Linear and non-linear methods for automatic seizure detection in scalp electro-encephalogram recordings

Epileptic seizure detection using CHB-MIT dataset: The overlooked perspectives

Epilepsy is a life-threatening neurological condition. Manual detection of epileptic seizures (ES) is laborious and burdensome. Machine learning techniques applied to electroencephalography (EEG) signals are widely used for automatic seizure detection. Some key factors are worth considering for the real-world applicability of such systems: (i) continuous EEG data typically has a higher class imbalance; (ii) higher variability across subjects is present in physiological signals such as EEG; and (iii) seizure event detection is more practical than random segment detection. Most prior studies failed to address these crucial factors altogether for seizure detection. In this study, we intend to investigate a generalized cross-subject seizure event detection system using the continuous EEG signals from the CHB-MIT dataset that considers all these overlooked aspects. A 5-second non-overlapping window is used to extract 92 features from 22 EEG channels; however, the most significant 32 features from each channel are used in experimentation. Seizure classification is done using a Random Forest (RF) classifier for segment detection, followed by a post-processing method used for event detection. Adopting all the above-mentioned essential aspects, the proposed event detection system achieved 72.63% and 75.34% sensitivity for subject-wise 5-fold and leave-one-out analyses, respectively. This study presents the real-world scenario for ES event detectors and furthers the understanding of such detection systems.

Dual-Modal Information Bottleneck Network for Seizure Detection

In recent years, deep learning has shown very competitive performance in seizure detection. However, most of the currently used methods either convert electroencephalogram (EEG) signals into spectral images and employ 2D-CNNs, or split the one-dimensional (1D) features of EEG signals into many segments and employ 1D-CNNs. Moreover, these investigations are further constrained by the absence of consideration for temporal links between time series segments or spectrogram images. Therefore, we propose a Dual-Modal Information Bottleneck (Dual-modal IB) network for EEG seizure detection. The network extracts EEG features from both time series and spectrogram dimensions, allowing information from different modalities to pass through the Dual-modal IB, requiring the model to gather and condense the most pertinent information in each modality and only share what is necessary. Specifically, we make full use of the information shared between the two modality representations to obtain key information for seizure detection and to remove irrelevant feature between the two modalities. In addition, to explore the intrinsic temporal dependencies, we further introduce a bidirectional long-short-term memory (BiLSTM) for Dual-modal IB model, which is used to model the temporal relationships between the information after each modality is extracted by convolutional neural network (CNN). For CHB-MIT dataset, the proposed framework can achieve an average segment-based sensitivity of 97.42%, specificity of 99.32%, accuracy of 98.29%, and an average event-based sensitivity of 96.02%, false detection rate (FDR) of 0.70/h. We release our code at https://github.com/LLLL1021/Dual-modal-IB.

Seizure Detection and Prediction by Parallel Memristive Convolutional Neural Networks

During the past two decades, epileptic seizure detection and prediction algorithms have evolved rapidly. However, despite significant performance improvements, their hardware implementation using conventional technologies, such as Complementary Metal-Oxide-Semiconductor (CMOS), in power and area-constrained settings remains a challenging task; especially when many recording channels are used. In this paper, we propose a novel low-latency parallel Convolutional Neural Network (CNN) architecture that has between 2-2,800x fewer network parameters compared to State-Of-The-Art (SOTA) CNN architectures and achieves 5-fold cross validation accuracy of 99.84% for epileptic seizure detection, and 99.01% and 97.54% for epileptic seizure prediction, when evaluated using the University of Bonn Electroencephalogram (EEG), CHB-MIT and SWEC-ETHZ seizure datasets, respectively. We subsequently implement our network onto analog crossbar arrays comprising Resistive Random-Access Memory (RRAM) devices, and provide a comprehensive benchmark by simulating, laying out, and determining hardware requirements of the CNN component of our system. We parallelize the execution of convolution layer kernels on separate analog crossbars to enable 2 orders of magnitude reduction in latency compared to SOTA hybrid Memristive-CMOS Deep Learning (DL) accelerators. Furthermore, we investigate the effects of non-idealities on our system and investigate Quantization Aware Training (QAT) to mitigate the performance degradation due to low Analog-to-Digital Converter (ADC)/Digital-to-Analog Converter (DAC) resolution. Finally, we propose a stuck weight offsetting methodology to mitigate performance degradation due to stuck [Formula: see text] memristor weights, recovering up to 32% accuracy, without requiring retraining. The CNN component of our platform is estimated to consume approximately 2.791 W of power while occupying an area of 31.255 mm² in a 22 nm FDSOI CMOS process.

Representation based on ordinal patterns for seizure detection in EEG signals

EEG signals carry rich information about brain activity and play an important role in the diagnosis and recognition of epilepsy. Numerous algorithms using EEG signals to detect seizures have been developed in recent decades. However, most of them require well-designed features that highly depend on domain-specific knowledge and algorithm expertise. In this study, we introduce the unigram ordinal pattern (UniOP) and bigram ordinal pattern (BiOP) representations to capture the different underlying dynamics of time series, which only assumes that time series derived from different dynamics can be characterized by repeated ordinal patterns. Specifically, we first transform each subsequence in a time series into the corresponding ordinal pattern in terms of the ranking of values and consider the distribution of ordinal patterns of all subsequences as the UniOP representation. Furthermore, we consider the distribution of the cooccurrence of ordinal patterns as the BiOP representation to characterize the contextual information for each ordinal pattern. We then combine the proposed representations with the nearest neighbor algorithm to evaluate its effectiveness on three publicly available seizure datasets. The results on the Bonn EEG dataset demonstrate that this method provides more than 90% accuracy, sensitivity, and specificity in most cases and outperforms several state-of-the-art methods, which proves its ability to capture the key information of the underlying dynamics of EEG time series at healthy, seizure-free, and seizure states. The results on the second dataset are comparable with the state-of-the-art method, showing the good generalization ability of the proposed method. All performance metrics on the third dataset are approximately 89%, which demonstrates that the proposed representations are suitable for large-scale datasets.

Epileptic Seizure Detection with EEG Textural Features and Imbalanced Classification Based on EasyEnsemble Learning

Imbalance data classification is a challenging task in automatic seizure detection from electroencephalogram (EEG) recordings when the durations of non-seizure periods are much longer than those of seizure activities. An imbalanced learning model is proposed in this paper to improve the identification of seizure events in long-term EEG signals. To better represent the underlying microstructure distributions of EEG signals while preserving the non-stationary nature, discrete wavelet transform (DWT) and uniform 1D-LBP feature extraction procedure are introduced. A learning framework is then designed by the ensemble of weakly trained support vector machines (SVMs). Under-sampling is employed to split the imbalanced seizure and non-seizure samples into multiple balanced subsets where each of them is utilized to train an individual SVM classifier. The weak SVMs are incorporated to build a strong classifier which emphasizes seizure samples and in the meantime analyzing the imbalanced class distribution of EEG data. Final seizure detection results are obtained in a multi-level decision fusion process by considering temporal and frequency factors. The model was validated over two long-term and one short-term public EEG databases. The model achieved a [Formula: see text]-mean of 97.14% with respect to epoch-level assessment, an event-level sensitivity of 96.67%, and a false detection rate of 0.86/h on the long-term intracranial database. An epoch-level [Formula: see text]-mean of 95.28% and event-level false detection rate of 0.81/h were yielded over the long-term scalp database. The comparisons with 14 published methods demonstrated the improved detection performance for imbalanced EEG signals and the generalizability of the proposed model.

Epilepsy Seizure Prediction on EEG Using Common Spatial Pattern and Convolutional Neural Network

Epilepsy seizure prediction paves the way of timely warning for patients to take more active and effective intervention measures. Compared to seizure detection that only identifies the inter-ictal state and the ictal state, far fewer researches have been conducted on seizure prediction because the high similarity makes it challenging to distinguish between the pre-ictal state and the inter-ictal state. In this paper, a novel solution on seizure prediction is proposed using common spatial pattern (CSP) and convolutional neural network (CNN). Firstly, artificial pre-ictal EEG signals based on the original ones are generated by combining the segmented pre-ictal signals to solve the trial imbalance problem between the two states. Secondly, a feature extractor employing wavelet packet decomposition and CSP is designed to extract the distinguishing features in both the time domain and the frequency domain. It can improve overall accuracy while reducing the training time. Finally, a shallow CNN is applied to discriminate between the pre-ictal state and the inter-ictal state. Our proposed solution is evaluated on 23 patients' data from Boston Children's Hospital-MIT scalp EEG dataset by employing a leave-one-out cross-validation, and it achieves a sensitivity of 92.2% and false prediction rate of 0.12/h. Experimental result demonstrates that the proposed approach outperforms most state-of-the-art methods.

CONVOLUTIONAL LONG-SHORT TERM MEMORY NETWORKS MODEL FOR LONG DURATION EEG SIGNAL CLASSIFICATION

Background and objective: Deep learning structures have recently achieved remarkable success in the field of machine learning. Convolutional neural networks (CNN) in image processing and long-short term memory (LSTM) in the time-series analysis are commonly used deep learning algorithms. Healthcare applications of deep learning algorithms provide important contributions for computer-aided diagnosis research. In this study, convolutional long-short term memory (CLSTM) network was used for automatic classification of EEG signals and automatic seizure detection.

Methods: A new nine-layer deep network model consisting of convolutional and LSTM layers was designed. The signals processed in the convolutional layers were given as an input to the LSTM network whose outputs were processed in densely connected neural network layers. The EEG data is appropriate for a model having 1-D convolution layers. A bidirectional model was employed in the LSTM layer.

Results: Bonn University EEG database with five different datasets was used for experimental studies. In this database, each dataset contains 23.6[Formula: see text]s duration 100 single channel EEG segments which consist of 4097 dimensional samples (173.61[Formula: see text]Hz). Eight two-class and three three-class clinical scenarios were examined. When the experimental results were evaluated, it was seen that the proposed model had high accuracy on both binary and ternary classification tasks.

Conclusions: The proposed end-to-end learning structure showed a good performance without using any hand-crafted feature extraction or shallow classifiers to detect the seizures. The model does not require filtering, and also automatically learns to filter the input as well. As a result, the proposed model can process long duration EEG signals without applying segmentation, and can detect epileptic seizures automatically by using the correlation of ictal and interictal signals of raw data.

Automatic Computer-Based Detection of Epileptic Seizures

Automatic computer-based seizure detection and warning devices are important for objective seizure documentation, for SUDEP prevention, to avoid seizure related injuries and social embarrassments as a consequence of seizures, and to develop on demand epilepsy therapies. Automatic seizure detection systems can be based on direct analysis of epileptiform discharges on scalp-EEG or intracranial EEG, on the detection of motor manifestations of epileptic seizures using surface electromyography (sEMG), accelerometry (ACM), video detection systems and mattress sensors and finally on the assessment of changes of physiologic parameters accompanying epileptic seizures measured by electrocardiography (ECG), respiratory monitors, pulse oximetry, surface temperature sensors, and electrodermal activity. Here we review automatic seizure detection based on scalp-EEG, ECG, and sEMG. Different seizure types affect preferentially different measurement parameters. While EEG changes accompany all types of seizures, sEMG and ACM are suitable mainly for detection of seizures with major motor manifestations. Therefore, seizure detection can be optimized by multimodal systems combining several measurement parameters. While most systems provide sensitivities over 70%, specificity expressed as false alarm rates still needs to be improved. Patients' acceptance and comfort of a specific device are of critical importance for its long-term application in a meaningful clinical way.

Various epileptic seizure detection techniques using biomedical signals: a review

Epilepsy is a chronic chaos of the central nervous system that influences individual’s daily life by putting it at risk due to repeated seizures. Epilepsy affects more than 2% people worldwide of which developing countries are affected worse. A seizure is a transient irregularity in the brain’s electrical activity that produces disturbing physical symptoms such as a lapse in attention and memory, a sensory illusion, etc. Approximately one out of every three patients have frequent seizures, despite treatment with multiple anti-epileptic drugs. According to a survey, population aged 65 or above in European Union is predicted to rise from 16.4% (2004) to 29.9% (2050) and also this tremendous increase in aged population is also predicted for other countries by 2050. In this paper, seizure detection techniques are classified as time, frequency, wavelet (time–frequency), empirical mode decomposition and rational function techniques. The aim of this review paper is to present state-of-the-art methods and ideas that will lead to valid future research direction in the field of seizure detection.

Seizures Start without Common Signatures of Critical Transition

Complex dynamical systems may exhibit sudden autonomous changes from one state to another. Such changes that occur rapidly in comparison to the regular dynamics have been termed critical transitions. Examples of such phenomena can be found in many complex systems: changes in climate and ocean circulation, changes in wildlife populations, financial crashes, as well as in medical conditions like asthma attacks and depression. It has been recognized that critical transitions, even if they arise in completely different contexts and situations, share several common attributes and also generic early-warning signals that indicate that a critical transition is approaching. In the present study, we review briefly the general characteristics that have been observed in systems prior to critical transitions and apply these general indicators to nearly 300 epileptic seizures collected from human subjects using invasive EEG. Only in about 8% of the patients was evidence of critical transitions found. In the remaining majority of cases no early warning signals that behaved consistently prior to seizures were observed. These results do not rule out the possibility of critical transitions to seizure but point to limited relevance of their early warning signals in the context of human epilepsy observed using intracranial EEG recordings.

Detection of seizure and epilepsy using higher order statistics in the EMD domain

In this paper, a method using higher order statistical moments of EEG signals calculated in the empirical mode decomposition (EMD) domain is proposed for detecting seizure and epilepsy. The appropriateness of these moments in distinguishing the EEG signals is investigated through an extensive analysis in the EMD domain. An artificial neural network is employed as the classifier of the EEG signals wherein these moments are used as features. The performance of the proposed method is studied using a publicly available benchmark database for various classification cases that include healthy, interictal (seizure-free interval) and ictal (seizure), healthy and seizure, nonseizure and seizure, and interictal and ictal, and compared with that of several recent methods based on time-frequency analysis and statistical moments. It is shown that the proposed method can provide, in almost all the cases, 100% accuracy, sensitivity, and specificity, especially in the case of discriminating seizure activities from the nonseizure ones for patients with epilepsy while being much faster as compared to the time-frequency analysis-based techniques.

Improving seizure detection performance reporting: analysing the duration needed for a detection

Improving seizure detection performance relies not only on novel signal processing approaches but also on new accurate, reliable and comparable performance reporting to give researchers better and fairer tools for understanding the true algorithm operation. This paper investigates the sensitivity of current performance metrics to the duration of data that must be marked as candidate seizure activity before a seizure detection is made. The results demonstrate that not all metrics are insensitive to this high level choice in the algorithm design, and provide new approaches for comparing between reported algorithm performances in a robust and reliable manner.

Comparison of fractal dimension estimation algorithms for epileptic seizure onset detection

Fractal dimension (FD) is a natural measure of the irregularity of a curve. In this study the performances of three waveform FD estimation algorithms (i.e. Katz's, Higuchi's and the k-nearest neighbour (k-NN) algorithm) were compared in terms of their ability to detect the onset of epileptic seizures in scalp electroencephalogram (EEG). The selection of parameters involved in FD estimation, evaluation of the accuracy of the different algorithms and assessment of their robustness in the presence of noise were performed based on synthetic signals of known FD. When applied to scalp EEG data, Katz's and Higuchi's algorithms were found to be incapable of producing consistent changes of a single type (either a drop or an increase) during seizures. On the other hand, the k-NN algorithm produced a drop, starting close to the seizure onset, in most seizures of all patients. The k-NN algorithm outperformed both Katz's and Higuchi's algorithms in terms of robustness in the presence of noise and seizure onset detection ability. The seizure detection methodology, based on the k-NN algorithm, yielded in the training data set a sensitivity of 100% with 10.10 s mean detection delay and a false positive rate of 0.27 h(-1), while the corresponding values in the testing data set were 100%, 8.82 s and 0.42 h(-1), respectively. The above detection results compare favourably to those of other seizure onset detection methodologies applied to scalp EEG in the literature. The methodology described, based on the k-NN algorithm, appears to be promising for the detection of the onset of epileptic seizures based on scalp EEG.

Detecting epileptic seizures in long-term human EEG: a new approach to automatic online and real-time detection and classification of polymorphic seizure patterns

Epileptic seizures can cause a variety of temporary changes in perception and behavior. In the human EEG they are reflected by multiple ictal patterns, where epileptic seizures typically become apparent as characteristic, usually rhythmic signals, often coinciding with or even preceding the earliest observable changes in behavior. Their detection at the earliest observable onset of ictal patterns in the EEG can, thus, be used to start more-detailed diagnostic procedures during seizures and to differentiate epileptic seizures from other conditions with seizure-like symptoms. Recently, warning and intervention systems triggered by the detection of ictal EEG patterns have attracted increasing interest. Since the workload involved in the detection of seizures by human experts is quite formidable, several attempts have been made to develop automatic seizure detection systems. So far, however, none of these found widespread application. Here, we present a novel procedure for generic, online, and real-time automatic detection of multimorphologic ictal-patterns in the human long-term EEG and its validation in continuous, routine clinical EEG recordings from 57 patients with a duration of approximately 43 hours and additional 1,360 hours of seizure-free EEG data for the estimation of the false alarm rates. We analyzed 91 seizures (37 focal, 54 secondarily generalized) representing the six most common ictal morphologies (alpha, beta, theta, and delta- rhythmic activity, amplitude depression, and polyspikes). We found that taking the seizure morphology into account plays a crucial role in increasing the detection performance of the system. Moreover, besides enabling a reliable (mean false alarm rate<0.5/h, for specific ictal morphologies<0.25/h), early and accurate detection (average correct detection rate>96%) within the first few seconds of ictal patterns in the EEG, this procedure facilitates the automatic categorization of the prevalent seizure morphologies without the necessity to adapt the proposed system to specific patients.

Predicting epileptic seizures with a mental simulation task: a prospective study

The unpredictability of seizures is one of the most disabling aspects of epilepsy. We assessed the possibility of forecasting seizure onset by means of a behavioral task using mental simulation of biological and nonbiological movements. Four patients with epilepsy were included in a longitudinal study. In two patients with right temporal lobe epilepsy (RTLE), performance of the task changed depending on the epileptic cycle. In the preictal phase, timing precision decreased during simulation of biological movements and increased during simulation of nonbiological movements. In two other patients, the timing precision during simulation of biological movements was chronically decreased. The observed phase dependency of performance in patients with RTLE may indicate changes in brain states hours in advance of seizure onset. We suggest that our paradigm might be a simple and accurate method for the prediction of epileptic seizures. The validity of the approach for different types of epilepsy should be assessed in future studies.

Permutation entropy to detect vigilance changes and preictal states from scalp EEG in epileptic patients. A preliminary study

Permutation entropy (PE) was recently introduced as a very fast and robust algorithm to detect dynamic complexity changes in time series. It was also suggested as a useful screening algorithm for epileptic events in EEG data. In the present work, we tested its efficacy on scalp EEG data recorded from three epileptic patients. With a receiver operating characteristics (ROC) analysis, we evaluated the separability of amplitude distributions of PE resulting from preictal and interictal phases. Moreover, the dependency of PE on vigilance state was tested by correlation coefficients. A good separability of interictal and preictal phase was found, nevertheless PE was shown to be sensitive to changes in vigilance state. The changes of PE during the preictal phase and at seizure onset coincided with changes in vigilance state, restricting its possible use for seizure prediction on scalp EEG; this finding however suggests its possible usefulness for an automated classification of vigilance states.

State-of-the-Art of Seizure Prediction

SUMMARY

Although there are numerous studies exploring basic neuronal mechanisms that are likely to be associated with seizures, to date no definite information is available as to how, when, or why a seizure occurs in humans. The fact that seizures occur without warning in the majority of cases is one of the most disabling aspects of epilepsy. If it were possible to identify preictal precursors from the EEG of epilepsy patients, therapeutic possibilities and quality of life could improve dramatically. The last three decades have witnessed a rapid increase in the development of new EEG analysis techniques that appear to be capable of defining seizure precursors. Since the 1970s, studies on seizure prediction have advanced from preliminary descriptions of preictal phenomena and proof of principle studies via controlled studies to studies on continuous multiday recordings. At present, it is unclear whether prospective algorithms can predict seizures. If prediction algorithms are to be used in invasive seizure intervention techniques in humans, they must be proven to perform considerably better than a random predictor. The authors present an overview of the field of seizure prediction, its history, accomplishments, recent controversies, and potential for future development.

Automatic seizure detection based on time-frequency analysis and artificial neural networks

The recording of seizures is of primary interest in the evaluation of epileptic patients. Seizure is the phenomenon of rhythmicity discharge from either a local area or the whole brain and the individual behavior usually lasts from seconds to minutes. Since seizures, in general, occur infrequently and unpredictably, automatic detection of seizures during long-term electroencephalograph (EEG) recordings is highly recommended. As EEG signals are nonstationary, the conventional methods of frequency analysis are not successful for diagnostic purposes. This paper presents a method of analysis of EEG signals, which is based on time-frequency analysis. Initially, selected segments of the EEG signals are analyzed using time-frequency methods and several features are extracted for each segment, representing the energy distribution in the time-frequency plane. Then, those features are used as an input in an artificial neural network (ANN), which provides the final classification of the EEG segments concerning the existence of seizures or not. We used a publicly available dataset in order to evaluate our method and the evaluation results are very promising indicating overall accuracy from 97.72% to 100%.

Nonlinear dynamical analysis of EEG and MEG: review of an emerging field

Many complex and interesting phenomena in nature are due to nonlinear phenomena. The theory of nonlinear dynamical systems, also called 'chaos theory', has now progressed to a stage, where it becomes possible to study self-organization and pattern formation in the complex neuronal networks of the brain. One approach to nonlinear time series analysis consists of reconstructing, from time series of EEG or MEG, an attractor of the underlying dynamical system, and characterizing it in terms of its dimension (an estimate of the degrees of freedom of the system), or its Lyapunov exponents and entropy (reflecting unpredictability of the dynamics due to the sensitive dependence on initial conditions). More recently developed nonlinear measures characterize other features of local brain dynamics (forecasting, time asymmetry, determinism) or the nonlinear synchronization between recordings from different brain regions. Nonlinear time series has been applied to EEG and MEG of healthy subjects during no-task resting states, perceptual processing, performance of cognitive tasks and different sleep stages. Many pathologic states have been examined as well, ranging from toxic states, seizures, and psychiatric disorders to Alzheimer's, Parkinson's and Cre1utzfeldt-Jakob's disease. Interpretation of these results in terms of 'functional sources' and 'functional networks' allows the identification of three basic patterns of brain dynamics: (i) normal, ongoing dynamics during a no-task, resting state in healthy subjects; this state is characterized by a high dimensional complexity and a relatively low and fluctuating level of synchronization of the neuronal networks; (ii) hypersynchronous, highly nonlinear dynamics of epileptic seizures; (iii) dynamics of degenerative encephalopathies with an abnormally low level of between area synchronization. Only intermediate levels of rapidly fluctuating synchronization, possibly due to critical dynamics near a phase transition, are associated with normal information processing, whereas both hyper-as well as hyposynchronous states result in impaired information processing and disturbed consciousness.