEG source imaging (ESI) is an advancing and growing application for presurgical epilepsy evaluation. For far too long, localization of seizures with scalp EEG has continued to rely on visual inspection of tracings arranged in a variety of montages allowing, at best, rough estimates of seizure onset regions. This most critical step is arguably the weakest point in epilepsy localization for surgical decision-making in clinical practice today. This review covers the methods and strategies that have been developed and tested for the performance of ictal ESI. It highlights practical issues and solutions toward sound implementation while covering differing methods to tackle the challenges specific to ictal ESI-noise and artifact reduction, component analysis, and other tools to increase seizure-specific signal for analysis. Further, validation studies to date-those with both high and low density numbers of electrodes-are summarized, providing a glimpse at the relative accuracy of ictal ESI in all types of focal epilepsy patients. Finally, given the added noninvasive information (greater degree of spatial resolution compared with standard ictal EEG review), the role of ictal ESI and its clinical utility in the presurgical evaluation is discussed.

Pitfalls in scalp EEG: Current obstacles and future directions

Although electroencephalography (EEG) serves a critical role in the evaluation and management of seizure disorders, it is commonly misinterpreted, resulting in avoidable medical, social, and financial burdens to patients and health care systems. Overinterpretation of sharply contoured transient waveforms as being representative of interictal epileptiform abnormalities lies at the core of this problem. However, the magnitude of these errors is amplified by the high prevalence of paroxysmal events exhibited in clinical practice that compel investigation with EEG. Neurology training programs, which vary considerably both in the degree of exposure to EEG and the composition of EEG didactics, have not effectively addressed this widespread issue. Implementation of competency-based curricula in lieu of traditional educational approaches may enhance proficiency in EEG interpretation amongst general neurologists in the absence of formal subspecialty training. Efforts in this regard have led to the development of a systematic, high-fidelity approach to the interpretation of epileptiform discharges that is readily employable across medical centers. Additionally, machine learning techniques hold promise for accelerating accurate and reliable EEG interpretation, particularly in settings where subspecialty interpretive EEG services are not readily available. This review highlights common diagnostic errors in EEG interpretation, limitations in current educational paradigms, and initiatives aimed at resolving these challenges.

Estimating Patient-Level Uncertainty in Seizure Detection Using Group-Specific Out-of-Distribution Detection Technique

Epilepsy is a chronic neurological disorder affecting around 1% of the global population, characterized by recurrent epileptic seizures. Accurate diagnosis and treatment are crucial for reducing mortality rates. Recent advancements in machine learning (ML) algorithms have shown potential in aiding clinicians with seizure detection in electroencephalography (EEG) data. However, these algorithms face significant challenges due to the patient-specific variability in seizure patterns and the limited availability of high-quality EEG data for training, causing erratic predictions. These erratic predictions are harmful, especially for high-stake domains in healthcare, negatively affecting patients. Therefore, ensuring safety in AI is of the utmost importance. In this study, we propose a novel ensemble method for uncertainty quantification to identify patients with low-confidence predictions in ML-based seizure detection algorithms. Our approach aims to mitigate high-risk predictions in previously unseen seizure patients, thereby enhancing the robustness of existing seizure detection algorithms. Additionally, our method can be implemented with most of the deep learning (DL) models. We evaluated the proposed method against established uncertainty detection techniques, demonstrating its effectiveness in identifying patients for whom the model's predictions are less certain. Our proposed method managed to achieve 87%, 89% and 75% in accuracy, specificity and sensitivity, respectively. This study represents a novel attempt to improve the reliability and robustness of DL algorithms in the domain of seizure detection. This study underscores the value of integrating uncertainty quantification into ML algorithms for seizure detection, offering clinicians a practical tool to gauge the applicability of ML models for individual patients.

Acceptability by End-users of a Standardized Structured Format for Reporting EEG

The report of the electroencephalogram (EEG) results has traditionally been made using free-text formats with a huge variation in descriptions due to several factors. Recently, the International Federation of Clinical Neurophysiology (IFCN) endorsed the use of the Standardized Computer-based Organized Reporting of EEG (SCORE). This system has many advantages, but only some concerns have been investigated so far. This study aimed to assess the end-users acceptability of this proposed EEG report format. A 16-item electronic survey was sent to physicians who use EEG services of a medical diagnosis clinic. Physicians had been receiving the EEG reports in free-text formats from the same three board-certified electroencephalographers for the past three years. In January 2019, the report changed to the SCORE format. The survey assessed five main topics: physician information and historical use of EEG; personal preferences; comparative aspects of the formats; impact of the new format on clinical decision-making; and satisfaction. Thirty-two of 52 have responded to the survey (61%). On average, 81% of the responders have received enough reports with the new format to reliably complete the survey. Every responder prefers the standardized compared to the free-text format. Twenty-five responders like the inclusion of the head model, and interestingly, five suggest including another legend to differentiate "slow activity" from "other abnormal activity". Virtually all responders would recommend the new format, but one-third read only the conclusion. Our findings suggest high acceptability of this standardized report format. Despite the limitations of this study, we hope these findings contribute to the improvement and expansion of standardized EEG reporting systems.

Developing a Standardized Approach to Grading the Level of Brain Dysfunction on EEG

PURPOSE

To assess variability in interpretation of electroencephalogram (EEG) background activity and qualitative grading of cerebral dysfunction based on EEG findings, including which EEG features are deemed most important in this determination.

METHODS

A web-based survey (Qualtrics) was disseminated to electroencephalographers practicing in institutions participating in the Critical Care EEG Monitoring Research Consortium between May 2017 and August 2018. Respondents answered 12 questions pertaining to their training and EEG interpretation practices and graded 40 EEG segments (15-second epochs depicting patients' most stimulated state) using a 6-grade scale. Fleiss' Kappa statistic evaluated interrater agreement.

RESULTS

Of 110 respondents, 78.2% were attending electroencephalographers with a mean of 8.3 years of experience beyond training. Despite 83% supporting the need for a standardized approach to interpreting the degree of dysfunction on EEG, only 13.6% used a previously published or an institutional grading scale. The overall interrater agreement was fair ( k = 0.35). Having Critical Care EEG Monitoring Research Consortium nomenclature certification (40.9%) or EEG board certification (70%) did not improve interrater agreement ( k = 0.26). Predominant awake frequencies and posterior dominant rhythm were ranked as the most important variables in grading background dysfunction, followed by continuity and reactivity.

CONCLUSIONS

Despite the preference for a standardized grading scale for background EEG interpretation, the lack of interrater agreement on levels of dysfunction even among experienced academic electroencephalographers unveils a barrier to the widespread use of EEG as a clinical and research neuromonitoring tool. There was reasonable agreement on the features that are most important in this determination. A standardized approach to grading cerebral dysfunction, currently used by the authors, and based on this work, is proposed.

Enhancing Performance of Convolutional Neural Network-Based Epileptic Electroencephalogram Diagnosis by Asymmetric Stochastic Resonance

Epilepsy is a chronic disorder that leads to transient neurological dysfunction and is clinically diagnosed primarily by electroencephalography. Several intelligent systems have been proposed to automatically detect seizures, among which deep convolutional neural networks (CNNs) have shown better performance than traditional machine-learning algorithms. Owing to artifacts and noise, the raw electroencephalogram (EEG) must be preprocessed to improve the signal-to-noise ratio prior to being fed into the CNN classifier. However, because of the spectrum overlapping of uncontrollable noise with EEG, traditional filters cause information loss in EEG; thus, the potential of classifiers cannot be fully exploited. In this study, we propose a stochastic resonance-effect-based EEG preprocessing module composed of three asymmetrical overdamped bistable systems in parallel. By setting different asymmetries for the three parallel units, the inherent noise can be transferred to the different spectral components of the EEG through the asymmetric stochastic resonance effect. In this process, the proposed preprocessing module not only avoids the loss of information of EEG but also provides a CNN with high-quality EEG of diversified frequency information to enhance its performance. By combining the proposed preprocessing module with a residual neural network, we developed an intelligent diagnostic system for predicting seizure onset. The developed system achieved an average sensitivity of 98.96% on the CHB-MIT dataset and 95.45% on the Siena dataset, with a false prediction rate of 0.048/h and 0.033/h, respectively. In addition, a comparative analysis demonstrated the superiority of the developed diagnostic system with the proposed preprocessing module over other existing methods.

Predicting seizure outcome after epilepsy surgery: Do we need more complex models, larger samples, or better data?

OBJECTIVE

The accurate prediction of seizure freedom after epilepsy surgery remains challenging. We investigated if (1) training more complex models, (2) recruiting larger sample sizes, or (3) using data-driven selection of clinical predictors would improve our ability to predict postoperative seizure outcome using clinical features. We also conducted the first substantial external validation of a machine learning model trained to predict postoperative seizure outcome.

METHODS

We performed a retrospective cohort study of 797 children who had undergone resective or disconnective epilepsy surgery at a tertiary center. We extracted patient information from medical records and trained three models-a logistic regression, a multilayer perceptron, and an XGBoost model-to predict 1-year postoperative seizure outcome on our data set. We evaluated the performance of a recently published XGBoost model on the same patients. We further investigated the impact of sample size on model performance, using learning curve analysis to estimate performance at samples up to N = 2000. Finally, we examined the impact of predictor selection on model performance.

RESULTS

Our logistic regression achieved an accuracy of 72% (95% confidence interval [CI] = 68%-75%, area under the curve [AUC] = .72), whereas our multilayer perceptron and XGBoost both achieved accuracies of 71% (95% CIMLP  = 67%-74%, AUCMLP  = .70; 95% CIXGBoost own  = 68%-75%, AUCXGBoost own  = .70). There was no significant difference in performance between our three models (all p > .4) and they all performed better than the external XGBoost, which achieved an accuracy of 63% (95% CI = 59%-67%, AUC = .62; pLR  = .005, pMLP  = .01, pXGBoost own  = .01) on our data. All models showed improved performance with increasing sample size, but limited improvements beyond our current sample. The best model performance was achieved with data-driven feature selection.

SIGNIFICANCE

We show that neither the deployment of complex machine learning models nor the assembly of thousands of patients alone is likely to generate significant improvements in our ability to predict postoperative seizure freedom. We instead propose that improved feature selection alongside collaboration, data standardization, and model sharing is required to advance the field.

Utilization of temporal autoencoder for semi-supervised intracranial EEG clustering and classification

Manual visual review, annotation and categorization of electroencephalography (EEG) is a time-consuming task that is often associated with human bias and requires trained electrophysiology experts with specific domain knowledge. This challenge is now compounded by development of measurement technologies and devices allowing large-scale heterogeneous, multi-channel recordings spanning multiple brain regions over days, weeks. Currently, supervised deep-learning techniques were shown to be an effective tool for analyzing big data sets, including EEG. However, the most significant caveat in training the supervised deep-learning models in a clinical research setting is the lack of adequate gold-standard annotations created by electrophysiology experts. Here, we propose a semi-supervised machine learning technique that utilizes deep-learning methods with a minimal amount of gold-standard labels. The method utilizes a temporal autoencoder for dimensionality reduction and a small number of the expert-provided gold-standard labels used for kernel density estimating (KDE) maps. We used data from electrophysiological intracranial EEG (iEEG) recordings acquired in two hospitals with different recording systems across 39 patients to validate the method. The method achieved iEEG classification (Pathologic vs. Normal vs. Artifacts) results with an area under the receiver operating characteristic (AUROC) scores of 0.862 ± 0.037, 0.879 ± 0.042, and area under the precision-recall curve (AUPRC) scores of 0.740 ± 0.740, 0.714 ± 0.042. This demonstrates that semi-supervised methods can provide acceptable results while requiring only 100 gold-standard data samples in each classification category. Subsequently, we deployed the technique to 12 novel patients in a pseudo-prospective framework for detecting Interictal epileptiform discharges (IEDs). We show that the proposed temporal autoencoder was able to generalize to novel patients while achieving AUROC of 0.877 ± 0.067 and AUPRC of 0.705 ± 0.154.

Reliability of visual review of intracranial electroencephalogram in identifying the seizure onset zone: A systematic review and implications for the accuracy of automated methods

Visual review of intracranial electroencephalography (iEEG) is often an essential component for defining the zone of resection for epilepsy surgery. Unsupervised approaches using machine and deep learning are being employed to identify seizure onset zones (SOZs). This prompts a more comprehensive understanding of the reliability of visual review as a reference standard. We sought to summarize existing evidence on the reliability of visual review of iEEG in defining the SOZ for patients undergoing surgical workup and understand its implications for algorithm accuracy for SOZ prediction. We performed a systematic literature review on the reliability of determining the SOZ by visual inspection of iEEG in accordance with best practices. Searches included MEDLINE, Embase, Cochrane Library, and Web of Science on May 8, 2022. We included studies with a quantitative reliability assessment within or between observers. Risk of bias assessment was performed with QUADAS-2. A model was developed to estimate the effect of Cohen kappa on the maximum possible accuracy for any algorithm detecting the SOZ. Two thousand three hundred thirty-eight articles were identified and evaluated, of which one met inclusion criteria. This study assessed reliability between two reviewers for 10 patients with temporal lobe epilepsy and found a kappa of .80. These limited data were used to model the maximum accuracy of automated methods. For a hypothetical algorithm that is 100% accurate to the ground truth, the maximum accuracy modeled with a Cohen kappa of .8 ranged from .60 to .85 (F-2). The reliability of reviewing iEEG to localize the SOZ has been evaluated only in a small sample of patients with methodologic limitations. The ability of any algorithm to estimate the SOZ is notably limited by the reliability of iEEG interpretation. We acknowledge practical limitations of rigorous reliability analysis, and we propose design characteristics and study questions to further investigate reliability.

Quantitative artifact reduction and pharmacologic paralysis improve detection of EEG epileptiform activity in critically ill patients

OBJECTIVE

Epileptiform activity is common in critically ill patients, but movement-related artifacts-including electromyography (EMG) and myoclonus-can obscure EEG, limiting detection of epileptiform activity. We sought to determine the ability of pharmacologic paralysis and quantitative artifact reduction (AR) to improve epileptiform discharge detection.

METHODS

Retrospective analysis of patients who underwent continuous EEG monitoring with pharmacologic paralysis. Four reviewers read each patient's EEG pre- and post- both paralysis and AR, and indicated the presence of epileptiform discharges. We compared the interrater reliability (IRR) of identifying discharges at baseline, post-AR, and post-paralysis, and compared the performance of AR and paralysis according to artifact type.

RESULTS

IRR of identifying epileptiform discharges at baseline was slight (N = 30; κ = 0.10) with a trend toward increase post-AR (κ = 0.26, p = 0.053) and a significant increase post-paralysis (κ = 0.51, p = 0.001). AR was as effective as paralysis at improving IRR of identifying discharges in those with high EMG artifact (N = 15; post-AR κ = 0.63, p = 0.009; post-paralysis κ = 0.62, p = 0.006) but not with primarily myoclonus artifact (N = 15).

CONCLUSIONS

Paralysis improves detection of epileptiform activity in critically ill patients when movement-related artifact obscures EEG features. AR improves detection as much as paralysis when EMG artifact is high, but is ineffective when the primary source of artifact is myoclonus.

SIGNIFICANCE

In the appropriate setting, both AR and paralysis facilitate identification of epileptiform activity in critically ill patients.

EEG-based methods for recovery prognosis of patients with disorders of consciousness: A systematic review

OBJECTIVE

Disorders of consciousness (DoC) are acquired conditions of severely altered consciousness. Electroencephalography (EEG)-derived biomarkers have been studied as clinical predictors of consciousness recovery. Therefore, this study aimed to systematically review the methods, features, and models used to derive prognostic EEG markers in patients with DoC in a rehabilitation setting.

METHODS

We conducted a systematic literature search of EEG-based strategies for consciousness recovery prognosis in five electronic databases.

RESULTS

The search resulted in 2964 papers. After screening, 15 studies were included in the review. Our analyses revealed that simpler experimental settings and similar filtering cut-off frequencies are preferred. The results of studies were categorised by extracting qualitative and quantitative features. The quantitative features were further classified into evoked/event-related potentials, spectral measures, entropy measures, and graph-theory measures. Despite the variety of methods, features from all categories, including qualitative ones, exhibited significant correlations with DoC prognosis. Moreover, no agreement was found on the optimal set of EEG-based features for the multivariate prognosis of patients with DoC, which limits the computational methods applied for outcome prediction and correlation analysis to classical ones. Nevertheless, alpha power, reactivity, and higher complexity metrics were often found to be predictive of consciousness recovery.

CONCLUSIONS

This study's findings confirm the essential role of qualitative EEG and suggest an important role for quantitative EEG. Their joint use could compensate for their reciprocal limitations.

SIGNIFICANCE

This study emphasises the need for further efforts toward guidelines on standardised EEG analysis pipeline, given the already proven role of EEG markers in the recovery prognosis of patients with DoC.

EEG-Update

Durch die rasante Entwicklung digitaler Computertechniken und neuer Analysemethoden hat sich ein neuer Ansatz zur Analyse der Hirnströme (quantitatives EEG) ergeben, die in verschiedenen klinischen Bereichen der Neurologie und Psychiatrie bereits Ergebnisse zeigen. Die neuen Möglichkeiten der Analyse des EEG durch Einsatz künstlicher Intelligenz (Deep Learning) und großer Datenmengen (Big Data) sowie telemedizinischer Datenübermittlung und Interaktion wird den Einsatz der Methode vermutlich in den nächsten Jahren erweitern.

Electric Source Imaging in Presurgical Evaluation of Epilepsy: An Inter-Analyser Agreement Study

Electric source imaging (ESI) estimates the cortical generator of the electroencephalography (EEG) signals recorded with scalp electrodes. ESI has gained increasing interest for the presurgical evaluation of patients with drug-resistant focal epilepsy. In spite of a standardised analysis pipeline, several aspects tailored to the individual patient involve subjective decisions of the expert performing the analysis, such as the selection of the analysed signals (interictal epileptiform discharges and seizures, identification of the onset epoch and time-point of the analysis). Our goal was to investigate the inter-analyser agreement of ESI in presurgical evaluations of epilepsy, using the same software and analysis pipeline. Six experts, of whom five had no previous experience in ESI, independently performed interictal and ictal ESI of 25 consecutive patients (17 temporal, 8 extratemporal) who underwent presurgical evaluation. The overall agreement among experts for the ESI methods was substantial (AC1 = 0.65; 95% CI: 0.59–0.71), and there was no significant difference between the methods. Our results suggest that using a standardised analysis pipeline, newly trained experts reach similar ESI solutions, calling for more standardisation in this emerging clinical application in neuroimaging.

Electroencephalographic Seizures in Emergency Department Patients After Treatment for Convulsive Status Epilepticus

PURPOSE

It is unknown how often and how early EEG is obtained in patients presenting with status epilepticus. The Established Status Epilepticus Treatment Trial enrolled patients with benzodiazepine-refractory seizures and randomized participants to fosphenytoin, levetiracetam, or valproate. The use of early EEG, including frequency of electrographic seizures, was determined in Established Status Epilepticus Treatment Trial participants.

METHODS

Secondary analysis of 475 enrollments at 58 hospitals to determine the frequency of EEG performed within 24 hours of presentation. The EEG type, the prevalence of electrographic seizures, and characteristics associated with obtaining early EEG were recorded. Chi-square and Wilcoxon rank-sum tests were calculated as appropriate for univariate and bivariate comparisons. Odds ratios are reported with 95% confidence intervals.

RESULTS

A total of 278 of 475 patients (58%) in the Established Status Epilepticus Treatment Trial cohort underwent EEG within 24 hours (median time to EEG: 5 hours [interquartile range: 3-10]). Electrographic seizure prevalence was 14% (95% confidence interval, 10%-19%; 39/278) in the entire cohort and 13% (95% confidence interval, 7%-21%) in the subgroup of patients meeting the primary outcome of the Established Status Epilepticus Treatment Trial (clinical treatment success within 60 minutes of randomization). Among subjects diagnosed with electrographic seizures (39), 15 (38%; 95% confidence interval, 25%-54%) had no clinical correlate on the video EEG recording.

CONCLUSIONS

Electrographic seizures may occur in patients who stop seizing clinically after treatment of convulsive status epilepticus. Clinical correlates might not be present during electrographic seizures. These findings support early initiation of EEG recordings in patients suffering from convulsive status epilepticus, including those with clinical evidence of treatment success.

Alpha-power in electroencephalography as good outcome predictor for out-of-hospital cardiac arrest survivors

This study aimed to investigate the utility of quantitative EEG biomarkers for predicting good neurologic outcomes in OHCA survivors treated with targeted temperature management (TTM) using power spectral density (PSD), event-related spectral perturbation (ERSP), and spectral entropy (SE). This observational registry-based study was conducted at a tertiary care hospital in Korea using data of adult nontraumatic comatose OHCA survivors who underwent standard EEG and treated with TTM between 2010 and 2018. Good neurological outcome at 1 month (Cerebral Performance Category scores 1 and 2) was the primary outcome. The linear mixed model analysis was performed for PSD, ESRP, and SE values of all and each frequency band. Thirteen of the 54 comatose OHCA survivors with TTM and EEG were excluded due to poor EEG quality or periodic/rhythmic pattern, and EEG data of 41 patients were used for analysis. The median time to EEG was 21 h, and the rate of the good neurologic outcome at 1 month was 52.5%. The good neurologic outcome group was significantly younger and showed higher PSD and ERSP and lower SE features for each frequency than the poor outcome group. After age adjustment, only the alpha-PSD was significantly higher in the good neurologic outcome group (1.13 ± 1.11 vs. 0.09 ± 0.09, p = 0.031) and had best performance with 0.903 of the area under the curve for predicting good neurologic outcome. Alpha-PSD best predicts good neurologic outcome in OHCA survivors and is an early biomarker for prognostication. Larger studies are needed to conclusively confirm these findings.

Improved access to rapid electroencephalography at a community hospital reduces inter-hospital transfers for suspected non-convulsive seizures

OBJECTIVE

Patients with suspected non-convulsive seizures are optimally evaluated with EEG. However, limited EEG infrastructure at community hospitals often necessitates transfer for long-term EEG monitoring (LTM). Novel point-of-care EEG systems could expedite management of nonconvulsive seizures and reduce unnecessary transfers. We aimed to describe the impact of rapid access to EEG using a novel EEG device with remote expert interpretation (tele-EEG) on rates of transfer for LTM.

METHODS

We retrospectively identified a cohort of patients who underwent Rapid-EEG (Ceribell Inc., Mountain View, CA) monitoring as part of a new standard-of-care at a community hospital. Rapid-EEGs were initially reviewed on-site by a community hospital neurologist before transitioning to tele-EEG review by epileptologists at an affiliated academic hospital. We compared the rate of transfer for LTM after Rapid-EEG/tele-EEG implementation to the expected rate if rapid access to EEG was unavailable.

RESULTS

Seventy-four patients underwent a total of 118 Rapid-EEG studies (10 with seizure, 18 with highly epileptiform patterns, 90 with slow/normal activity). Eighty-one studies (69%), including 9 of 10 studies that detected seizures, occurred after-hours when EEG was previously unavailable. Based on historical practice patterns, we estimated that Rapid-EEG potentially obviated transfer for LTM in 31 of 33 patients (94%); both completed transfers occurred before the transition to tele-EEG review.

SIGNIFICANCE

Rapid access to EEG led to the detection of seizures that would otherwise have been missed and reduced inter-hospital transfers for LTM. We estimate that the reduction in inter-hospital transportation costs alone would be in excess of $39,000 ($1,274 per patient). Point-of-care EEG systems may support a hub-and-spoke model for managing non-convulsive seizures (similar to that utilized in this study and analogous to existing acute stroke infrastructures), with increased EEG capacity at community hospitals and tele-EEG interpretation by specialists at academic hospitals that can accept transfers for LTM.

Measuring expertise in identifying interictal epileptiform discharges

OBJECTIVE

Interictal epileptiform discharges on EEG are integral to diagnosing epilepsy. However, EEGs are interpreted by readers with and without specialty training, and there is no accepted method to assess skill in interpretation. We aimed to develop a test to quantify IED recognition skills.

METHODS

A total of 13,262 candidate IEDs were selected from EEGs and scored by eight fellowship-trained reviewers to establish a gold standard. An online test was developed to assess how well readers with different training levels could distinguish candidate waveforms. Sensitivity, false positive rate and calibration were calculated for each reader. A simple mathematical model was developed to estimate each reader's skill and threshold in identifying an IED, and to develop receiver operating characteristics curves for each reader. We investigated the number of IEDs needed to measure skill level with acceptable precision.

RESULTS

Twenty-nine raters completed the test; nine experts, seven experienced non-experts and thirteen novices. Median calibration errors for experts, experienced non-experts and novices were -0.056, 0.012, 0.046; median sensitivities were 0.800, 0.811, 0.715; and median false positive rates were 0.177, 0.272, 0.396, respectively. The number of test questions needed to measure those scores was 549. Our analysis identified that novices had a higher noise level (uncertainty) compared to experienced non-experts and experts. Using calculated noise and threshold levels, receiver operating curves were created, showing increasing median area under the curve from novices (0.735), to experienced non-experts (0.852) and experts (0.891).

SIGNIFICANCE

Expert and non-expert readers can be distinguished based on ability to identify IEDs. This type of assessment could also be used to identify and correct differences in thresholds in identifying IEDs.

Automated spike detection: Which software package?

PURPOSE

We assessed three commercial automated spike detection software packages (Persyst, Encevis and BESA) to see which had the best performance.

METHODS

Thirty prolonged EEG records from people aged at least 16 years were collected and 30-minute representative epochs were selected. Interictal epileptiform discharges (IEDs) were marked by three human experts and by all three software packages. For each 30-minutes selection and for each 10-second epoch we measured whether or not IEDs had occurred. We defined the gold standard as the combined detections of the experts. Kappa scores, sensitivity and specificity were estimated for each software package.

RESULTS

Sensitivity for Persyst in the default setting was 95% for 30-minute selections and 82% for 10-second epochs. Sensitivity for Encevis was 86% (30-minute selections) and 61% (10-second epochs). The specificity for both packages was 88% for 30-minute selections and 96%-99% for the 10-second epochs. Interrater agreement between Persyst and Encevis and the experts was similar than between experts (0.67-0.83 versus 0.63-0.67). Sensitivity for BESA was 40% and specificity 100%. Interrater agreement (0.25) was low.

CONCLUSIONS

IED detection by the Persyst automated software is better than the Encevis and BESA packages, and similar to human review, when reviewing 30-minute selections and 10-second epochs. This findings may help prospective users choose a software package.

Pitfalls in EEG Analysis in Patients With Nonconvulsive Status Epilepticus: A Preliminary Study

Objective: Electroencephalography (EEG) interpretations through visual (by human raters) and automated (by computer technology) analysis were still not reliable for the diagnosis of nonconvulsive status epilepticus (NCSE). This study aimed to identify typical pitfalls in the EEG analysis and make suggestions as to how those pitfalls might be avoided. Methods: We analyzed the EEG recordings of individuals who had clinically confirmed or suspected NCSE. Epileptiform EEG activity during seizures (ictal discharges) was visually analyzed by 2 independent raters. We investigated whether unreliable EEG visual interpretations quantified by low interrater agreement can be predicted by the characteristics of ictal discharges and individuals' clinical data. In addition, the EEG recordings were automatically analyzed by in-house algorithms. To further explore the causes of unreliable EEG interpretations, 2 epileptologists analyzed EEG patterns most likely misinterpreted as ictal discharges based on the differences between the EEG interpretations through the visual and automated analysis. Results: Short ictal discharges with a gradual onset (developing over 3 s in length) were liable to be misinterpreted. An extra 2 min of ictal discharges contributed to an increase in the kappa statistics of >0.1. Other problems were the misinterpretation of abnormal background activity (slow-wave activities, other abnormal brain activity, and the ictal-like movement artifacts), continuous interictal discharges, and continuous short ictal discharges. Conclusion: A longer duration criterion for NCSE-EEGs than 10 s that is commonly used in NCSE working criteria is recommended. Using knowledge of historical EEGs, individualized algorithms, and context-dependent alarm thresholds may also avoid the pitfalls.

AiED: Artificial intelligence for the detection of intracranial interictal epileptiform discharges

OBJECTIVE

Deep learning provides an appealing solution for the ongoing challenge of automatically classifying intracranial interictal epileptiform discharges (IEDs). We report results from an automated method consisting of a template-matching algorithm and convolutional neural network (CNN) for the detection of intracranial IEDs ("AiED").

METHODS

1000 intracranial electroencephalogram (EEG) epochs extracted randomly from 307 subjects with refractory epilepsy were annotated independently by two expert neurophysiologists. These annotated epochs were divided into 1062 two-second epochs with IEDs and 1428 two-second epochs without IEDs, which were transformed into spectrograms prior to training the neural network. The highest performing network was validated on an annotated external test set.

RESULTS

The final network had an F1-score of 0.95 (95% CI: 0.91-0.98) and an average Area Under the Receiver Operating Characteristic of 0.98 (95% CI: 0.96-1.00). For the external test set, it showed an overall F1-score of 0.71, correctly identifying 100% of all high-amplitude IED complexes, 96.23% of all high-amplitude isolated IEDs, and 66.15% of all IEDs of atypical morphology.

CONCLUSIONS

Template-matching combined with a CNN offers a fast, robust method for detecting intracranial IEDs.

SIGNIFICANCE

"AiED" is generalizable and achieves comparable performance to human reviewers; it may support clinical and research EEG analyses.

The Levels of Circulating MicroRNAs at 6-Hour Cardiac Arrest Can Predict 6-Month Poor Neurological Outcome

Early prognostication in cardiac arrest survivors is challenging for physicians. Unlike other prognostic modalities, biomarkers are easily accessible and provide an objective assessment method. We hypothesized that in cardiac arrest patients with targeted temperature management (TTM), early circulating microRNA (miRNA) levels are associated with the 6-month neurological outcome. In the discovery phase, we identified candidate miRNAs associated with cardiac arrest patients who underwent TTM by comparing circulating expression levels in patients and healthy controls. Next, using a larger cohort, we validated the prognostic values of the identified early miRNAs by measuring the serum levels of miRNAs, neuron-specific enolase (NSE), and S100 calcium-binding protein B (S100B) 6 h after cardiac arrest. The validation cohort consisted of 54 patients with TTM. The areas under the curve (AUCs) for poor outcome were 0.85 (95% CI (confidence interval), 0.72–0.93), 0.82 (95% CI, 0.70–0.91), 0.78 (95% CI, 0.64–0.88), and 0.77 (95% CI, 0.63–0.87) for miR-6511b-5p, -125b-1-3p, -122-5p, and -124-3p, respectively. When the cut-off was based on miRNA levels predicting poor outcome with 100% specificity, sensitivities were 67.7% (95% CI, 49.5–82.6), 50.0% (95% CI, 32.4–67.7), 35.3% (95% CI, 19.7–53.5), and 26.5% (95% CI, 12.9–44.4) for the above miRNAs, respectively. The models combining early miRNAs with protein biomarkers demonstrated superior prognostic performance to those of protein biomarkers.

HFOApp: A MATLAB Graphical User Interface for High-Frequency Oscillation Marking

Epilepsy affects 3.4 million people in the United States, and, despite the availability of numerous antiepileptic drugs, 36% of patients have uncontrollable seizures, which severely impact quality of life. High-frequency oscillations (HFOs) are a potential biomarker of epileptogenic tissue that could be useful in surgical planning. As a result, research into the efficacy of HFOs as a clinical tool has increased over the last 2 decades. However, detection and identification of these transient rhythms in intracranial electroencephalographic recordings remain time-consuming and challenging. Although automated detection algorithms have been developed, their results are widely inconsistent, reducing reliability. Thus, manual marking of HFOs remains the gold standard, and manual review of automated results is required. However, manual marking and review are time consuming and can still produce variable results because of their subjective nature and the limitations in functionality of existing open-source software. Our goal was to develop a new software with broad application that improves on existing open-source HFO detection applications in usability, speed, and accuracy. Here, we present HFOApp: a free, open-source, easy-to-use MATLAB-based graphical user interface for HFO marking. This toolbox offers a high degree of intuitive and ergonomic usability and integrates interactive automation-assist options with manual marking, significantly reducing the time needed for review and manual marking of recordings, while increasing inter-rater reliability. The toolbox also features simultaneous multichannel detection and marking. HFOApp was designed as an easy-to-use toolbox for clinicians and researchers to quickly and accurately mark, quantify, and characterize HFOs within electrophysiological datasets.

To spindle or not to spindle: A replication study into spindling excessive beta as a transdiagnostic EEG feature associated with impulse control

BACKGROUND

Frontocentral Spindling Excessive Beta (SEB), a spindle-like beta-activity observed in the electroencephalogram (EEG), has been transdiagnostically associated with more problems with impulse control and sleep maintenance. The current study aims to replicate and elaborate on these findings.

METHODS

Participants reporting sleep problems (n = 31) or Attention-Deficit/Hyperactivity Disorder (ADHD) symptoms (n = 48) were included. Baseline ADHD-Rating Scale (ADHD-RS), Pittsburgh Sleep Quality Index (PSQI), Holland Sleep Disorder Questionnaire (HSDQ), and EEG were assessed. Analyses were confined to adults with frontocentral SEB.

RESULTS

Main effects of SEB showed more impulse control problems (d = 0.87) and false positive errors (d = 0.55) in participants with SEB. No significant associations with sleep or interactions with Sample were observed.

DISCUSSION

This study partially replicates an earlier study and demonstrates that participants exhibiting SEB report more impulse control problems, independent of diagnosis. Future studies should focus on automating SEB classification and further investigate the transdiagnostic nature of SEB.

Seizure Detection: Interreader Agreement and Detection Algorithm Assessments Using a Large Dataset

PURPOSE

To compare the seizure detection performance of three expert humans and two computer algorithms in a large set of epilepsy monitoring unit EEG recordings.

METHODS

One hundred twenty prolonged EEGs, 100 containing clinically reported EEG-evident seizures, were evaluated. Seizures were marked by the experts and algorithms. Pairwise sensitivity and false-positive rates were calculated for each human-human and algorithm-human pair. Differences in human pairwise performance were calculated and compared with the range of algorithm versus human performance differences as a type of statistical modified Turing test.

RESULTS

A total of 411 individual seizure events were marked by the experts in 2,805 hours of EEG. Mean, pairwise human sensitivities and false-positive rates were 84.9%, 73.7%, and 72.5%, and 1.0, 0.4, and 1.0/day, respectively. Only the Persyst 14 algorithm was comparable with humans-78.2% and 1.0/day. Evaluation of pairwise differences in sensitivity and false-positive rate demonstrated that Persyst 14 met statistical noninferiority criteria compared with the expert humans.

CONCLUSIONS

Evaluating typical prolonged EEG recordings, human experts had a modest level of agreement in seizure marking and low false-positive rates. The Persyst 14 algorithm was statistically noninferior to the humans. For the first time, a seizure detection algorithm and human experts performed similarly.

Chasing Certainty After Cardiac Arrest: Can a Technological Innovation Solve a Moral Dilemma?

When information on a coma patient’s expected outcome is uncertain, a moral dilemma arises in clinical practice: if life-sustaining treatment is continued, the patient may survive with unacceptably poor neurological prospects, but if withdrawn a patient who could have recovered may die. Continuous electroencephalogram-monitoring (cEEG) is expected to substantially improve neuroprognostication for patients in coma after cardiac arrest. This raises expectations that decisions whether or not to withdraw will become easier. This paper investigates that expectation, exploring cEEG’s impacts when it becomes part of a socio-technical network in an Intensive Care Unit (ICU). Based on observations in two ICUs in the Netherlands and one in the USA that had cEEG implemented for research, we interviewed 25 family members, healthcare professionals, and surviving patients. The analysis focuses on (a) the way patient outcomes are constructed, (b) the kind of decision support these outcomes provide, and (c) how cEEG affects communication between professionals and relatives. We argue that cEEG can take away or decrease the intensity of the dilemma in some cases, while increasing uncertainty for others. It also raises new concerns. Since its actual impacts furthermore hinge on how cEEG is designed and implemented, we end with recommendations for ensuring responsible development and implementation.

Time-Frequency Decomposition of Scalp Electroencephalograms Improves Deep Learning-Based Epilepsy Diagnosis

Epilepsy diagnosis based on Interictal Epileptiform Discharges (IEDs) in scalp electroencephalograms (EEGs) is laborious and often subjective. Therefore, it is necessary to build an effective IED detector and an automatic method to classify IED-free versus IED EEGs. In this study, we evaluate features that may provide reliable IED detection and EEG classification. Specifically, we investigate the IED detector based on convolutional neural network (ConvNet) with different input features (temporal, spectral, and wavelet features). We explore different ConvNet architectures and types, including 1D (one-dimensional) ConvNet, 2D (two-dimensional) ConvNet, and noise injection at various layers. We evaluate the EEG classification performance on five independent datasets. The 1D ConvNet with preprocessed full-frequency EEG signal and frequency bands (delta, theta, alpha, beta) with Gaussian additive noise at the output layer achieved the best IED detection results with a false detection rate of 0.23/min at 90% sensitivity. The EEG classification system obtained a mean EEG classification Leave-One-Institution-Out (LOIO) cross-validation (CV) balanced accuracy (BAC) of 78.1% (area under the curve (AUC) of 0.839) and Leave-One-Subject-Out (LOSO) CV BAC of 79.5% (AUC of 0.856). Since the proposed classification system only takes a few seconds to analyze a 30-min routine EEG, it may help in reducing the human effort required for epilepsy diagnosis.

EEG microstate features for schizophrenia classification

Electroencephalography (EEG) microstate analysis is a method wherein spontaneous EEG activity is segmented at sub-second levels to analyze quasi-stable states. In particular, four archetype microstates and their features are known to reflect changes in brain state in neuropsychiatric diseases. However, previous studies have only reported differences in each microstate feature and have not determined whether microstate features are suitable for schizophrenia classification. Therefore, it is necessary to validate microstate features for schizophrenia classification. Nineteen microstate features, including duration, occurrence, and coverage as well as thirty-one conventional EEG features, including statistical, frequency, and temporal characteristics were obtained from resting-state EEG recordings of 14 patients diagnosed with schizophrenia and from 14 healthy (control) subjects. Machine-learning based multivariate analysis was used to evaluate classification performance. EEG recordings of patients and controls showed different microstate features. More importantly, when differentiating among patients and controls, EEG microstate features outperformed conventional EEG ones. The performance of the microstate features exceeded that of conventional EEG, even after optimization using recursive feature elimination. EEG microstate features applied with conventional EEG features also showed better classification performance than conventional EEG features alone. The current study is the first to validate the use of microstate features to discriminate schizophrenia, suggesting that EEG microstate features are useful for schizophrenia classification.

Determining the Spike-Wave Index Using Automated Detection Software

PURPOSE

The spike-wave index (SWI) is a key feature in the diagnosis of electrical status epilepticus during slow-wave sleep. Estimating the SWI manually is time-consuming and is subject to interrater and intrarater variability. Use of automated detection software would save time. Thereby, this software will consistently detect a certain EEG phenomenon as epileptiform and is not influenced by human factors. To determine noninferiority in calculating the SWI, we compared the performance of a commercially available spike detection algorithm (P13 software, Persyst Development Corporation, San Diego, CA) with human expert consensus.

METHODS

The authors identified all prolonged EEG recordings for the diagnosis or follow-up of electrical status epilepticus during slow-wave sleep carried out from January to December 2018 at an epilepsy tertiary referral center. The SWI during the first 10 minutes of sleep was estimated by consensus of two human experts. This was compared with the SWI calculated by the automated spike detection algorithm using the three available sensitivity settings: "low," "medium," and "high." In the software, these sensitivity settings are denoted as perception values.

RESULTS

Forty-eight EEG recordings from 44 individuals were analyzed. The SWIs estimated by human experts did not differ from the SWIs calculated by the automated spike detection algorithm in the "low" perception mode (P = 0.67). The SWIs calculated in the "medium" and "high" perception settings were, however, significantly higher than the human expert estimated SWIs (both P < 0.001).

CONCLUSIONS

Automated spike detection (P13) is a useful tool in determining SWI, especially when using the "low" sensitivity setting. Using such automated detection tools may save time, especially when reviewing larger epochs.

The influence of information about the circumstances of EEG recordings on the ability to identify seizure patterns

Purpose To quantify the influence of prior knowledge about the patient and the EEG circumstances on the EEG-based seizure detection rate. Methods A sample of 95 EEGs with epileptic seizure patterns matched with 95 seizure-free control sequences were extracted from EEG video monitoring data. They were stripped of all additional information. These plain EEG recordings were evaluated by two board certified EEG reviewers. The results were compared with the interpretations of the original video monitoring evaluations. Results Using the plain EEG sequences, epileptic seizure patterns were detected with a sensitivity and specificity of 0.758 and 0.958, respectively. The classification of the seizure pattern localization and lateralization differed in 56% and 50%, respectively, from the results of the video monitoring evaluations. Conclusion Additional information about the patient and the events during an EEG recording leads to a clinically and statistically significant increase in the seizure detection rates. These results imply that the human evaluation of a plain EEG without further information may not be seen as the gold standard in EEG evaluation. The performance estimation of automated EEG evaluation methods should take this into account.

Prevention of Epilepsy in Infants with Tuberous Sclerosis Complex in the EPISTOP Trial

OBJECTIVE

Epilepsy develops in 70 to 90% of children with tuberous sclerosis complex (TSC) and is often resistant to medication. Recently, the concept of preventive antiepileptic treatment to modify the natural history of epilepsy has been proposed. EPISTOP was a clinical trial designed to compare preventive versus conventional antiepileptic treatment in TSC infants.

METHODS

In this multicenter study, 94 infants with TSC without seizure history were followed with monthly video electroencephalography (EEG), and received vigabatrin either as conventional antiepileptic treatment, started after the first electrographic or clinical seizure, or preventively when epileptiform EEG activity before seizures was detected. At 6 sites, subjects were randomly allocated to treatment in a 1:1 ratio in a randomized controlled trial (RCT). At 4 sites, treatment allocation was fixed; this was denoted an open-label trial (OLT). Subjects were followed until 2 years of age. The primary endpoint was the time to first clinical seizure.

RESULTS

In 54 subjects, epileptiform EEG abnormalities were identified before seizures. Twenty-seven were included in the RCT and 27 in the OLT. The time to the first clinical seizure was significantly longer with preventive than conventional treatment [RCT: 364 days (95% confidence interval [CI] = 223-535) vs 124 days (95% CI = 33-149); OLT: 426 days (95% CI = 258-628) vs 106 days (95% CI = 11-149)]. At 24 months, our pooled analysis showed preventive treatment reduced the risk of clinical seizures (odds ratio [OR] = 0.21, p = 0.032), drug-resistant epilepsy (OR = 0.23, p = 0.022), and infantile spasms (OR = 0, p < 0.001). No adverse events related to preventive treatment were noted.

INTERPRETATION

Preventive treatment with vigabatrin was safe and modified the natural history of seizures in TSC, reducing the risk and severity of epilepsy. ANN NEUROL 2021;89:304-314.

Electroencephalography Technologist Inter-rater Agreement and Interpretation of Pediatric Critical Care Electroencephalography

OBJECTIVES

Electroencephalography (EEG) technologists commonly screen continuous EEG. Until now, the inter-rater agreement or sensitivity for important EEG findings has been unknown in this group.

METHODS

Twenty-nine EEG technologists and three clinical neurophysiologists interpreted 90 five-minute samples of pediatric critical care EEG. Inter-rater agreement was examined with Cohen's kappa and Fleiss' kappa for EEG findings. A gold-standard consensus agreement was developed for examining sensitivity and specificity for seizures or discontinuity. Kruskal-Wallis tests with Benjamani-Hochberg corrections for multiple comparisons were utilized to examine associations between correct scoring and certification status and years of experience.

RESULTS

Aggregate agreement was moderate for seizures and fair for EEG background continuity among EEG technologists. Individual agreement for seizures and continuity varied from slight to substantial. For individual EEG technologists, sensitivity for seizures ranged from 44 to 93% and sensitivity for continuity ranged from 81 to 100%. Raters with Certified Long Term Monitoring credentials were more likely to identify seizures correctly.

SIGNIFICANCE

This is the first study to evaluate inter-rater agreement and interpretation correctness among EEG technologists interpreting pediatric critical care EEG. EEG technologists demonstrated better aggregate agreement for seizure detection than other EEG findings, yet individual results and internal consistency varied widely. These data provide important insight into the common practice of utilizing EEG technologists for screening critical care EEG.

Automated Adult Epilepsy Diagnostic Tool Based on Interictal Scalp Electroencephalogram Characteristics: A Six-Center Study

The diagnosis of epilepsy often relies on a reading of routine scalp electroencephalograms (EEGs). Since seizures are highly unlikely to be detected in a routine scalp EEG, the primary diagnosis depends heavily on the visual evaluation of Interictal Epileptiform Discharges (IEDs). This process is tedious, expert-centered, and delays the treatment plan. Consequently, the development of an automated, fast, and reliable epileptic EEG diagnostic system is essential. In this study, we propose a system to classify EEG as epileptic or normal based on multiple modalities extracted from the interictal EEG. The ensemble system consists of three components: a Convolutional Neural Network (CNN)-based IED detector, a Template Matching (TM)-based IED detector, and a spectral feature-based classifier. We evaluate the system on datasets from six centers from the USA, Singapore, and India. The system yields a mean Leave-One-Institution-Out (LOIO) cross-validation (CV) area under curve (AUC) of 0.826 (balanced accuracy (BAC) of 76.1%) and Leave-One-Subject-Out (LOSO) CV AUC of 0.812 (BAC of 74.8%). The LOIO results are found to be similar to the interrater agreement (IRA) reported in the literature for epileptic EEG classification. Moreover, as the proposed system can process routine EEGs in a few seconds, it may aid the clinicians in diagnosing epilepsy efficiently.

Epilepsy surveillance in normocephalic children with and without prenatal Zika virus exposure

Children with Congenital Zika Syndrome and microcephaly are at high risk for epilepsy; however, the risk is unclear in normocephalic children with prenatal Zika virus (ZIKV) exposure [Exposed Children (EC)]. In this prospective cohort study, we performed epilepsy screening in normocephalic EC alongside a parallel group of normocephalic unexposed children [Unexposed Children (UC)]. We compared the incidence rate of epilepsy among EC and UC at one year of life to global incidence rates. Pregnant women were recruited from public health centers during the ZIKV outbreak in Grenada, West Indies and assessed for prior ZIKV infection using a plasmonic-gold platform that measures IgG antibodies in serum. Normocephalic children born to mothers with positive ZIKV results during pregnancy were classified as EC and those born to mothers with negative ZIKV results during and after pregnancy were classified as UC. Epilepsy screening procedures included a pediatric epilepsy screening questionnaire and video electroencephalography (vEEG). vEEG was collected using a multi-channel microEEG® system for a minimum of 20 minutes along with video recording of participant behavior time-locked to the EEG. vEEGs were interpreted independently by two pediatric epileptologists, who were blinded to ZIKV status, via telemedicine platform. Positive screening cases were referred to a local pediatrician for an epilepsy diagnostic evaluation. Epilepsy screens were positive in 2/71 EC (IR: 0.028; 95% CI: 0.003-0.098) and 0/71 UC. In both epilepsy-positive cases, questionnaire responses and interictal vEEGs were consistent with focal, rather than generalized, seizures. Both children met criteria for a clinical diagnosis of epilepsy and good seizure control was achieved with carbamazepine. Our results indicate that epilepsy rates are modestly elevated in EC. Given our small sample size, results should be considered preliminary. They support the use of epilepsy screening procedures in larger epidemiological studies of children with congenital ZIKV exposure, even in the absence of microcephaly, and provide guidance for conducting epilepsy surveillance in resource limited settings.

Proceedings of the 15th Antiepileptic Drug and Device Trials Meeting: State of the Science

On May 22-24, 2019, the 15th Antiepileptic Drug and Device (AEDD) Trials Conference was held, which focused on current issues related to AEDD development from preclinical models to clinical prognostication. The conference featured regulatory agencies, academic laboratories, and healthcare companies involved in emerging epilepsy therapies and research. The program included discussions around funding and support for investigations in epilepsy and neurologic research, clinical trial design and integrated outcome measures for people with epilepsy, and drug development and upcoming disease-modifying therapies. Finally, the conference included updates from the preclinical, clinical, and device pipeline. Summaries of the talks are provided in this paper, with the various pipeline therapeutics in the listed tables to be outlined in a subsequent publication.

Eignung eines mobilen Trockenelektroden-EEG-Gerätes im Rahmen der Epilepsiediagnostik

EEG-Aufzeichnungen bei Verdacht auf Epilepsie erfolgen routinemäßig mit einer durchschnittlichen Ableitezeit von 20–30 Min. mittels stationärer Geräte. Längere und häufigere Ableitungen, auch in der Häuslichkeit der Patienten, erhöhen die Wahrscheinlichkeit, Ereignisse zu erfassen. Die technische Qualität und medizinische Auswertbarkeit der EEG-Aufzeichnungen sind Grundvoraussetzungen für ein Home-Monitoring. Die HOMEEPI Studie prüft die technische Verwertbarkeit und Wirksamkeit eines mobilen EEG-Gerätes mit Trockenelektroden (Fourier ONE) im Vergleich zu einem konventionellen EEG-Gerät bei 49 Patienten mit Verdacht auf Epilepsie. Die Studienergebnisse basieren auf Intra- und Interratervergleichen und belegen eine vergleichbare Qualität der EEG-Aufzeichnungen und eine hohe Übereinstimmungsrate der medizinischen Befunde.

Machine-learning-based diagnostics of EEG pathology

Machine learning (ML) methods have the potential to automate clinical EEG analysis. They can be categorized into feature-based (with handcrafted features), and end-to-end approaches (with learned features). Previous studies on EEG pathology decoding have typically analyzed a limited number of features, decoders, or both. For a I) more elaborate feature-based EEG analysis, and II) in-depth comparisons of both approaches, here we first develop a comprehensive feature-based framework, and then compare this framework to state-of-the-art end-to-end methods. To this aim, we apply the proposed feature-based framework and deep neural networks including an EEG-optimized temporal convolutional network (TCN) to the task of pathological versus non-pathological EEG classification. For a robust comparison, we chose the Temple University Hospital (TUH) Abnormal EEG Corpus (v2.0.0), which contains approximately 3000 EEG recordings. The results demonstrate that the proposed feature-based decoding framework can achieve accuracies on the same level as state-of-the-art deep neural networks. We find accuracies across both approaches in an astonishingly narrow range from 81 to 86%. Moreover, visualizations and analyses indicated that both approaches used similar aspects of the data, e.g., delta and theta band power at temporal electrode locations. We argue that the accuracies of current binary EEG pathology decoders could saturate near 90% due to the imperfect inter-rater agreement of the clinical labels, and that such decoders are already clinically useful, such as in areas where clinical EEG experts are rare. We make the proposed feature-based framework available open source and thus offer a new tool for EEG machine learning research.

Comparison of mobile and clinical EEG sensors through resting state simultaneous data collection

Development of mobile sensors brings new opportunities to medical research. In particular, mobile electroencephalography (EEG) devices can be potentially used in low cost screening for epilepsy and other neurological and psychiatric disorders. The necessary condition for such applications is thoughtful validation in the specific medical context. As part of validation and quality assurance, we developed a computer-based analysis pipeline, which aims to compare the EEG signal acquired by a mobile EEG device to the one collected by a medically approved clinical-grade EEG device. Both signals are recorded simultaneously during 30 min long sessions in resting state. The data are collected from 22 patients with epileptiform abnormalities in EEG. In order to compare two multichannel EEG signals with differently placed references and electrodes, a novel data processing pipeline is proposed. It allows deriving matching pairs of time series which are suitable for similarity assessment through Pearson correlation. The average correlation of 0.64 is achieved on a test dataset, which can be considered a promising result, taking the positions shift due to the simultaneous electrode placement into account.