The value of linear and non-linear quantitative EEG analysis in paediatric epilepsy surgery: a machine learning approach.

Epilepsy surgery is effective for patients with medication-resistant seizures, however 20-40% of them are not seizure free after surgery. Aim of this study is to evaluate the role of linear and non-linear EEG features to predict post-surgical outcome. We included 123 paediatric patients who underwent epilepsy surgery at Bambino Gesù Children Hospital (January 2009-April 2020). All patients had long term video-EEG monitoring. We analysed 1-min scalp interictal EEG (wakefulness and sleep) and extracted 13 linear and non-linear EEG features (power spectral density (PSD), Hjorth, approximate entropy, permutation entropy, Lyapunov and Hurst value). We used a logistic regression (LR) as feature selection process. To quantify the correlation between EEG features and surgical outcome we used an artificial neural network (ANN) model with 18 architectures. LR revealed a significant correlation between PSD of alpha band (sleep), Mobility index (sleep) and the Hurst value (sleep and awake) with outcome. The fifty-four ANN models gave a range of accuracy (46-65%) in predicting outcome. Within the fifty-four ANN models, we found a higher accuracy (64.8% ± 7.6%) in seizure outcome prediction, using features selected by LR. The combination of PSD of alpha band, mobility and the Hurst value positively correlate with good surgical outcome.

Changes in interhemispheric coherence after total corpus callosotomy: a scalp EEG study in children with non-lesional generalized epilepsy.

PURPOSE: Coherence analysis in electroencephalography (EEG) allows measurement of the degree of consistency of amplitude between pairs of electrodes. Theoretically, disconnective epilepsy surgery should decrease coherence between corresponding areas. The study aimed to evaluate postoperative changes in interhemispheric coherence values after corpus callosotomy (CC). METHODS: Non-lesional, drug-resistant, generalized epilepsy patients who underwent total CC were retrospectively collected. To evaluate coherence, we divided the scalp interictal EEG into "baseline" and "discharge" states after excluding periods with artifacts. Interhemispheric coherence values were obtained between eight pairs of symmetrically opposite scalp electrodes in six different frequency bands. We analyzed both pre- and postoperative EEG sessions and calculated the percentage of difference (POD) in coherence values. RESULTS: We collected 13 patients and analyzed 2496 interhemispheric coherence values. Preoperative coherence values differed significantly between baseline and discharge states (p = 0.0003), but postoperative values did not (p = 0.11). For baseline state, coherence values were decreased after CC and median POD was - 22.3% (p < 0.0001). Delta frequency showed the most decreased POD (-44.3%, p = 0.0009). Median POD was lowest in the Fp1-Fp2 pair of electrodes. For discharge state, coherence values were decreased after CC and median POD was - 24.7% (p < 0.0001). Delta frequency again showed the most decreased POD (-55.9%, p = 0.0016). Median POD was lowest in the F7-F8 pair. CONCLUSION: After total CC, interhemispheric coherence decreased significantly in both baseline and discharge states. The most decreased frequency band was the delta band, which may be used as a representative frequency band in future studies.

Detection of seizure onset in childhood absence epilepsy.

OBJECTIVE: This study aims to detect the seizure onset, in childhood absence epilepsy, as early as possible. Indeed, interfering with absence seizures with sensory simulation has been shown to be possible on the condition that the stimulation occurs soon enough after the seizure onset. METHODS: We present four variations (two supervised, two unsupervised) of an algorithm designed to detect the onset of absence seizures from 4 scalp electrodes, and compare their performance with that of a state-of-the-art algorithm. We exploit the characteristic shape of spike-wave discharges to detect the seizure onset. Their performance is assessed on clinical electroencephalograms from 63 patients with confirmed childhood absence epilepsy. RESULTS: The proposed approaches succeed in early detection of the seizure onset, contrary to the classical detection algorithm. Indeed, the results clearly show the superiority of the proposed methods for small delays of detection, under 750 ms from the onset. CONCLUSION: The performance of the proposed unsupervised methods is equivalent to that of the supervised ones. The use of only four electrodes makes the pipeline suitable to be embedded in a wearable device. SIGNIFICANCE: The proposed pipelines perform early detection of absence seizures, which constitutes a prerequisite for a closed-loop system.

EEG-based spatiotemporal dynamics of fast ripple networks and hubs in infantile epileptic spasms.

OBJECTIVE: Infantile epileptic spasms (IS) are epileptic seizures that are associated with increased risk for developmental impairments, adult epilepsies, and mortality. Here, we investigated coherence-based network dynamics in scalp EEG of infants with IS to identify frequency-dependent networks associated with spasms. We hypothesized that there is a network of increased fast ripple connectivity during the electrographic onset of clinical spasms, which is distinct from controls. METHODS: We retrospectively analyzed peri-ictal and interictal EEG recordings of 14 IS patients. The data was compared with 9 age-matched controls. Wavelet phase coherence (WPC) was computed between 0.2 and 400 Hz. Frequency- and time-dependent brain networks were constructed using this coherence as the strength of connection between two EEG channels, based on graph theory principles. Connectivity was evaluated through global efficiency (GE) and channel-based closeness centrality (CC), over frequency and time. RESULTS: GE in the fast ripple band (251-400 Hz) was significantly greater following the onset of spasms in all patients (P < 0.05). Fast ripple networks during the first 10s from spasm onset show enhanced anteroposterior gradient in connectivity (posterior > central > anterior, Kruskal-Wallis P < 0.001), with maximum CC over the centroparietal channels in 10/14 patients. Additionally, this anteroposterior gradient in CC connectivity is observed during spasms but not during the interictal awake or asleep states of infants with IS. In controls, anteroposterior gradient in fast ripple CC was noted during arousals and wakefulness but not during sleep. There was also a simultaneous decrease in GE in the 5-8 Hz range after the onset of spasms (P < 0.05), of unclear biological significance. SIGNIFICANCE: We identified an anteroposterior gradient in the CC connectivity of fast ripple hubs during spasms. This anteroposterior gradient observed during spasms is similar to the anteroposterior gradient in the CC connectivity observed in wakefulness or arousals in controls, suggesting that this state change is related to arousal networks.

Correspondence between scalp-EEG and stereoelectroencephalography seizure-onset patterns in patients with MRI-negative drug-resistant focal epilepsy.

OBJECTIVE: Our objective was to evaluate the relationship between scalp-EEG and stereoelectroencephalography (SEEG) seizure-onset patterns (SOP) in patients with MRI-negative drug-resistant focal epilepsy. METHODS: We analyzed retrospectively 41 patients without visible lesion on brain MRI who underwent video-EEG followed by SEEG. We defined five types of SOPs on scalp-EEG and eight types on SEEG. We examined how various clinical variables affected scalp-EEG SOPs. RESULTS: The most prevalent scalp SOPs were rhythmic sinusoidal activity (56.8%), repetitive epileptiform discharges (22.7%), and paroxysmal fast activity (15.9%). The presence of paroxysmal fast activity on scalp-EEG was always seen without delay from clinical onset and correlated with the presence of low-voltage fast activity in SEEG (sensitivity = 22.6%, specificity = 100%). The main factor explaining the discrepancy between the scalp and SEEG SOPs was the delay between clinical and scalp-EEG onset. There was a correlation between the scalp and SEEG SOPs when the scalp onset was simultaneous with the clinical onset (p = 0.026). A significant delay between clinical and scalp discharge onset was observed in 25% of patients and featured always with a rhythmic sinusoidal activity on scalp, corresponding to similar morphology of the discharge on SEEG. The presence of repetitive epileptiform discharges on scalp was associated with an underlying focal cortical dysplasia (sensitivity = 30%, specificity = 90%). There was no significant association between the scalp SOP and the epileptogenic zone location (deep or superficial), or surgical outcome. SIGNIFICANCE: In patients with MRI-negative focal epilepsy, scalp SOP could suggest the SEEG SOP and some etiology (focal cortical dysplasia) but has no correlation with surgical prognosis. Scalp SOP correlates with the SEEG SOP in cases of simultaneous EEG and clinical onset; otherwise, scalp SOP reflects the propagation of the SEEG discharge. PLAIN LANGUAGE SUMMARY: We looked at the correspondence between the electrical activity recorded during the start of focal seizure using scalp and intracerebral electrodes in patients with no visible lesion on MRI. If there is a fast activity on scalp, it reflects similar activity inside the brain. We found a good correspondence between scalp and intracerebral electrical activity for cases without significant delay between clinical and scalp electrical onset (seen in 75% of the cases we studied). Visualizing repetitive epileptic activity on scalp could suggest a particular cause of the epilepsy: a subtype of brain malformation called focal cortical dysplasia.

Scalp EEG-recorded high-frequency oscillations can predict seizure activity in Panayiotopoulos syndrome.

OBJECTIVE: We studied the relationship between the clinical course of Panayiotopoulos syndrome (PS) and high-frequency oscillations (HFOs) captured during interictal scalp electroencephalography (EEG) to determine the feasibility of using HFOs to detect seizure activity in PS. METHODS: We analyzed the interictal scalp EEGs of 18 children with PS. Age parameters, seizure frequencies, and antiepileptic drugs were compared between the HFO-positive (HFOPG) and HFO-negative (HFONG) groups. RESULTS: Thirteen patients (72.2%) had HFOs while five patients (27.8%) had no HFOs in 194 interictal EEG records. We found no statistically significant differences in the mean age of epilepsy onset and last seizure, seizure frequency, or frequency of status epilepticus. However, the seizure activity period of the HFOPG was significantly longer than that of the HFONG. Patients with an HFO duration longer than 2 years were intractable to treatment. In most cases, seizures did not occur in the absence of HFOs, even when the spikes remained. CONCLUSIONS: HFOs are related to the seizure activity period in patients with PS. SIGNIFICANCE: We propose that HFOs are a biomarker of epileptogenicity and an indicator for drug reduction because seizures did not occur if HFOs disappeared even if the spikes remained.

Graph neural networks in EEG spike detection.

OBJECTIVE: This study develops new machine learning architectures that are more adept at detecting interictal epileptiform discharges (IEDs) in scalp EEG. A comparison of results using the average precision (AP) metric is made with the proposed models on two datasets obtained from Baptist Hospital of Miami and Temple University Hospital. METHODS: Applying graph neural networks (GNNs) on functional connectivity (FC) maps of different frequency sub-bands to yield a novel architecture we call FC-GNN. Attention mechanism is applied on a complete graph to let the neural network select its important edges, hence bypassing the extraction of features, a model we refer to as CA-GNN. RESULTS: On the Baptist Hospital dataset, the results were as follows: Vanilla Self-Attention →0.9029±0.0431, Hierarchical Attention →0.8546±0.0587, Vanilla Visual Geometry Group (VGG) →0.92±0.0618, Satelight →0.9219±0.046, FC-GNN →0.9731±0.0187, and CA-GNN →0.9788±0.0125. In the same order, the results on the Temple University Hospital dataset are 0.9692, 0.9113, 0.97, 0.9575, 0.963, and 0.9879. CONCLUSION: Based on the good results they yield, GNNs prove to have a strong potential in detecting epileptogenic activity. SIGNIFICANCE: This study opens the door for the discovery of the powerful role played by GNNs in capturing IEDs, which is an essential step for identifying the epileptogenic networks of the affected brain and hence improving the prospects for more accurate 3D source localization.

Scalp high-frequency oscillations differentiate neonates with seizures from healthy neonates.

OBJECTIVE: We aimed to investigate (1) whether an automated detector can capture scalp high-frequency oscillations (HFO) in neonates and (2) whether scalp HFO rates can differentiate neonates with seizures from healthy neonates. METHODS: We considered 20 neonates with EEG-confirmed seizures and four healthy neonates. We applied a previously validated automated HFO detector to determine scalp HFO rates in quiet sleep. RESULTS: Etiology in neonates with seizures included hypoxic-ischemic encephalopathy in 11 cases, structural vascular lesions in 6, and genetic causes in 3. The HFO rates were significantly higher in neonates with seizures (0.098 ± 0.091 HFO/min) than in healthy neonates (0.038 ± 0.025 HFO/min; P = 0.02) with a Hedge's g value of 0.68 indicating a medium effect size. The HFO rate of 0.1 HFO/min/ch yielded the highest Youden index in discriminating neonates with seizures from healthy neonates. In neonates with seizures, etiology, status epilepticus, EEG background activity, and seizure patterns did not significantly impact HFO rates. SIGNIFICANCE: Neonatal scalp HFO can be detected automatically and differentiate neonates with seizures from healthy neonates. Our observations have significant implications for neuromonitoring in neonates. This is the first step in establishing neonatal HFO as a biomarker for neonatal seizures.

Painful Cutaneous Laser Stimulation for Temporal Summation of Pain Assessment.

Variability in pain sensitivity arises not only from the differences in peripheral sensory receptors but also from the differences in central nervous system (CNS) pain inhibition and facilitation mechanisms. Temporal summation of pain (TSP) is an experimental protocol commonly used in human studies of pain facilitation but is susceptible to confounding when elicited with the skin-contact thermode, which adds the responses of touch-related Aß low-threshold mechanoreceptors to nociceptive receptors. In the present study, we evaluate an alternative method involving the use of a contactless cutaneous laser for TSP assessment. We show that repetitive laser stimulations with a one second inter-stimulus interval evoked reliable TSP responses in a significant proportion of healthy subjects (N = 36). Female subjects (N = 18) reported greater TSP responses than male subjects confirming earlier studies of sex differences in central nociceptive excitability. Furthermore, repetitive laser stimulations during TSP induction elicited increased time-frequency electroencephalography (EEG) responses. The present study demonstrates that repetitive laser stimulation may be an alternative to skin-contact methods for TSP assessment in patients and healthy controls. PERSPECTIVE: Temporal summation of pain (TSP) is an experimental protocol commonly used in human studies of pain facilitation. We show that contactless cutaneous laser stimulation is a reliable alternative to the skin contact approaches during TSP assessment.

Psychogenic Nonepileptic Seizures-High Mortality Rate Is a 'Wake-Up Call'.

Patients with epilepsy have an elevated mortality rate compared to the general population and now studies are showing a comparable death ratio in patients diagnosed with psychogenic nonepileptic seizures. The latter is a top differential diagnosis for epilepsy and the unexpected mortality rate in these patients underscores the importance of an accurate diagnosis. Experts have called for more studies to elucidate this finding but the explanation is already available, embedded in the existing data. To illustrate, a review of the diagnostic practice in epilepsy monitoring units, of the studies examining mortality in PNES and epilepsy patients, and of the general clinical literature on the two populations was conducted. The analysis reveals that the scalp EEG test result, which distinguishes a psychogenic from an epileptic seizure, is highly fallible; that the clinical profiles of the PNES and epilepsy patient populations are virtually identical; and that both are dying of natural and non-natural causes including sudden unexpected death associated with confirmed or suspected seizure activity. The recent data showing a similar mortality rate simply constitutes more confirmatory evidence that the PNES population consists largely of patients with drug-resistant scalp EEG-negative epileptic seizures. To reduce the morbidity and mortality in these patients, they must be given access to treatments for epilepsy.

Brain functional connectivity-based prediction of vagus nerve stimulation efficacy in pediatric pharmacoresistant epilepsy.

OBJECTIVE: Although vagus nerve stimulation (VNS) is a common and widely used therapy for pharmacoresistant epilepsy, the reported efficacy of VNS in pediatric patients varies, so it is unclear which children will respond to VNS therapy. This study aimed to identify functional brain network features associated with VNS action to distinguish VNS responders from nonresponders using scalp electroencephalogram (EEG) data. METHODS: Twenty-three children were included in this study, 16 in the discovery cohort and 7 in the test cohort. Using partial correlation value as a measure of whole-brain functional connectivity, we identified the differential edges between responders and nonresponders. Results derived from this were used as input to generate a support vector machine-learning classifier to predict VNS outcomes. RESULTS: The postcentral gyrus in the left and right parietal lobe regions was identified as the most significant differential brain region between VNS responders and nonresponders (p < 0.001). The resultant classifier demonstrated a mean AUC value of 0.88, a mean sensitivity rate of 91.4%, and a mean specificity rate of 84.3% on fivefold cross-validation in the discovery cohort. In the testing cohort, our study demonstrated an AUC value of 0.91, a sensitivity rate of 86.6%, and a specificity rate of 79.3%. Furthermore, for prediction accuracy, our model can achieve 81.4% accuracy at the epoch level and 100% accuracy at the patient level. SIGNIFICANCE: This study provides the first treatment response prediction model for VNS using scalp EEG data with ictal recordings and offers new insights into its mechanism of action. Our results suggest that brain functional connectivity features can help predict therapeutic response to VNS therapy. With further validation, our model could facilitate the selection of targeted pediatric patients and help avoid risky and costly procedures for patients who are unlikely to benefit from VNS therapy.

[Construction of an epileptic seizure prediction model using a semi-supervised method of generative adversarial and long short term memory network combined with Stockwell transform].

OBJECTIVE: To propose a semi-supervised epileptic seizure prediction model (ST-WGAN-GP-Bi-LSTM) to enhance the prediction performance by improving time-frequency analysis of electroencephalogram (EEG) signals, enhancing the stability of the unsupervised feature learning model and improving the design of back-end classifier. METHODS: Stockwell transform (ST) of the epileptic EEG signals was performed to locate the time-frequency information by adaptive adjustment of the resolution and retaining the absolute phase to obtain the time-frequency inputs. When there was no overlap between the generated data distribution and the real EEG data distribution, to avoid failure of feature learning due to a constant JS divergence, Wasserstein GAN was used as a feature learning model, and the cost function based on EM distance and gradient penalty strategy was adopted to constrain the unsupervised training process to allow the generation of a high-order feature extractor. A temporal prediction model was finally constructed based on a bi-directional long short term memory network (Bi-LSTM), and the classification performance was improved by obtaining the temporal correlation between high-order time-frequency features. The CHB-MIT scalp EEG dataset was used to validate the proposed patient-specific seizure prediction method. RESULTS: The AUC, sensitivity, and specificity of the proposed method reached 90.40%, 83.62%, and 86.69%, respectively. Compared with the existing semi-supervised methods, the propose method improved the original performance by 17.77%, 15.41%, and 53.66%. The performance of this method was comparable to that of a supervised prediction model based on CNN. CONCLUSION: The utilization of ST, WGAN-GP, and Bi-LSTM effectively improves the prediction performance of the semi-supervised deep learning model, which can be used for optimization of unsupervised feature extraction in epileptic seizure prediction.

Association between postictal EEG suppression, postictal autonomic dysfunction, and sudden unexpected death in epilepsy: Evidence from intracranial EEG.

OBJECTIVE: The association between postictal electroencephalogram (EEG) suppression (PES), autonomic dysfunction, and Sudden Unexpected Death in Epilepsy (SUDEP) remains poorly understood. We compared PES on simultaneous intracranial and scalp-EEG and evaluated the association of PES with postictal heart rate variability (HRV) and SUDEP outcome. METHODS: Convulsive seizures were analyzed in patients with drug-resistant epilepsy at 5 centers. Intracranial PES was quantified using the Hilbert transform. HRV was quantified using root mean square of successive differences of interbeat intervals, low-frequency to high-frequency power ratio, and RR-intervals. RESULTS: There were 64 seizures from 63 patients without SUDEP and 11 seizures from 6 SUDEP patients. PES occurred in 99% and 87% of seizures on intracranial-EEG and scalp-EEG, respectively. Mean PES duration in intracranial and scalp-EEG was similar. Intracranial PES was regional (<90% of channels) in 46% of seizures; scalp PES was generalized in all seizures. Generalized PES showed greater decrease in postictal parasympathetic activity than regional PES. PES duration and extent were similar between patients with and without SUDEP. CONCLUSIONS: Regional intracranial PES can be present despite scalp-EEG demonstrating generalized or no PES. Postictal autonomic dysfunction correlates with the extent of PES. SIGNIFICANCE: Intracranial-EEG demonstrates changes in autonomic regulatory networks not seen on scalp-EEG.

Quantitative signal quality assessment for large-scale continuous scalp electroencephalography from a big data perspective.

Objective. Despite electroencephalography (EEG) being a widely used neuroimaging technique with an excellent temporal resolution, in practice, the signals are heavily contaminated by artifacts masking responses of interest in an experiment. It is thus essential to guarantee a prompt and effective detection of artifacts that provides quantitative quality assessment (QA) on raw EEG data. This type of pipeline is crucial for large-scale EEG studies. However, current EEG QA studies are still limited.Approach. In this study, combined from a big data perspective, we therefore describe a quantitative signal quality assessment pipeline, a stable and general threshold-based QA pipeline that automatically integrates artifact detection and new QA measures to assess continuous resting-state raw EEG data. One simulation dataset and two resting-state EEG datasets from 42 healthy subjects and 983 clinical patients were utilized to calibrate the QA pipeline.Main Results. The results demonstrate that (1) the QA indices selected are sensitive: they almost strictly and linearly decrease as the noise level increases; (2) stable, replicable QA thresholds are valid for other experimental and clinical EEG datasets; and (3) use of the QA pipeline on these datasets reveals that high-frequency noises are the most common noises in EEG practice. The QA pipeline is also deployed in the WeBrain cloud platform (https://webrain.uestc.edu.cn/, the Chinese EEG Brain Consortium portal).Significance. These findings suggest that the proposed QA pipeline may be a stable and promising approach for quantitative EEG signal quality assessment in large-scale EEG studies.

Prediction of patients with idiopathic generalized epilepsy from healthy controls using machine learning from scalp EEG recordings.

Epilepsy detection is essential for patients with epilepsy and their families, as well as for researchers and medical staff. The use of electroencephalogram (EEG) as a tool to support the diagnosis of patients with epilepsy is fundamental. Today, machine learning (ML) techniques are widely applied in neuroscience. The main objective of our study is to differentiate patients with idiopathic generalized epilepsy from healthy controls by applying machine learning techniques on interictal electroencephalographic recordings. Our research predicts which patients have idiopathic generalized epilepsy from a scalp EEG study. In addition, this study focuses on using the extreme gradient boosting (XGB) method applied to scalp EEG. XGB is one of the variants of gradient boosting and is a supervised learning algorithm. This type of system is developed to increase performance and processing speed. Through this proposed method, an attempt is made to recognize patterns from scalp EEG recordings that would allow the detection of IGE with high accuracy and differentiate IGE patients from healthy controls, creating an additional tool to support clinicians in their decision-making. Among the ML methods applied, the proposed XGB method achieves a better prediction of the distinct features in EEG signals from patients with IGE. XGB was 6.26% more accurate than the k-Nearest Neighbours method and was more accurate than the support vector machine (10.61%), decision tree (9.71%) and Gaussian Naïve Bayes (11.83%). Besides, the proposed XGB method showed the highest area under the curve (AUC 98%) and balanced accuracy (98.13%) of all methods tested. Application of ML technique in EEG of patients with epilepsy is very recent and is emerging with promising results. In this research work, we showed the usefulness of ML techniques to identify and predict generalized epilepsy from healthy controls in scalp EEG studies. These findings could help develop automated tools that integrate these ML techniques to assist clinicians in differentiating between patients with IGE from healthy controls in daily practice.

EEG-based high-performance depression state recognition.

Depression is a global disease that is harmful to people. Traditional identification methods based on various scales are not objective and accurate enough. Electroencephalogram (EEG) contains abundant physiological information, which makes it a new research direction to identify depression state. However, most EEG-based algorithms only extract the original EEG features and ignore the complex spatiotemporal information interactions, which will reduce performance. Thus, a more accurate and objective method for depression identification is urgently needed. In this work, we propose a novel depression identification model: W-GCN-GRU. In our proposed method, we censored six sensitive features based on Spearman's rank correlation coefficient and assigned different weight coefficients to each sensitive feature by AUC for the weighted fusion of sensitive features. In particular, we use the GCN and GRU cascade networks based on weighted sensitive features as depression recognition models. For the GCN, we creatively took the brain function network based on the correlation coefficient matrix as the adjacency matrix input and the weighted fused sensitive features were used as the node feature matrix input. Our proposed model performed well on our self-collected dataset and the MODMA datasets with a accuracy of 94.72%, outperforming other methods. Our findings showed that feature dimensionality reduction, weighted fusion, and EEG spatial information all had great effects on depression recognition.

Construction of an epileptic seizure prediction model using a semi-supervised method of generative adversarial and long short term memory network combined with Stockwell transform / 南方医科大学学报

OBJECTIVE@#To propose a semi-supervised epileptic seizure prediction model (ST-WGAN-GP-Bi-LSTM) to enhance the prediction performance by improving time-frequency analysis of electroencephalogram (EEG) signals, enhancing the stability of the unsupervised feature learning model and improving the design of back-end classifier.@*METHODS@#Stockwell transform (ST) of the epileptic EEG signals was performed to locate the time-frequency information by adaptive adjustment of the resolution and retaining the absolute phase to obtain the time-frequency inputs. When there was no overlap between the generated data distribution and the real EEG data distribution, to avoid failure of feature learning due to a constant JS divergence, Wasserstein GAN was used as a feature learning model, and the cost function based on EM distance and gradient penalty strategy was adopted to constrain the unsupervised training process to allow the generation of a high-order feature extractor. A temporal prediction model was finally constructed based on a bi-directional long short term memory network (Bi-LSTM), and the classification performance was improved by obtaining the temporal correlation between high-order time-frequency features. The CHB-MIT scalp EEG dataset was used to validate the proposed patient-specific seizure prediction method.@*RESULTS@#The AUC, sensitivity, and specificity of the proposed method reached 90.40%, 83.62%, and 86.69%, respectively. Compared with the existing semi-supervised methods, the propose method improved the original performance by 17.77%, 15.41%, and 53.66%. The performance of this method was comparable to that of a supervised prediction model based on CNN.@*CONCLUSION@#The utilization of ST, WGAN-GP, and Bi-LSTM effectively improves the prediction performance of the semi-supervised deep learning model, which can be used for optimization of unsupervised feature extraction in epileptic seizure prediction.

Recurrent Hippocampo-neocortical sleep-state divergence in humans.

Sleep is assumed to be a unitary, global state in humans and most other animals that is coordinated by executive centers in the brain stem, hypothalamus, and basal forebrain. However, the common observation of unihemispheric sleep in birds and marine mammals, as well as the recently discovered nonpathological regional sleep in rodents, calls into question whether the whole human brain might also typically exhibit different states between brain areas at the same time. We analyzed sleep states independently from simultaneously recorded hippocampal depth electrodes and cortical scalp electrodes in eight human subjects who were implanted with depth electrodes for pharmacologically intractable epilepsy evaluation. We found that the neocortex and hippocampus could be in nonsimultaneous states, on average, one-third of the night and that the hippocampus often led in asynchronous state transitions. Nonsimultaneous bout lengths varied from 30 s to over 30 min. These results call into question the conclusions of studies, across phylogeny, that measure only surface cortical state but seek to assess the functions and drivers of sleep states throughout the brain.

Interictal and seizure-onset EEG patterns in malformations of cortical development: A systematic review.

OBJECTIVES: Malformations of cortical development (MCDs) are common causes of drug-resistant epilepsy. The mechanisms underlying the associated epileptogenesis and ictogenesis remain poorly elucidated. EEG can help in understanding these mechanisms. We systematically reviewed studies reporting scalp or intracranial EEG features of MCDs to characterise interictal and seizure-onset EEG patterns across different MCD types. METHODS: We conducted a systematic review in accordance with PRISMA guidelines. MEDLINE, PubMed, and Cochrane databases were searched for studies describing interictal and seizure-onset EEG patterns in MCD patients. A classification framework was implemented to group EEG features into 20 predefined patterns, comprising nine interictal (five, scalp EEG; four, intracranial EEG) and 11 seizure-onset (five, scalp EEG; six, intracranial EEG) patterns. Logistic regression was used to estimate the odds ratios (OR) of each seizure-onset pattern being associated with specific MCD types. RESULTS: Our search yielded 1682 studies, of which 27 comprising 936 MCD patients were included. Of the nine interictal EEG patterns, five (three, scalp EEG; two, intracranial EEG) were detected in ≥2 MCD types, while four (rhythmic epileptiform discharges type 1 and type 2 on scalp EEG; repetitive bursting spikes and sporadic spikes on intracranial EEG) were seen only in focal cortical dysplasia (FCD). Of the 11 seizure-onset patterns, eight (three, scalp EEG; five, intracranial EEG) were found in ≥2 MCD types, whereas three were observed only in FCD (suppression on scalp EEG; delta brush on intracranial EEG) or tuberous sclerosis complex (TSC; focal fast wave on scalp EEG). Among scalp EEG seizure-onset patterns, paroxysmal fast activity (OR = 0.13; 95% CI: 0.03-0.53; p = 0.024) and repetitive epileptiform discharges (OR = 0.18; 95% CI: 0.05-0.61; p = 0.036) were less likely to occur in TSC than FCD. Among intracranial EEG seizure-onset patterns, low-voltage fast activity was more likely to be detected in heterotopia (OR = 19.3; 95% CI: 6.22-60.1; p < 0.001), polymicrogyria (OR = 6.70; 95% CI: 2.25-20.0; p = 0.004) and TSC (OR = 4.27; 95% CI: 1.88-9.70; p = 0.005) than FCD. SIGNIFICANCE: Different MCD types can share similar interictal or seizure-onset EEG patterns, reflecting common underlying biological mechanisms. However, selected EEG patterns appear to point to distinct MCD types, suggesting certain differences in their neuronal networks.

Automatic BASED scoring on scalp EEG in children with infantile spasms using convolutional neural network.

In recent years, the Burden of Amplitudes and Epileptiform Discharges (BASED) score has been used as a reliable, accurate, and feasible electroencephalogram (EEG) grading scale for infantile spasms. However, manual EEG annotation is, in general, very time-consuming, and BASED scoring is no exception. Convolutional neural networks (CNNs) have proven their great potential in many EEG classification problems. However, very few research studies have focused on the use of CNNs for BASED scoring, a challenging but vital task in the diagnosis and treatment of infantile spasms. This study proposes an automatic BASED scoring framework using EEG and a deep CNN. The feasibility of using CNN for automatic BASED scoring was investigated in 36 patients with infantile spasms by annotating their long-term EEG data with four levels of the BASED score (scores 5, 4, 3, and ≤2). In the validation set, the accuracy was 96.9% by applying a multi-layer CNN to classify the EEG data as a 4-label problem. The extensive experiments have demonstrated that our proposed approach offers high accuracy and, hence, is an important step toward an automatic BASED scoring algorithm. To the best of our knowledge, this is the first attempt to use a CNN to construct a BASED-based scoring model.