Automatic seizure detection in long-term scalp EEG using an adaptive thresholding technique: A validation study for clinical routine

ObjectiveIn a previous study we proposed a robust method for automatic seizure detection in scalp EEG recordings. The goal of the current study was to validate an improved algorithm in a much larger group of patients in order to show its general applicability in clinical routine.MethodsFor the detection of seizures we developed an algorithm based on Short Time Fourier Transform, calculating the integrated power in the frequency band 2.5–12 Hz for a multi-channel seizure detection montage referenced against the average of Fz-Cz-Pz. For identification of seizures an adaptive thresholding technique was applied. Complete data sets of each patient were used for analyses for a fixed set of parameters.Results159 patients (117 temporal-lobe epilepsies (TLE), 35 extra-temporal lobe epilepsies (ETLE), 7 other) were included with a total of 25,278 h of EEG data, 794 seizures were analyzed. The sensitivity was 87.3% and number of false detections per hour (FpH) was 0.22/h. The sensitivity for TLE patients was 89.9% and FpH = 0.19/h; for ETLE patients sensitivity was 77.4% and FpH = 0.25/h.ConclusionsThe seizure detection algorithm provided high values for sensitivity and selectivity for unselected large EEG data sets without a priori assumptions of seizure patterns.SignificanceThe algorithm is a valuable tool for fast and effective screening of long-term scalp EEG recordings.     

Application of a multivariate seizure detection and prediction method to non-invasive and intracranial long-term EEG recordings.

OBJECTIVE: Retrospective evaluation and comparison of performances of a multivariate method for seizure detection and prediction on simultaneous long-term EEG recordings from scalp and intracranial electrodes. METHODS: Two multivariate techniques based on simulated leaky integrate-and-fire neurons were investigated in order to detect and predict seizures. Both methods were applied and assessed on 423h of EEG and 26 seizures in total, recorded simultaneously from the scalp and intracranially continuously over several days from six patients with pharmacorefractory epilepsy. RESULTS: Features generated from simultaneous scalp and intracranial EEG data showed a similar dynamical behavior. Significant performances with sensitivities of up to 73%/62% for scalp/invasive EEG recordings given an upper limit of 0.15 false detections per hour were obtained. Up to 59%/50% of all seizures could be predicted from scalp/invasive EEG, given a maximum number of 0.15 false predictions per hour. A tendency to better performances for scalp EEG was obtained for the detection algorithm. CONCLUSIONS: The investigated methods originally developed for non-invasive EEG were successfully applied to intracranial EEG. Especially, concerning seizure detection the method shows a promising performance which is appropriate for practical applications in EEG monitoring. Concerning seizure prediction a significant prediction performance is indicated and a modification of the method is suggested. SIGNIFICANCE: This study evaluates simultaneously recorded non-invasive and intracranial continuous long-term EEG data with respect to seizure detection and seizure prediction for the first time.     

A system to detect the onset of epileptic seizures in scalp EEG.

OBJECTIVE: A new method for automatic seizure detection and onset warning is proposed. The system is based on determining the seizure probability of a section of EEG. Operation features a user-tuneable threshold to exploit the trade-off between sensitivity and detection delay and an acceptable false detection rate. METHODS: The system was designed using 652 h of scalp EEG, including 126 seizures in 28 patients. Wavelet decomposition, feature extraction and data segmentation were employed to compute the a priori probabilities required for the Bayesian formulation used in training, testing and operation. RESULTS: Results based on the analysis of separate testing data (360 h of scalp EEG, including 69 seizures in 16 patients) initially show a sensitivity of 77.9%, a false detection rate of 0.86/h and a median detection delay of 9.8 s. Results after use of the tuning mechanism show a sensitivity of 76.0%, a false detection rate of 0.34/h and a median detection delay of 10 s. Missed seizures are characterized mainly by subtle or focal activity, mixed frequencies, short duration or some combination of these traits. False detections are mainly caused by short bursts of rhythmic activity, rapid eye blinking and EMG artifact caused by chewing. Evaluation of the traditional seizure detection method of using both data sets shows a sensitivity of 50.1%, a false detection rate of 0.5/h and a median detection delay of 14.3 s. CONCLUSIONS: The system performed well enough to be considered for use within a clinical setting. In patients having an unacceptable level of false detection, the tuning mechanism provided an important reduction in false detections with minimal loss of detection sensitivity and detection delay. SIGNIFICANCE: During prolonged EEG monitoring of epileptic patients, the continuous recording may be marked where seizures are likely to have taken place. Several methods of automatic seizure detection exist, but few can operate as an on-line seizure alert system. We propose a seizure detection system that can alert medical staff to the onset of a seizure and hence improve clinical diagnosis.     

Automatic multimodal detection for long-term seizure documentation in epilepsy

Objective

This study investigated sensitivity and false detection rate of a multimodal automatic seizure detection algorithm and the applicability to reduced electrode montages for long-term seizure documentation in epilepsy patients.

Long-term EEG-video-audio monitoring: computer detection of focal EEG seizure patterns.

Twelve individuals with medically refractory partial seizures had undergone EEG-video-audio (EVA) monitoring over 1-15 (mean 10.5) days. We selectively reexamined available 15-channel EEGs (video-cassettes) totaling 461 h and containing 253 EEG focal seizures. Computer analysis (CA) of these bipolar records was performed using a mimetic method of seizure detection at 6 successive computer settings. We determined the computer parameters at which this method correctly detected a reasonably large percentage of seizures (81.42%) while generating an acceptable rate of false positive results (5.38/h). These parameters were adopted as the default setting for identifying focal EEG seizure patterns in all subsequent long-term bipolar scalp and sphenoidal recordings. Factors hindering or facilitating automatic seizure identification are discussed. It is concluded that on-line computer detection of focal EEG seizure patterns by this method offers a satisfactory alternative to and represents a distinct improvement over the extremely time consuming and fatiguing off-line fast visual review (FVR). Combining CA with seizure signaling (SS) by the patients and other observers increased the correct detections to 85.38% CA is best used in conjunction with SS.     

Using simulated neuronal cell models for detection of epileptic seizures in foramen ovale and scalp EEG.

OBJECTIVE: To demonstrate a novel approach for real-time and automatic detection of epileptic seizures in EEG recorded with foramen ovale (Fov) or scalp electrodes. METHODS: Our seizure detection method is based on simulated leaky integrate and fire units (LIFU), which are classical simple neuronal cell models. The LIFUs are connected to a signal preprocessing stage and increase their spiking rates in response to rhythmic and synchronous EEG signals as typically occur at the onset and during seizures. RESULTS: We analyzed 22 short-term (10+/-3 min) and 4 long-term (18+/-7 h) Fov or scalp EEGs of 10 patients with drug resistant partial epilepsy. Seizures (n=36) were marked by increases of the LIFUs spiking rates above a preset threshold. The durations of increased spiking rates due to seizures were always longer than 10 s (36+/-21 s) and allowed separation from artifacts, which caused only short durations (1.2+/-0.6 s) of high spiking rates. The LIFUs correctly detected all the seizures and produced no false alarms. In the long term Fov EEGs seizure detection occurred before the onset of clinical signs (41+/-22 s). CONCLUSIONS: By using simulated neuronal cell models it is possible to automatically detect epileptic seizures in scalp and Fov EEG with high sensitivity and specificity.     

Seizure state detection of temporal lobe seizures by autoregressive spectral analysis of scalp EEG

ObjectiveTo investigate a novel application of autoregression (AR) spectral techniques for seizure detection from scalp EEG.MethodsEEGs were recorded from twelve patients with left temporal lobe epilepsy. The Burg maximum entropy AR method was applied to the signals from four electrodes near the epileptic focus for each patient, and the AR spectra were parameterized based on scalp EEG features described by a neurologist, thus mimicking clinical seizure identification. The parameters measured spectral peak power, sharpness, and location in a delta/low theta frequency range. An optimized nonlinear seizure detection index, which accounted for spatial and temporal persistence of behavior, was then calculated.ResultsPerformance was optimized using recordings from two patients (315 h, 18 seizures). For the remaining 10 patients (1624 h, 83 seizures) results are presented as a Receiver Operating Characteristic graph, yielding an overall event-based true positive rate of 91.57% and epoch-based false positive rate of 3.97%.ConclusionsPerformance of the AR seizure identification method is comparable to other approaches. Techniques such as artifact removal are expected to improve performance.SignificanceThere is a real potential for this seizure detection method to be of practical clinical use in long-term monitoring.     

Detecting temporal lobe seizures from scalp EEG recordings: a comparison of various features.

OBJECTIVE: Sixteen different features are evaluated in their potential ability to detect seizures from scalp EEG recordings containing temporal lobe (TL) seizures. Features include spectral measures, non-linear methods (e.g. zero-crossings), phase synchronization and the recently introduced Brain Symmetry Index (BSI). Besides an individual comparison, several combinations of features are evaluated as well in their potential ability to detect TL seizures. METHODS: Sixteen long-term scalp EEG recordings, containing TL seizures from patients suffering from temporal lobe epilepsy (TLE), were analyzed. For each EEG, all 16 features were determined for successive 10s epochs of the recording. All epochs were labeled by experts for the presence or absence of seizure activity. In addition, triplet combinations of various features were evaluated using pattern recognition tools. Final performance was evaluated by the sensitivity and specificity (False Alarm Rate (FAR)), using ROC curves. RESULTS: In those TL seizures characterized by unilateral epileptiform discharges, the BSI was the best single feature. Except for one low-voltage EEG with many artifacts, the sensitivity found ranged from 0.55 to 0.90 at a FAR of approximately 1/h. Using three features increased the sensitivity to 0.77-0.97. In patients with bilateral electroencephalographic changes, the single best feature most often found was a measure for the number of minima and maxima (mmax) in the recording, yielding sensitivities of approximately 0.30-0.96 at FAR approximately 1/h. Using three features increased the sensitivity to 0.38-0.99, at the same FAR. In various recordings, it was even possible to obtain sensitivities of 0.70-0.95 at a FAR = 0. CONCLUSIONS: The Brain Symmetry Index is the most relevant individual feature to detect electroencephalographic seizure activity in TLE with unilateral epileptiform discharges. In patients with bilateral discharges, mmax performs best. Using a triplet of features significantly improves the performance of the detector. SIGNIFICANCE: Improved seizure detection can improve patient care in both the epilepsy monitoring unit and the intensive care unit.     

Improvement in seizure detection performance by automatic adaptation to the EEG of each patient

An important problem in the use of automatic seizure detection during long-term epilepsy monitoring is that false detections can be very frequent, often because a paroxysmal but non-epileptiform pattern occurs repeatedly in a particular patient. We therefore introduce a method to reduce such patient-specific false seizure detections. The program “learns” about the false detections occurring in the first day of a prolonged monitoring session and attempts to eliminate similar patterns occurring during the remainder of the session. This method was evaluated in 20 patients having particularly high false detection rates. Seventy EEG sessions from 10 patients with scalp electrodes and 64 sessions from 10 patients with depth electrodes, covering a total of 2600 h were used in the evaluation. False detections were reduced by 61% (50% in scalp recordings and 71% in depth recordings), with only a 5% probability of losing true seizures. The average false detection rate in these patients fell from 3.25/h to 1.26/h. This significant reduction in false detections could also lead to lower detection thresholds and consequently to the detection of more true seizures.     

Wavelet based automatic seizure detection in intracerebral electroencephalogram.

BACKGROUND: Automatic seizure detection is often used during long-term monitoring, and is particularly important during intracerebral investigations. Existing methods make many false detections, particularly in intracerebral electroencephalogram (EEG) because of frequent large amplitude rhythmic activity bursts that are non-epileptiform. OBJECTIVE: To develop a seizure detection method for intracerebral monitoring that is as sensitive as existing methods but has fewer false detections. METHODS: To capture the rhythmic nature of seizure discharges, we developed a wavelet-based method, examining how different frequency ranges fluctuate compared to the background. In particular, the system remembers rhythmic bursts occurring commonly in the background to avoid detecting them as seizures. RESULTS: The method was evaluated on test data from 11 patients, including 229 h and 66 seizures, and its performance compared to the method of Gotman (Electroencephalogr clin Neurophysiol 76 (1990) 317). Detection sensitivity was unchanged at close to 90%, but false detections were reduced from 2.4 to 0.3/h. CONCLUSIONS: Perfect sensitivity is unlikely because the morphology of seizure discharges is so variable. Nevertheless, the 87% sensitivity obtained in the combined training and testing data is quite high. We reduced the average false alarm rate to one per 3 h of recording, or 6 per 24-h period. Given how rapidly one can decide visually that a detection is erroneous, false detections should not cause any burden to the reviewer. SIGNIFICANCE: In intracerebral EEG it is possible to detect seizures automatically with high sensitivity and high specificity.     

Seizure detection using scalp‐EEG

Scalp electroencephalography (EEG)–based seizure‐detection algorithms applied in a clinical setting should detect a broad range of different seizures with high sensitivity and selectivity and should be easy to use with identical parameter settings for all patients. Available algorithms provide sensitivities between 75% and 90%. EEG seizure patterns with short duration, low amplitude, circumscribed focal activity, high frequency, and unusual morphology as well as EEG seizure patterns obscured by artifacts are generally difficult to detect. Therefore, detection algorithms generally perform worse on seizures of extratemporal origin as compared to those of temporal lobe origin. Specificity (false‐positive alarms) varies between 0.1 and 5 per hour. Low false‐positive alarm rates are of critical importance for acceptance of algorithms in a clinical setting. Reasons for false‐positive alarms include physiological and pathological interictal EEG activities as well as various artifacts. To achieve a stable, reproducible performance (especially concerning specificity), algorithms need to be tested and validated on a large amount of EEG data comprising a complete temporal assessment of all interictal EEG. Patient‐specific algorithms can further improve sensitivity and specificity but need parameter adjustments and training for individual patients. Seizure alarm systems need to provide on‐line calculation with short detection delays in the order of few seconds. Scalp‐EEG–based seizure detection systems can be helpful in an everyday clinical setting in the epilepsy monitoring unit, but at the current stage cannot replace continuous supervision of patients and complete visual review of the acquired data by specially trained personnel. In an outpatient setting, application of scalp‐EEG–based seizure‐detection systems is limited because patients won't tolerate wearing widespread EEG electrode arrays for long periods in everyday life. Recently developed subcutaneous EEG electrodes may offer a solution in this respect.     

Automatic seizure detection using diffusion distance and BLDA in intracranial EEG

Approximately 1% of the world's population suffers from epilepsy. An automatic seizure detection system is of great significance in the monitoring and diagnosis of epilepsy. In this paper, a novel method is proposed for automatic seizure detection in intracranial EEG recordings. The EEG recordings are divided into 4-s epochs, and then wavelet decomposition with five scales is performed to the EEG epochs. Detail signals at scales 3, 4, and 5 are selected to form a signal distribution. The diffusion distances are extracted as features, and Bayesian linear discriminant analysis (BLDA) is used as the classifier. A total of 193.75 h of intracranial EEG recordings from 21 patients having 87 seizures are employed to evaluate the system, and the average sensitivity of 94.99%, specificity of 98.74%, and false-detection rate of 0.24/h are achieved. The seizure detection system based on diffusion distance yields a high sensitivity as well as a low false-detection rate for long-term EEG recordings.     

Seizure detection method based on fractal dimension and gradient boosting

Automatic seizure detection technology is necessary and crucial for the long-term electroencephalography (EEG) monitoring of patients with epilepsy. This article presents a patient-specific method for the detection of epileptic seizures. The fractal dimensions of preprocessed multichannel EEG were firstly estimated using a k-nearest neighbor algorithm. Then, the feature vector constructed for each epoch was fed into a trained gradient boosting classifier. After a series of postprocessing, including smoothing, threshold processing, collar operation, and union of seizure detections in a short time interval, a binary decision was made to determine whether the epoch belonged to seizure status or not. Both the epoch-based and event-based assessments were used for the performance evaluation of this method on the EEG data of 21 patients from the Freiburg dataset. An average epoch-based sensitivity of 91.01% and a specificity of 95.77% were achieved. For the event-based assessment, this method obtained an average sensitivity of 94.05%, with a false detection rate of 0.27/h.     

Accuracy of seizure detection using abbreviated EEG during polysomnography.

The purpose of this study was to determine the validity of abbreviated EEG montages for seizure detection during polysomnography. Three electroencephalographers reviewed files containing seizures or nonepileptic events using 8- and 18-channel montages. Files were rated as to whether they contained seizures and assigned a "probability of seizure" score from 0% to 100% reflecting the confidence that it was a seizure. Readers then localized seizures as temporal, frontal, parieto-occipital, or nonlocalized and provided a probability of correct localization with 0% to 100% confidence. Data were analyzed using the Adjusted McNemar Test method of Obochuwski. The probability of seizure score was measured using the receiver operating characteristic curve. Observed agreement was 78% and 84% for 8- and 18-channel montages, respectively. Readers were better able to distinguish seizures from nonepileptic events using the 18-channel montage (P = 0.004). Seizures localized to the temporal and parieto-occipital regions were more likely to be correctly identified and localized. Readers were able to correctly localize 27% and 49% of seizures using the 8- and 18-channel montages, respectively (P < 0.001). Abbreviated EEG montages are inadequate in the differentiation of seizures and nonepileptic events arising from sleep during polysomnography. This seems to be particularly true in frontal lobe epilepsy.     

Seizure detection using heart rate variability: A prospective validation study

Although several validated seizure detection algorithms are available for convulsive seizures, detection of nonconvulsive seizures remains challenging. In this phase 2 study, we have validated a predefined seizure detection algorithm based on heart rate variability (HRV) using patient-specific cutoff values. The validation data set was independent from the previously published data set. Electrocardiography (ECG) was recorded using a wearable device (ePatch) in prospectively recruited patients. The diagnostic gold standard was inferred from video–EEG monitoring. Because HRV-based seizure detection is suitable only for patients with marked ictal autonomic changes, we defined responders as the patients who had a>50 beats/min ictal change in heart rate. Eleven of the 19 included patients with seizures (57.9%) fulfilled this criterion. In this group, the algorithm detected 20 of the 23 seizures (sensitivity: 87.0%). The algorithm detected all but one of the 10 recorded convulsive seizures and all of the 8 focal impaired awareness seizures, and it missed 2 of the 4 focal aware seizures. The median sensitivity per patient was 100% (in nine patients all seizures were detected). The false alarm rate was 0.9/24 h (0.22/night). Our results suggest that HRV-based seizure detection has high performance in patients with marked autonomic changes.     

Nearest neighbor phase synchronization as a measure to detect seizure activity from scalp EEG recordings.

The author presents results from the application of a particular measure for synchronization between brain areas (i.e., phase synchronization) in its behavior to detect epileptic seizure activity from scalp EEG recordings. The primary motivation for the current study was to contribute to the development of physiologic measures that both transform the EEG to a visual domain that allows a more intuitive interpretation of the interictal and ictal EEG and allows automated analysis, both relevant for real-time monitoring. EEGs from 16 patients experiencing temporal lobe and generalized seizures were analyzed. Nearest neighbor phase synchronization (NNPS) values for several frequency bands were determined. Additional analysis of the NNPS in the delta band, using different thresholds, allows construction of receiver operating characteristics (ROC) curves for each EEG analyzed. The common value for the sensitivity and the specificity, Q*, was used as a measure for the test accuracy, with values of Q* near 1.0, indicating ROC curves with sensitivity and specificity both approaching 1.0. It was found that Q* = 0.48 to 0.87, depending on the EEG analyzed, indicating that the proposed method allows seizure detection in a significant portion of the EEGs studied. Nearest neighbor phase synchronization was typically increased during seizure activity and seems to be a promising method to detect seizure activity from scalp EEG recordings. The proposed visualization allows an intuitive interpretation of the EEG and may assist in real-time monitoring.     

A fuzzy rule-based system for epileptic seizure detection in intracranial EEG

ObjectiveWe present a method for automatic detection of seizures in intracranial EEG recordings from patients suffering from medically intractable focal epilepsy.MethodsWe designed a fuzzy rule-based seizure detection system based on knowledge obtained from experts’ reasoning. Temporal, spectral, and complexity features were extracted from IEEG segments, and spatio-temporally integrated using the fuzzy rule-based system for seizure detection. A total of 302.7 h of intracranial EEG recordings from 21 patients having 78 seizures was used for evaluation of the system.ResultsThe system yielded a sensitivity of 98.7%, a false detection rate of 0.27/h, and an average detection latency of 11 s. There was only one missed seizure. Most of false detections were caused by high-amplitude rhythmic activities. The results from the system correlate well with those from expert visual analysis.ConclusionThe fuzzy rule-based seizure detection system enabled us to deal with imprecise boundaries between interictal and ictal IEEG patterns.SignificanceThis system may serve as a good seizure detection tool with high sensitivity and low false detection rate for monitoring long-term IEEG.     

A multistage knowledge-based system for EEG seizure detection in newborn infants

OBJECTIVE: Automatic seizure detection has attracted attention as a method to obtain valuable information concerning the duration, timing, and frequency of seizures. Methods currently used to detect EEG seizures in adults show high false detection rates in neonates because they lack information about specific age-dependent features of normal and pathological EEG and artifacts. This paper describes a novel multistage knowledge-based seizure detection system for newborn infants to identify and classify normal and pathological newborn EEGs as well as seizures with a reduced false detection rate. METHODS: We developed the system in a way to make comprehensive use of spatial and temporal contextual information obtained from multichannel EEGs. The system development consists of six major stages: (i) EEG data collection and bandpass filtering; (ii) automatic artifact detection; (iii) feature extraction from segments of non-seizure and seizure activities; (iv) feature selection via the relevance and redundancy analysis; (v) EEG classification and pattern recognition using a trained multilayer back-propagation neural network; and (v) knowledge-based decision-making to examine each of possible EEG patterns from a multi-channel perspective. The system was developed and tested with the EEG recordings of 10 newborns aged between 39 and 42 weeks. RESULTS: The overall sensitivity, selectivity, and average detection rate of the system were 74%, 70.1%, and 79.7%, respectively. The average false detection of 1.55/h was also achieved by the system with a feature reduction up to 80%. CONCLUSIONS: The expert rule-based decision-making subsystem accompanying the classifier helped to reduce the false detection rate, reject a wide variety of artifacts, and discriminate various patterns of EEG. SIGNIFICANCE: This paper may serve as a guide for the selection of discriminative features to improve the accuracy of conventional seizure detection systems for routine clinical EEG interpretation and brain activity monitoring in newborns especially those hospitalized in the neonatal intensive care units.     

The effect of reducing EEG electrode number on the visual interpretation of the human expert for neonatal seizure detection

Objectives

To measure changes in the visual interpretation of the EEG by the human expert for neonatal seizure detection when reducing the number of recording electrodes.