Context-based automated detection of epileptogenic sharp transients in the EEG: elimination of false positives

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.

Epileptic seizure focus detection from interictal electroencephalogram: a survey

Electroencephalogram (EEG) is one of most effective clinical diagnosis modalities for the localization of epileptic focus. Most current AI solutions use this modality to analyze the EEG signals in an automated manner to identify the epileptic seizure focus. To develop AI system for identifying the epileptic focus, there are many recently-published AI solutions based on biomarkers or statistic features that utilize interictal EEGs. In this review, we survey these solutions and find that they can be divided into three main categories: (i) those that use of biomarkers in EEG signals, including high-frequency oscillation, phase-amplitude coupling, and interictal epileptiform discharges, (ii) others that utilize feature-extraction methods, and (iii) solutions based upon neural networks (an end-to-end approach). We provide a detailed description of seizure focus with clinical diagnosis methods, a summary of the public datasets that seek to reduce the research gap in epilepsy, recent novel performance evaluation criteria used to evaluate the AI systems, and guidelines on when and how to use them. This review also suggests a number of future research challenges that must be overcome in order to design more efficient computer-aided solutions to epilepsy focus detection.

Computerized application for epilepsy in China: Does the era of artificial intelligence comes?

Epilepsy, one of the most common neurological diseases in China, is notorious for its spontaneous, unprovoked and recurrent seizures. The etiology of epilepsy varies among individual patients, including congenital gene mutation, traumatic injury, infections, etc. This heterogeneity partly hampered the accurate diagnosis and choice of appropriate treatments. Encouragingly, great achievements have been achieved in computational science, making it become a key player in medical fields gradually and bringing new hope for rapid and accurate diagnosis as well as targeted therapies in epilepsy. Here, we historically review the advances of computerized applications in epilepsy-especially those tremendous findings achieved in China-for different purposes including seizure prediction, localization of epileptogenic zone, post-surgical prognosis, etc. Special attentions are paid to the great progress based on artificial intelligence (AI), which is more "sensitive", "smart" and "in-depth" than human capacities. At last, we give a comprehensive discussion about the disadvantages and limitations of current computerized applications for epilepsy and propose some future directions as further stepping stones to embrace "the era of AI" in epilepsy.

Fast parametric curve matching (FPCM) for automatic spike detection

Epilepsy is a widely spread neurological disease, whose treatment often requires resection of the pathological cortical tissue. Interictal spike analysis observed in the non-invasively collected EEG or MEG data offers a way to localize epileptogenic cortical structures for surgery planning purposes. While a plethora of automatic spike detection techniques have been developed each with its own assumptions and limitations, non of them is ideal and the best results are achieved when the output of several automatic spike detectors are combined. This is especially true in the low signal-to-noise ratio conditions. To this end we propose a novel biomimetic approach for automatic spike detection based on a constrained mixed spline machinery that we dub as fast parametric curve matching (FPCM). Using the peak-wave shape parametrization, the constrained parametric morphological model is constructed and convolved with the observed multichannel data to very efficiently determine mixed spline parameters corresponding to each time-point in the dataset. Then the logical predicates that directly map to the expected interictal event morphology allow us to accomplish the spike detection task. The results of simulations mimicking typical low SNR scenario show the robustness and high ROC AUC values of the FPCM method as compared to the spike detection performed by the means of more conventional approaches such as wavelet decomposition, template matching or simple amplitude thresholding. Applied to the real MEG and EEG data from the human patients and to ECoG data from the rat, the FPCM technique demonstrates reliable detection of the interictal events and localization of epileptogenic zones concordant with independent conclusions made by the epileptologist. Since the FPCM is computationally light, tolerant to high amplitude artifacts and flexible to accommodate verbalized descriptions of the arbitrary target morphology, it may complement the existing arsenal of means for analysis of noisy interictal datasets.

A review on the pattern detection methods for epilepsy seizure detection from EEG signals

Over several years, research had been conducted for the detection of epileptic seizures to support an automatic diagnosis system to comfort the clinicians' encumbrance. In this regard, a number of research papers have been published for the identification of epileptic seizures. A thorough review of all these papers is required. So, an attempt has been made to review on the pattern detection methods for epilepsy seizure detection from EEG signals. More than 150 research papers have been discussed to determine the techniques for detecting epileptic seizures. Further, the literature review confirms that the pattern recognition techniques required to detect epileptic seizures varies across the electroencephalogram (EEG) datasets of different conditions. This is mostly owing to the fact that EEG detected under different conditions have different characteristics. This consecutively necessitates the identification of the pattern recognition technique to efficiently differentiate EEG epileptic data from the EEG data of various conditions.

Epilepsy detection from EEG signals: a review

Over many decades, research is being attempted for the detection of epileptic seizure to support for automatic diagnosis system to help clinicians from burdensome work. In this respect, an enormous number of research papers is published for identification of epileptic seizure. It is difficult to present a detailed review of all these literature. Therefore, in this paper, an attempt has been made to review the detection of an epileptic seizure. More than 100 research papers have been discussed to discern the techniques for detecting the epileptic seizure. Further, the literature survey shows that the pattern recognition required to detect epileptic seizure varies with different conditions of EEG datasets. This is mainly due to the fact that EEG detected under different conditions has different characteristics. This is, in turn, necessitates the identification of pattern recognition technique to effectively distinguish EEG epileptic data from a various condition of EEG data.

CLASSIFICATION OF EEG SIGNALS FOR DETECTION OF EPILEPTIC SEIZURES BASED ON WAVELETS AND STATISTICAL PATTERN RECOGNITION

The electroencephalogram (EEG) signal is very important in the diagnosis of epilepsy. Long-term EEG recordings of an epileptic patient contain a huge amount of EEG data. The detection of epileptic activity is, therefore, a very demanding process that requires a detailed analysis of the entire length of the EEG data, usually performed by an expert. This paper describes an automated classification of EEG signals for the detection of epileptic seizures using wavelet transform and statistical pattern recognition. The decision making process is comprised of three main stages: (a) feature extraction based on wavelet transform, (b) feature space dimension reduction using scatter matrices and (c) classification by quadratic classifiers. The proposed methodology was applied on EEG data sets that belong to three subject groups: (a) healthy subjects, (b) epileptic subjects during a seizure-free interval and (c) epileptic subjects during a seizure. An overall classification accuracy of 99% was achieved. The results confirmed that the proposed algorithm has a potential in the classification of EEG signals and detection of epileptic seizures, and could thus further improve the diagnosis of epilepsy.

EEG signals classification of epileptic patients via feature selection and voting criteria in intelligent method

Epileptic disease can be diagnosed by using intelligent methods on the Electroencephalograph (EEG) signals. In this paper, wavelet packet transform (WPT) was used in each of the frequency bands and wavelet coefficients were obtained, then the energy and entropy function was done on the wavelet coefficients and used as initial feature vectors. In the next step, eight and 15 features from 30 initial energy and entropy features were selected as the final features because their receiver operating characteristic (ROC) curve areas were higher than others. There were seven classifier inputs. These seven classifiers consisted of four artificial neural networks (ANN) with different structures, support vector machines (SVM), K-nearest neighbours (KNN) and a hybrid network. Each classifier was trained by 0.5, 0.8 and 0.9 EEG signals. After the training process, a fusion network based on a voting criteria was used to make the algorithm robust against the possible changes in each classifier and increase the classification accuracy. Finally, the algorithm was tested by other EEG signals. As a result, normal and epileptic classes were detected with total classification accuracy of 99-100%.

Detecting epileptic seizure from scalp EEG using Lyapunov spectrum

One of the inherent weaknesses of the EEG signal processing is noises and artifacts. To overcome it, some methods for prediction of epilepsy recently reported in the literature are based on the evaluation of chaotic behavior of intracranial electroencephalographic (EEG) recordings. These methods reduced noises, but they were hazardous to patients. In this study, we propose using Lyapunov spectrum to filter noise and detect epilepsy on scalp EEG signals only. We determined that the Lyapunov spectrum can be considered as the most expected method to evaluate chaotic behavior of scalp EEG recordings and to be robust within noises. Obtained results are compared to the independent component analysis (ICA) and largest Lyapunov exponent. The results of detecting epilepsy are compared to diagnosis from medical doctors in case of typical general epilepsy.

Automatic EEG spike detection

Since the 1970s advances in science and technology during each succeeding decade have renewed the expectation of efficient, reliable automatic epileptiform spike detection (AESD). But even when reinforced with better, faster tools, clinically reliable unsupervised spike detection remains beyond our reach. Expert-selected spike parameters were the first and still most widely used for AESD. Thresholds for amplitude, duration, sharpness, rise-time, fall-time, after-coming slow waves, background frequency, and more have been used. It is still unclear which of these wave parameters are essential, beyond peak-peak amplitude and duration. Wavelet parameters are very appropriate to AESD but need to be combined with other parameters to achieve desired levels of spike detection efficiency. Artificial Neural Network (ANN) and expert-system methods may have reached peak efficiency. Support Vector Machine (SVM) technology focuses on outliers rather than centroids of spike and nonspike data clusters and should improve AESD efficiency. An exemplary spike/nonspike database is suggested as a tool for assessing parameters and methods for AESD and is available in CSV or Matlab formats from the author at brainvue@gmail.com. Exploratory Data Analysis (EDA) is presented as a graphic method for finding better spike parameters and for the step-wise evaluation of the spike detection process.

Detection of focal epileptiform events in the EEG by spatio-temporal dipole clustering

OBJECTIVE

Methods for the detection of epileptiform events can be broadly divided into two main categories: temporal detection methods that exploit the EEG's temporal characteristics, and spatial detection methods that base detection on the results of an implicit or explicit source analysis. We describe how the framework of a spatial detection method was extended to improve its performance by including temporal information. This results in a method that provides (i) automated localization of an epileptogenic focus and (ii) detection of focal epileptiform events in an EEG recording. For the detection, only one threshold value needs to be set.

METHODS

The method comprises five consecutive steps: (1) dipole source analysis in a moving window, (2) automatic selection of focal brain activity, (3) dipole clustering to arrive at the identification of the epileptiform cluster, (4) derivation of a spatio-temporal template of the epileptiform activity, and (5) template matching. Routine EEG recordings from eight paediatric patients with focal epilepsy were labelled independently by two experts. The method was evaluated in terms of (i) ability to identify the epileptic focus, (ii) validity of the derived template, and (iii) detection performance. The clustering performance was evaluated using a leave-one-out cross validation. Detection performance was evaluated using Precision-Recall curves and compared to the performance of two temporal (mimetic and wavelet based) and one spatial (dipole analysis based) detection methods.

RESULTS

The method succeeded in identifying the epileptogenic focus in seven of the eight recordings. For these recordings, the mean distance between the epileptic focus estimated by the method and the region indicated by the labelling of the experts was 8mm. Except for two EEG recordings where the dipole clustering step failed, the derived template corresponded to the epileptiform activity marked by the experts. Over the eight EEGs, the method showed a mean sensitivity and selectivity of 92 and 77%, respectively.

CONCLUSIONS

The method allows automated localization of the epileptogenic focus and shows good agreement with the region indicated by the labelling of the experts. If the dipole clustering step is successful, the method allows a detection of the focal epileptiform events, and gave a detection performance comparable or better to that of the other methods.

SIGNIFICANCE

The identification and quantification of epileptiform events is of considerable importance in the diagnosis of epilepsy. Our method allows the automatic identification of the epileptic focus, which is of value in epilepsy surgery. The method can also be used as an offline exploration tool for focal EEG activity, displaying the dipole clusters and corresponding time series.

Higher Order Spectral (HOS) analysis of epileptic EEG signals

Epilepsy is a neurological condition, which affects the nervous system. Automatic seizure detection is very important in clinical practice and has to be achieved by analyzing the Electroencephalogram (EEG). Seizures are the clinical manifestations of excessive and hypersynchronous activity of the neurons in the cerebral cortex and represent one of the most frequent malfunctions of the human central nervous system. Therefore, the search for precursors and predictors of a seizure in the human EEG is of utmost clinical relevance and may even lead to a deeper understanding of the seizure generating mechanisms. In this paper, the normal, pre-ictal (background) and ictal (epileptic) EEG signals are studied using higher order spectra. HOS based measures are shown to be able to distinguish epileptic EEG from normal and background EEG with high confident level (p-value of less than 0.05).

Multiway analysis of epilepsy tensors

MOTIVATION

The success or failure of an epilepsy surgery depends greatly on the localization of epileptic focus (origin of a seizure). We address the problem of identification of a seizure origin through an analysis of ictal electroencephalogram (EEG), which is proven to be an effective standard in epileptic focus localization.

SUMMARY

With a goal of developing an automated and robust way of visual analysis of large amounts of EEG data, we propose a novel approach based on multiway models to study epilepsy seizure structure. Our contributions are 3-fold. First, we construct an Epilepsy Tensor with three modes, i.e. time samples, scales and electrodes, through wavelet analysis of multi-channel ictal EEG. Second, we demonstrate that multiway analysis techniques, in particular parallel factor analysis (PARAFAC), provide promising results in modeling the complex structure of an epilepsy seizure, localizing a seizure origin and extracting artifacts. Third, we introduce an approach for removing artifacts using multilinear subspace analysis and discuss its merits and drawbacks.

RESULTS

Ictal EEG analysis of 10 seizures from 7 patients are included in this study. Our results for 8 seizures match with clinical observations in terms of seizure origin and extracted artifacts. On the other hand, for 2 of the seizures, seizure localization is not achieved using an initial trial of PARAFAC modeling. In these cases, first, we apply an artifact removal method and subsequently apply the PARAFAC model on the epilepsy tensor from which potential artifacts have been removed. This method successfully identifies the seizure origin in both cases.

Adaptive neuro-fuzzy inference system for classification of EEG signals using wavelet coefficients

This paper describes the application of adaptive neuro-fuzzy inference system (ANFIS) model for classification of electroencephalogram (EEG) signals. Decision making was performed in two stages: feature extraction using the wavelet transform (WT) and the ANFIS trained with the backpropagation gradient descent method in combination with the least squares method. Five types of EEG signals were used as input patterns of the five ANFIS classifiers. To improve diagnostic accuracy, the sixth ANFIS classifier (combining ANFIS) was trained using the outputs of the five ANFIS classifiers as input data. The proposed ANFIS model combined the neural network adaptive capabilities and the fuzzy logic qualitative approach. Some conclusions concerning the saliency of features on classification of the EEG signals were obtained through analysis of the ANFIS. The performance of the ANFIS model was evaluated in terms of training performance and classification accuracies and the results confirmed that the proposed ANFIS model has potential in classifying the EEG signals.

Detection of Interictal Spikes and Artifactual Data Through Orthogonal Transformations

This study introduces an integrated algorithm based on the Walsh transform to detect interictal spikes and artifactual data in epileptic patients using recorded EEG data. The algorithm proposes a unique mathematical use of Walsh-transformed EEG signals to identify those criteria that best define the morphologic characteristics of interictal spikes. EEG recordings were accomplished using the 10-20 system interfaced with the Electrical Source Imaging System with 256 channels (ESI-256) for enhanced preprocessing and on-line monitoring and visualization. The merits of the algorithm are: (1) its computational simplicity; (2) its integrated design that identifies and localizes interictal spikes while automatically removing or discarding the presence of different artifacts such as electromyography, electrocardiography, and eye blinks; and (3) its potential implication to other types of EEG analysis, given the mathematical basis of this algorithm, which can be patterned or generalized to other brain dysfunctions. The mathematics that were applied here assumed a dual role, that of transforming EEG signals into mutually independent bases and in ascertaining quantitative measures for those morphologic characteristics deemed important in the identification process of interictal spikes. Clinical experiments involved 31 patients with focal epilepsy. EEG data collected from 10 of these patients were used initially in a training phase to ascertain the reliability of the observable and formulated features that were used in the spike detection process. Three EEG experts annotated spikes independently. On evaluation of the algorithm using the 21 remaining patients in the testing phase revealed a precision (positive predictive value) of 92% and a sensitivity of 82%. Based on the 20- to 30-minute epochs of continuous EEG recording per subject, the false detection rate is estimated at 1.8 per hour of continuous EEG. These are positive results that support further development of this algorithm for prolonged EEG recordings on ambulatory subjects and to serve as a support mechanism to the decisions made by EEG experts.

Automatic Detection of Epileptiform Events in EEG by a Three-Stage Procedure Based on Artificial Neural Networks

This paper introduces a three-stage procedure based on artificial neural networks for the automatic detection of epileptiform events (EVs) in a multichannel electroencephalogram (EEG) signal. In the first stage, two discrete perceptrons fed by six features are used to classify EEG peaks into three subgroups: 1) definite epileptiform transients (ETs); 2) definite non-ETs; and 3) possible ETs and possible non-ETs. The pre-classification done in the first stage not only reduces the computation time but also increases the overall detection performance of the procedure. In the second stage, the peaks falling into the third group are aimed to be separated from each other by a nonlinear artificial neural network that would function as a postclassifier whose input is a vector of 41 consecutive sample values obtained from each peak. Different networks, i.e., a backpropagation multilayer perceptron and two radial basis function networks trained by a hybrid method and a support vector method, respectively, are constructed as the postclassifier and then compared in terms of their classification performances. In the third stage, multichannel information is integrated into the system for contributing to the process of identifying an EV by the electroencephalographers (EEGers). After the integration of multichannel information, the overall performance of the system is determined with respect to EVs. Visual evaluation, by two EEGers, of 19 channel EEG records of 10 epileptic patients showed that the best performance is obtained with a radial basis support vector machine providing an average sensitivity of 89.1%, an average selectivity of 85.9%, and a false detection rate (per hour) of 7.5.

Automatic spike detection in EEG by a two-stage procedure based on support vector machines

In this study, we introduce a two-stage procedure based on support vector machines for the automatic detection of epileptic spikes in a multi-channel electroencephalographic signal. In the first stage, a modified non-linear digital filter is used as a pre-classifier to classify the peaks into two subgroups: (i) spikes and spike like non-spikes (ii) trivial non-spikes. The pre-classification done in the first stage not only reduces the computation time but also increases the overall detection performance of the procedure. In the second stage, the peaks falling into the first group are aimed to be separated from each other by a support vector machine that would function as a post-classifier. Visual evaluation, by two experts, of 19 channel EEG records of 7 epileptic patients showed that the best performance is obtained providing 90.3% sensitivity, 88.1% selectivity and 9.5% false detection rate.

Gaussian mixture models of ECoG signal features for improved detection of epileptic seizures

PURPOSE

To investigate the potential for improving the performance of the Osorio-Frei seizure detection algorithm (OFA) by incorporating multiple FIR filters operating in parallel and Gaussian mixture models (GMM) for ECoG features distributions, thus creating "hybrid" system.

METHODS

The "hybrid" algorithm decomposes the signal into four subbands, using wavelets, after which relevant features are extracted for each subband. Following these steps, multivariate GMM are developed for seizure and non-seizure states, using training segments. State classification is based on thresholding of the likelihood ratio of seizure vs. non-seizure data. Multiple comparisons are performed between this "hybrid" and a modified version of the OFA suitable for this purpose, using as indices false positives (FP), false negatives (FN) and speed of detection.

RESULTS

GMM improved speed of detection over the modified OFA at negligible FP levels. The average detection delay from expert visually placed electrographic onset over all seizures was reduced from 4.8 s for modified OFA to 1.8 s for GMM (p < 0.002) Individualized training by subject proved superior to group-based training.

CONCLUSIONS

This work introduces multi-feature extraction from ECoG signals together with use of Gaussian mixtures to model them, as tools to improve automated seizure detection. At the clinical level, this approach appears to increase warning time and with it the window during which safety measures and seizure blockage may be implemented, at an affordable computational cost and with negligible FP rate.

Interictal Spike Detection Using the Walsh Transform

The objective of this study was to evaluate the feasibility of using the Walsh transformation to detect interictal spikes in electroencephalogram (EEG) data. Walsh operators were designed to formulate characteristics drawn from experimental observation, as provided by medical experts. The merits of the algorithm are: 1) in decorrelating the data to form an orthogonal basis and 2) simplicity of implementation. EEG recordings were obtained at a sampling frequency of 500 Hz using standard 10-20 electrode placements. Independent sets of EEG data recorded on 18 patients with focal epilepsy were used to train and test the algorithm. Twenty to thirty minutes of recordings were obtained with each subject awake, supine, and at rest. Spikes were annotated independently by two EEG experts. On evaluation, the algorithm identified 110 out of 139 spikes identified by either expert (True Positives = 79%) and missed 29 spikes (False Negatives = 21%). Evaluation of the algorithm revealed a Precision (Positive Predictive Value) of 85% and a Sensitivity of 79%. The encouraging preliminary results support its further development for prolonged EEG recordings in ambulatory subjects. With these results, the false detection (FD) rate is estimated at 7.2 FD per hour of continuous EEG recording.

Automated interictal spike detection and source localization in magnetoencephalography using independent components analysis and spatio-temporal clustering

OBJECTIVE

Magnetoencephalography (MEG) dipole localization of epileptic spikes is useful in epilepsy surgery for mapping the extent of abnormal cortex and to focus intracranial electrodes. Visually analyzing large amounts of data produces fatigue and error. Most automated techniques are based on matching of interictal spike templates or predictive filtering of the data and do not explicitly include source localization as part of the analysis. This leads to poor sensitivity versus specificity characteristics. We describe a fully automated method that combines time-series analysis with source localization to detect clusters of focal neuronal current generators within the brain that produce interictal spike activity.

METHODS

We first use an ICA (independent components analysis) method to decompose the multichannel MEG data and identify those components that exhibit spike-like characteristics. From these detected spikes we then find those whose spatial topographies across the array are consistent with focal neural sources, and determine the foci of equivalent current dipoles and their associated time courses. We then perform a clustering of the localized dipoles based on distance metrics that takes into consideration both their locations and time courses. The final step of refinement consists of retaining only those clusters that are statistically significant. The average locations and time series from significant clusters comprise the final output of our method.

RESULTS AND SIGNIFICANCE

Data were processed from 4 patients with partial focal epilepsy. In all three subjects for whom surgical resection was performed, clusters were found in the vicinity of the resectioned area.

CONCLUSIONS

The presented procedure is promising and likely to be useful to the physician as a more sensitive, automated and objective method to help in the localization of the interictal spike zone of intractable partial seizures. The final output can be visually verified by neurologists in terms of both the location and distribution of the dipole clusters and their associated time series. Due to the clinical relevance and demonstrated promise of this method, further investigation of this approach is warranted.

Improvement in the performance of automated spike detection using dipole source features for artefact rejection

OBJECTIVE

We evaluated the use of an efficient dipole source algorithm to improve performance of automated spike detection by identifying false detections caused by artefacts.

METHODS

Automated spike detections were acquired from 26 patients undergoing prolonged electroencephalograph (EEG) monitoring. Data from 6 patients were used to develop the method and data from 20 patients were used to test the method. To provide a standard against which to evaluate the results, an electroencephalographer (EEGer) visually categorized all automated detections before the dipole models were calculated for all events. The event categories (as defined by the EEGer) were then combined with properties of the dipole model and features were identified that differentiated spike and artefact detections. The resulting method was then applied to the testing data set.

RESULTS

Residual variance and eccentricity of the dipole models differentiated artefact and spike detections. A separate set of rules defining eye blink artefact was also developed. The combined criteria removed a mean of 53.2% of artefact from the testing data set. Some spike detections (4.3%) were also lost.

CONCLUSIONS

The features of the dipole source of a detected event can be used to differentiate artefacts from spikes. This algorithm is computationally light and could be implemented on-line.

A multistage, multimethod approach for automatic detection and classification of epileptiform EEG

An efficient system for detection of epileptic activity in ambulatory electroencephalogram (EEG) must be sensitive to abnormalities while keeping the false-detection rate to a low level. Such requirements could be fulfilled neither by single stage nor by simple method strategy, due to the extreme variety of EEG morphologies and frequency of artifacts. The present study proposes a robust system that combines multiple signal-processing methods in a multistage scheme, integrating adaptive filtering, wavelet transform, artificial neural network, and expert system. The system consists of two main stages: a preliminary screening stage in which data are reduced significantly, followed by an analytical stage. Unlike most systems that merely focus on sharp transients, our system also takes into account slow waves. A nonlinear filter for separation of nonstationary and stationary EEG components is also developed in this paper. The system was evaluated on testing data from 81 patients, totaling more than 800 hours of recordings. 90.0% of the epileptic events were correctly detected. The detection rate of sharp transients was 98.0% and overall false-detection rate was 6.1%. We conclude that our system has good performance in detecting epileptiform activities and the multistage multimethod approach is an appropriate way of solving this problem.

Epileptic transient detection: wavelets and time-frequency approaches

This paper is aimed at presenting the two main classes of nonstationary signal transforms that are currently used to analyze and to characterize EEG observations. Time-scale methods, or wavelet transforms, allow a time versus duration analysis to be performed whereas time-frequency methods allow spectral contents to be analyzed as a function of time. These two types of transform are well suited to the study of changes either localized or progressive that may be observed in EEG signal dynamics and that sign the evolution of underlying physiological mechanisms. The potential interest of these methods in nonstationary signal representation is illustrated through several academic examples. Then, methods are applied on real EEG signals to solve problems such that the detection of interictal transient signals (like spikes or spike-waves) and the recognition of signatures during ictal periods.

Computer-aided spatial classification of epileptic spikes

The authors present a method that can be used to identify exemplar spikes from prolonged EEG recordings. To achieve this they have calculated single dipole source models for each automatically detected spike-like waveform. They used a dipole source algorithm that is computationally light and can be run on-line during EEG acquisition. Although a single dipole source model may not provide anatomically accurate information about the location of generators of all epileptiform abnormalities, it does provide a novel spatial parameter that may be useful in its own right. The authors use this spatial parameter and present the relative spatial density of the dipole locations in the form of three planar projections of the spherical model (a view from above, a view from the right, and a view from behind) and allow users to define the x-, y-, and z-coordinates of points of interest within the spherical model. They then present 10 example waveforms of events that have dipole source model locations that occur close to that seed coordinate. Overall, they found that this method performs very well for frequent events, but does not perform well for rare events or for diffuse EEG abnormalities.

Wavelet analysis of transient biomedical signals and its application to detection of epileptiform activity in the EEG

Wavelet based signal analysis provides a powerful new means for the analysis of nonstationary signals such as the human EEG. The properties of the discrete wavelet transform are reviewed in illustrated application examples. The continuous wavelet transform is shown to provide better detection and representation of isolated transients. An approach to extract features of edges and transients from the continuous wavelet transform is outlined. Matching pursuit is presented as a more general transform method that covers both transients and oscillation spindles. A statistical model for the continuous wavelet transform of background EEG is found. A spike detection system based on this background model is presented. The performance of this detection system has been assessed in a preliminary clinical study of 11 EEG recordings containing epileptiform activity and shown to have a sensitivity of 84% and a selectivity of 12%. The spatial context of epileptiform activity will be incorporated to improve system performance.

Real-time detection of epileptiform activity in the EEG: a blinded clinical trial

The aim of this study was to determine the performance of a PC-based system for real-time detection and topographical mapping of epileptiform activity (EA) in the EEG during routine clinical recordings. The system incorporates a mimetic stage to locate candidate spikes (including sharp-waves) followed by two expert-system-based stages, which utilize spatial and wide-temporal contextual information in deciding whether candidate events are epileptiform or not. The data comprised 521 consecutive routine clinical EEG recordings (173 hours). Performance was evaluated by comparison with three independent electroencephalographers (EEGers-I). A second group of two EEGers (EEGers-II) separately interpreted the spike topographical maps and, for EEGs categorized as containing only questionable EA by the detection system, reviewed 6 sec segments of raw EEG centered on each questionable event. Thirty-eight of the EEGs were considered to contain definite EA by at least two of EEGers-I. The false detection rate of the system was 0.41 per hour. The system was found to have a sensitivity of 76% and a selectivity of 41% for EEGs containing definite EA. However, it only missed detection of EA in 5% of the recordings. EEGers-II agreed with EEGers-I on the distribution (generalized, lateralized, focal, multifocal) of EA in 79% of cases. This is by far the largest clinical evaluation of computerized spike detection reported in the literature and the only one to apply this in routine clinical recordings. The false detection rate is the lowest ever reported, suggesting that this multi-stage rule-based system is a powerful and practical tool in clinical electroencephalography and long-term EEG monitoring.

Multifold features determine linear equation for automatic spike detection applying neural nin interictal ECoG

OBJECTIVE

A 3-layer detection procedure was designed including preselection applying TEMPLAS software, feature extraction and artificial neural networks to determine a fast, precise and highly selective spike algorithm.

METHODS

Ten intraoperative ECoG recordings of patients with temporal lobe epilepsy were computer-assisted and evaluated by 3 experts upon preselected events. For each event, 19 features were extracted, normalized and fed into a two-layer and 3-layer feedforward, back-propagate network. The weights of the 5 best individual two-layer networks of patients were averaged separately to derive a mean network, where weights were pruned, rounded off and the configuration approximated by a linear equation.

RESULTS

In addition. when investigating latency histograms, a method for multi-channel artefact detection and elimination of too close intra-channel events could be found. Out of several training trails only the mean network and the linear equation were able to generalize. In comparison with the results of 19 publications, the developed solution and the estimated overall detection rates (spikes: 81%; non-spikes: 99.3%) were found to be of high quality. The processing time is short, and therefore, the method can be used to initiate other measurements.

CONCLUSION

The developed solution is a fast, precise and highly selective spike detection method.

Automatic detection of seizures and spikes

The recording of seizures and spikes is of primary importance in the evaluation of epileptic patients. This is not always an easy process because these events can be rare and are usually unpredictable. Since the earliest days of computer analysis of the EEG, researchers have developed methods for the automatic detection of spikes and, more recently, of seizures. The problems are complex because spikes and seizures are not clearly defined and have extremely varied morphologies. Nevertheless, it has been possible to develop automatic detection methods that can be of great assistance during long-term monitoring of epileptic patients. No method is absolutely fail-safe and all require human validation, but they save a considerable amount of time in the interpretation of long recordings. Recent developments include detection of the patterns specific to newborns, and the possibility of warning a patient or observer that a seizure is starting.

An automated system for epileptogenic focus localization in the electroencephalogram

This paper describes an automated system for the detection and localization of foci of epileptiform activity in the EEG. The system detects sharp EEG transients (STs) in the process, but the emphasis is on epileptic focus localization. A combination of techniques involving signal processing, pattern recognition, and the expert rules of an experienced electroencephalographer, involving considerable spatiotemporal context information, is applied to multichannel EEG data. An overall emphasis on minimizing the number of false-positive sharp transient detections drives the system design. Tested on data from 13 subjects with epileptiform activity and 5 controls, all areas of focal epileptiform activity were detected by the system, although not all of the contributing foci were reported separately. Two false-positive foci were detected as well due to nonfocal spike activity and normal spike-like activity not present in the training set. The system detected 95.7% of the epileptiform events constituting the correctly detected foci, with a false detection rate of 11.1%.

Detection of spikes with artificial neural networks using raw EEG

Artificial neural networks (ANN) using raw electroencephalogram (EEG) data were developed and tested off-line to detect transient epileptiform discharges (spike and spike/wave) and EMG activity in an ongoing EEG. In the present study, a feedforward ANN with a variable number of input and hidden layer units and two output units was used to optimize the detection system. The ANN system was trained and tested with the backpropagation algorithm using a large data set of exemplars. The effects of different EEG time windows and the number of hidden layer neurons were examined using rigorous statistical tests for optimum detection sensitivity and selectivity. The best ANN configuration occurred with an input time window of 150 msec (30 input units) and six hidden layer neurons. This input interval contained information on the wave component of the epileptiform discharge which improved detection. Two-dimensional receiver operating curves were developed to define the optimum threshold parameters for best detection. Comparison with previous networks using raw EEG showed improvement in both sensitivity and selectivity. This study showed that raw EEG can be successfully used to train ANNs to detect epileptogenic discharges with a high success rate without resorting to experimenter-selected parameters which may limit the efficiency of the system.

Multireference adaptive noise canceling applied to the EEG

The technique of multireference adaptive noise canceling (MRANC) is applied to enhance transient nonstationarities in the electroeancephalogram (EEG), with the adaptation implemented by means of a multilayer-perception artificial neural network (ANN). The method was applied to recorded EEG segments and the performance on documented nonstationarities recorded. The results show that the neural network (nonlinear) gives an improvement in performance (i.e., signal-to-noise ratio (SNR) of the nonstationarities) compared to a linear implementation of MRANC. In both cases an improvement in the SNR was obtained. The advantage of the spatial filtering aspect of MRANC is highlighted when the performance of MRANC is compared to that of the inverse auto-regressive filtering of the EEG, a purely temporal filter.

A patient-specific algorithm for the detection of seizure onset in long-term EEG monitoring: possible use as a warning device

During long-term electroencephalogram (EEG) monitoring of epileptic patients, a seizure warning system would allow patients and observers to take appropriate precautions. It would also allow observers to interact with patients early during the seizure, thus revealing clinically useful information. We designed patient-specific classifiers to detect seizure onsets. After a seizure and some nonseizure data are recorded in a patient, they are used to train a classifier. In subsequent monitoring sessions, EEG patterns have to pass this classifier to determine if a seizure onset occurs. If it does, an alarm is triggered. Extreme care has been taken to ensure a low false-alarm rate, since a high false-alarm rate would render the system ineffective. Features were extracted from the time and frequency domains and a modified nearest-neighbor (NN) classifier was used. The system reached an onset detection rate of 100% with an average delay of 9.35 a after onset. The average false-alarm rate was only 0.02/h. The method was evaluated in 12 patients with a total of 47 seizures. Results indicate that the system is effective and reasonably reliable. Computation load has been kept to a minimum so that real-time processing is possible.

Multiresolution decomposition of non-stationary EEG signals: a preliminary study

Wavelet representation is a recent development in the analysis of non-stationary signals. Its possibilities for use in the description of time-frequency characteristics of both transients in spontaneous EEG and time-varying rhythms in event related brain activity are explored here. By way of illustration, multiresolution decompositions of a wide variety of EEG transients are carried out in this work, including spike-and-waves, single spikes, sharp waves, blink artifacts, frontal intermittent rhythmic delta activity (FIRDA) and paroxysmal delta activity. Also, the application of the wavelet representation to study related spectra perturbations is illustrated with data from psychophysical experiments on the perception of image motion. The results demonstrate the capabilities of the wavelet transform, as an alternative to the Fourier transform, for the representation and analysis of non-stationary EEG signals.

Wavelet analysis of EEG for three-dimensional mapping of epileptic events

This paper is aimed at understanding epileptic patient disorders through the analysis of surface electroencephalograms (EEG). It deals with the detection of spikes or spike-waves based on a nonorthogonal wavelet transform. A multilevel structure is described that locates the temporal segments where abnormal events occur. These events are then visually interpreted by means of a 3D mapping technique. This 3D display makes use of a ray tracing scheme and combines both the functional (the EEG but also its wavelet representation) and the morphological data (acquired from computed tomography [CT] or magnetic resonance imaging [MRI] devices). The results show that a significant reduction of the clinical workload is obtained while the most important episodes are better reviewed and analyzed.

Practical detection of epileptiform discharges (EDs) in the EEG using an artificial neural network: a comparison of raw and parameterized EEG data

We have developed and tested "off-line" an artificial neural network (ANN) that successfully detects epileptiform discharges (EDs) when trained on EEG records marked by an electroencephalographer (EEGer). The system was trained on both parameterized and raw EEG data and can process 49 channels of EEG data in real time on an 80486/33 MHz personal computer, making it capable of processing EEG on-line in long-term monitoring units. Our detector consists of 2 stages: (1) a threshold detector identifies candidate EDs in 4-channel bipolar chains within the recording montage, parameterizes them and then passes these data to the second stage; (2) a 3-layer feed-forward ANN decides if a candidate wave form is an ED. The intersection of detector sensitivity and selectivity curves, or crossover threshold, for 10 patients from our Epilepsy Monitoring Unit occurred at 73% for parameterized EEG data and at 46% for "raw" EEG data. The ANN could be adapted to different EEGers' styles by changing the ANN output threshold for accepting candidate wave forms as EDs. In this "proof of principle" study the detector was trained on EEGs from 10 Johns Hopkins Hospital Epilepsy Monitoring Unit (JHH EMU) patients. We used different EEGs from the same patients for testing. Current testing should demonstrate that the ANN detector can generalize to previously "unseen" patients. This study shows that ANNs offer a practical solution for automated, real time ED detection that uses, standard, inexpensive computers, is easily adjustable to individual EEGer style and can produce sensitivities and selectivities similar to those of EEGers.

A multistage system to detect epileptiform activity in the EEG

A PC-based system has been developed to automatically detect epileptiform activity in sixteen-channel bipolar EEG's. The system consists of three stages: data collection, feature extraction, and event detection. The feature extractor employs a mimetic approach to detect candidate epileptiform transients on individual channels, while an expert system is used to detect focal and nonfocal multichannel epileptiform events. Considerable use of spatial and temporal contextual information present in the EEG aids both in the detection of epileptiform events and in the rejection of artifacts and background activity as events. Classification of events as definite or probable overcomes, to some extent, the problem of maintaining high detection rates while eliminating false detections. So far, the system has only been evaluated on development data but, although this does not provide a true measure of performance, the results are nevertheless impressive. Data from 11 patients, totaling 180 minutes of sixteen-channel bipolar EEG's, have been analyzed. A total of 45-71% (average 58%) of epileptiform events reported by the human expert in any EEG were detected as definite with no false detections (i.e., 100% selectivity) and 60-100% (average 80%) as either definite or probable but at the expense of up to nine false detections per hour. Importantly, the highest detection rates were achieved on EEG's containing little epileptiform activity and no false detections were made on normal EEG's.

An automated seizure monitoring system for patients with indwelling recording electrodes

An automated monitoring system has been developed to record from indwelling electrode arrays in patients undergoing evaluation for surgical treatment of intractable seizures. The functional aspects of this system's design are discussed and the range of electrical seizure patterns, other epileptiform events and artifacts that the system must handle are described. The system includes a flowing graphics image of as many as 32 channels of real-time ECoG, automatic seizure detection, recording of the ECoG for up to 6 min prior to seizure onset, and EEG for 792 clinical and subclinical seizures during 1578 h of monitoring with an artifact rate of 28% for all events recorded. However, system performance was judged upon the 86% accuracy for detection and recording of patient-specific seizure patterns (multiple seizures with the same pattern counted as 1) with 1.26 brief spike bursts and 0.67 artifacts/h.