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

A fast machine learning approach to facilitate the detection of interictal epileptiform discharges in the scalp electroencephalogram

BACKGROUND

Finding interictal epileptiform discharges (IEDs) in the EEG is a part of diagnosing epilepsy. Automated software for annotating EEGs of patients with suspected epilepsy can therefore help with reaching a diagnosis. A large amount of data is required for training and evaluating an effective IED detection system. IEDs occur infrequently in the most patients' EEG, therefore, interictal EEG recordings contain mostly background waveforms.

NEW METHOD

As the first step to detect IEDs, we propose a machine learning technique eliminating most EEG background data using an ensemble of simple fast classifiers based on several EEG features. This could save computation time for an IED detection method, allowing the remaining waveforms to be classified by more computationally intensive methods. We consider several efficient features and reject background by applying thresholds on them in consecutive steps.

RESULTS

We applied the proposed algorithm on a dataset of 156 EEGs (93 and 63 with and without IEDs, respectively). We were able to eliminate 78% of background waveforms while retaining 97% of IEDs on our cross-validated dataset.

COMPARISON WITH EXISTING METHODS

We applied support vector machine, k-nearest neighbours, and random forest classifiers to detect IEDs with and without initial background rejection. Results show that rejecting background by our proposed method speeds up the overall classification by a factor ranging from 1.8 to 4.7 for the considered classifiers.

CONCLUSIONS

The proposed method successfully reduces computation time of an IED detection system. Therefore, it is beneficial in speeding up IED detection especially when utilizing large EEG datasets.

Fast EEG spike detection via eigenvalue analysis and clustering of spatial amplitude distribution

OBJECTIVE

In the current study, we tested a proposed method for fast spike detection in electroencephalography (EEG).

APPROACH

We performed eigenvalue analysis in two-dimensional space spanned by gradients calculated from two neighboring samples to detect high-amplitude negative peaks. We extracted the spike candidates by imposing restrictions on parameters regarding spike shape and eigenvalues reflecting detection characteristics of individual medical doctors. We subsequently performed clustering, classifying detected peaks by considering the amplitude distribution at 19 scalp electrodes. Clusters with a small number of candidates were excluded. We then defined a score for eliminating spike candidates for which the pattern of detected electrodes differed from the overall pattern in a cluster. Spikes were detected by setting the score threshold.

MAIN RESULTS

Based on visual inspection by a psychiatrist experienced in EEG, we evaluated the proposed method using two statistical measures of precision and recall with respect to detection performance. We found that precision and recall exhibited a trade-off relationship. The average recall value was 0.708 in eight subjects with the score threshold that maximized the F-measure, with 58.6  ±  36.2 spikes per subject. Under this condition, the average precision was 0.390, corresponding to a false positive rate 2.09 times higher than the true positive rate. Analysis of the required processing time revealed that, using a general-purpose computer, our method could be used to perform spike detection in 12.1% of the recording time. The process of narrowing down spike candidates based on shape occupied most of the processing time.

SIGNIFICANCE

Although the average recall value was comparable with that of other studies, the proposed method significantly shortened the processing time.

Fast spike detection in EEG using eigenvalue analysis and clustering of spatial amplitude distribution

In the current study, we tested a proposed method for fast spike detection using a general-purpose computer. First, we performed eigenvalue analysis using a gradient calculated from two neighboring samples to detect high-amplitude negative peaks. Clustering was performed to classify detected peaks by considering amplitude distribution at scalp electrodes. Negative peaks were scored by considering electrodes in the detection process and the cluster to which each peak belonged. Spikes were detected using two parameters: score threshold, and the number of clusters. We then used precision and recall to eliminate overestimation of the performance of the method. The results revealed a tradeoff between precision and recall. Recall showed a maximum average value of 0.90 in two subjects. In contrast, average precision was 0.21, and the false positive rate was almost four times higher than the true positive rate on the condition that 64 and 54 spikes were included in two subjects. Analysis of required processing time revealed that our method could complete spike detection in approximately one-eighth of the recording time.

Automatic epileptic seizure detection using scalp EEG and advanced artificial intelligence techniques

The epilepsies are a heterogeneous group of neurological disorders and syndromes characterised by recurrent, involuntary, paroxysmal seizure activity, which is often associated with a clinicoelectrical correlate on the electroencephalogram. The diagnosis of epilepsy is usually made by a neurologist but can be difficult to be made in the early stages. Supporting paraclinical evidence obtained from magnetic resonance imaging and electroencephalography may enable clinicians to make a diagnosis of epilepsy and investigate treatment earlier. However, electroencephalogram capture and interpretation are time consuming and can be expensive due to the need for trained specialists to perform the interpretation. Automated detection of correlates of seizure activity may be a solution. In this paper, we present a supervised machine learning approach that classifies seizure and nonseizure records using an open dataset containing 342 records. Our results show an improvement on existing studies by as much as 10% in most cases with a sensitivity of 93%, specificity of 94%, and area under the curve of 98% with a 6% global error using a k-class nearest neighbour classifier. We propose that such an approach could have clinical applications in the investigation of patients with suspected seizure disorders.

Modeling electroencephalography waveforms with semi-supervised deep belief nets: fast classification and anomaly measurement

Clinical electroencephalography (EEG) records vast amounts of human complex data yet is still reviewed primarily by human readers. Deep belief nets (DBNs) are a relatively new type of multi-layer neural network commonly tested on two-dimensional image data but are rarely applied to times-series data such as EEG. We apply DBNs in a semi-supervised paradigm to model EEG waveforms for classification and anomaly detection. DBN performance was comparable to standard classifiers on our EEG dataset, and classification time was found to be 1.7-103.7 times faster than the other high-performing classifiers. We demonstrate how the unsupervised step of DBN learning produces an autoencoder that can naturally be used in anomaly measurement. We compare the use of raw, unprocessed data--a rarity in automated physiological waveform analysis--with hand-chosen features and find that raw data produce comparable classification and better anomaly measurement performance. These results indicate that DBNs and raw data inputs may be more effective for online automated EEG waveform recognition than other common techniques.

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.

Background activity originating from same area as events in the EEG of paediatric patients with focal epilepsy

The aim of this study was to investigate the presence of apparent non-epileptiform activity arising in the same brain area as epileptiform activity in the EEG of paediatric patients with focal epilepsy. The EEG from eight patients was analyzed by an automated method which detects epochs with a single underlying source having a dipolar potential distribution. The EEG with the highlighted detections was then rated by an EEGer with respect to epileptiform activity. Although EEGer-marked events and computer detections often coincided, in five out of the eight patients a substantial number of other detections were found to arise from the same area as the marked events. The morphology of a high proportion of these other detections did not resemble typical epileptiform activity.

A system for automatic artifact removal in ictal scalp EEG based on independent component analysis and Bayesian classification

OBJECTIVE

To devise an automated system to remove artifacts from ictal scalp EEG, using independent component analysis (ICA).

METHODS

A Bayesian classifier was used to determine the probability that 2s epochs of seizure segments decomposed by ICA represented EEG activity, as opposed to artifact. The classifier was trained using numerous statistical, spectral, and spatial features. The system's performance was then assessed using separate validation data.

RESULTS

The classifier identified epochs representing EEG activity in the validation dataset with a sensitivity of 82.4% and a specificity of 83.3%. An ICA component was considered to represent EEG activity if the sum of the probabilities that its epochs represented EEG exceeded a threshold predetermined using the training data. Otherwise, the component represented artifact. Using this threshold on the validation set, the identification of EEG components was performed with a sensitivity of 87.6% and a specificity of 70.2%. Most misclassified components were a mixture of EEG and artifactual activity.

CONCLUSIONS

The automated system successfully rejected a good proportion of artifactual components extracted by ICA, while preserving almost all EEG components. The misclassification rate was comparable to the variability observed in human classification.

SIGNIFICANCE

Current ICA methods of artifact removal require a tedious visual classification of the components. The proposed system automates this process and removes simultaneously multiple types of artifacts.

Efficient unsupervised algorithms for the detection of seizures in continuous EEG recordings from rats after brain injury

Long-term EEG monitoring in chronically epileptic animals produces very large EEG data files which require efficient algorithms to differentiate interictal spikes and seizures from normal brain activity, noise, and, artifact. We compared four methods for seizure detection based on (1) EEG power as computed using amplitude squared (the power method), (2) the sum of the distances between consecutive data points (the coastline method), (3) automated spike frequency and duration detection (the spike frequency method), and (4) data range autocorrelation combined with spike frequency (the autocorrelation method). These methods were used to analyze a randomly selected test set of 13 days of continuous EEG data in which 75 seizures were imbedded. The EEG recordings were from eight different rats representing two different models of chronic epilepsy (five kainate-treated and three hypoxic-ischemic). The EEG power method had a positive predictive value (PPV, or true positives divided by the sum of true positives and false positives) of 18% and a sensitivity (true positives divided by the sum of true positives and false negatives) of 95%, the coastline method had a PPV of 78% and sensitivity of 99.59, the spike frequency method had a PPV of 78% and a sensitivity of 95%, and the autocorrelation method yielded a PPV of 96% and a sensitivity of 100%. It is possible to detect seizures automatically in a prolonged EEG recording using computationally efficient unsupervised algorithms. Both the quality of the EEG and the analysis method employed affect PPV and sensitivity.

Proposal for electroencephalogram standardization in aircrew selection

Current diagnostic electroencephalogram (EEG) investigations in aircrew selection and certification lack both standardization and reference to universally applicable criteria for their effective use. Extrapolation from clinical EEG studies may not be appropriate. Recent studies on serial EEGs in aircrew are lacking, whereas follow-up of individuals who failed certification is nonexistent. Population-based EEG studies in healthy subjects are generally underpowered to establish the significance of pathological EEG findings. Advanced digital video/EEG recording, in combination with standardization of data exchange formats and automated detection of pathological grapho-elements, is cost effective when carried out for extended periods, e.g., during flight simulator sessions. Extensive databases of serial video/EEG records in aircrew may thus be easily obtained and validated over time. Prognostic inferences on the significance of pathological EEG discharges may subsequently be derived from these databases.

Slow-wave activity arising from the same area as epileptiform activity in the EEG of paediatric patients with focal epilepsy

OBJECTIVE

The aim of this study was to investigate the presence and characteristics of apparent non-epileptiform activity arising in the same brain area as epileptiform activity in the EEG of paediatric patients with focal epilepsy.

METHODS

The EEG from eight patients was analysed by an automated method which detects epochs with a single underlying source having a dipolar potential distribution. The EEG with the highlighted detections was then rated by a clinical neurophysiologist (EEGer) with respect to epileptiform activity.

RESULTS

Although EEGer-marked events and computer detections often coincided, in five out of the eight patients, a substantial number of other detections were found to arise from the same area as the marked events. The morphology of a high proportion of these other detections did not resemble typical epileptiform activity and had a frequency content mainly in the delta and theta ranges.

CONCLUSIONS

This is, to our knowledge, the first study to use an automated technique to demonstrate the presence of non-epileptiform activity arising from the same area as the epileptiform activity in the EEG of paediatric patients with focal epilepsy. This slow wave activity is likely to be related to the underlying epileptogenic process.

SIGNIFICANCE

This paper suggests a technique for automated detection of focal activity arising from epileptogenic foci. It also provides a new perspective on extracting clinical useful information from slow-wave background EEG activity.