Turky N. Alotaiby

A review of channel selection algorithms for EEG signal processing

Digital processing of electroencephalography (EEG) signals has now been popularly used in a wide variety of applications such as seizure detection/prediction, motor imagery classification, mental task classification, emotion classification, sleep state classification, and drug effects diagnosis. With the large number of EEG channels acquired, it has become apparent that efficient channel selection algorithms are needed with varying importance from one application to another. The main purpose of the channel selection process is threefold: (i) to reduce the computational complexity of any processing task performed on EEG signals by selecting the relevant channels and hence extracting the features of major importance, (ii) to reduce the amount of overfitting that may arise due to the utilization of unnecessary channels, for the purpose of improving the performance, and (iii) to reduce the setup time in some applications. Signal processing tools such as time-domain analysis, power spectral estimation, and wavelet transform have been used for feature extraction and hence for channel selection in most of channel selection algorithms. In addition, different evaluation approaches such as filtering, wrapper, embedded, hybrid, and human-based techniques have been widely used for the evaluation of the selected subset of channels. In this paper, we survey the recent developments in the field of EEG channel selection methods along with their applications and classify these methods according to the evaluation approach.

Seizure detection with common spatial pattern and Support Vector Machines

This paper extends the use of the Common Spatial Pattern (CSP) algorithm for epileptic Electroencephalography (EEG) seizure detection. The CSP algorithm is applied on EEG signal derivative, which contains reinforced details of the signal. The main idea of the proposed approach is to apply a differentiator on the multi-channel EEG signal, and hence the signal is segmented into overlapping segments. Each segment is projected on a CSP projection matrix to extract the training and testing features. In selecting the training period, a leave-one-hour-out cross validation strategy is adopted. A Support Vector Machine (SVM) classifier is then trained with the training features to classify inter-ictal and ictal segments. Two variants of the CSP are presented and tested in this paper; the original CSP and the Diagonal Loading CSP.

MEG data classification for healthy and epileptic subjects using linear discriminant analysis

Electroencephalogram (EEG) is the most commonly used clinical tool for the early diagnosis of epilepsy. However, with the recent advances in the magnetoencephalography (MEG) technology, a new source of information for the analysis of brain signals has been established. Epileptologists often spend considerable amount of time to review MEG recordings to determine whether or not a particular subject can be classified as an epileptic patient. This paper proposes a new algorithm for automatic classification of MEG data into two classes: data that belongs to healthy subjects and data that belongs to epileptic subjects. The classifier makes use of linear discriminant analysis (LDA) and considers features extracted from the signals of eight regions in the brain. The effectiveness of proposed classifier has been tested using real MEG data obtained from 15 healthy subjects and 18 epilepsy patients. The results obtained show good promise, which make the proposed classifier a valuable tool for analyzing brain signals in the initial assessment phases of subjects under epileptic symptoms.

Channel selection and seizure detection using a statistical approach

This paper proposes a novel patient-specific approach to channel selection and seizure detection based on estimating the histograms of multi-channel scalp electroencephalography (sEEG) signals. It consists of two main phases: training and testing. In the training phase, the signal is segmented into non-overlapping 10-second segments, with five histograms estimated for each segment. These histograms have multiple bins that are studied individually as random variables. Based on the histograms of these random variables for different signal activities and on predefined detection and false alarm probability thresholds, bin(s) are selected form certain channel distributions for seizure detection. In selecting the training hours, a leave-one-out cross-validation strategy is adopted. In the testing phase, those channel(s)-histogram(s)-bin(s) are used to classify each segment as ictal or non-ictal. This sequence is filtered with a moving average filter and compared to a patient-specific detection threshold. This method was evaluated using 309.9 h of sEEG including 26 seizures of five patients. It achieved an average sensitivity of 97.14% and an average specificity of 98.58%.

Online adaptive seizure prediction algorithm for scalp EEG

Epilepsy is a brain disorder, which affects around 1% of world population. The life of epilepsy patients can be improved by predicting seizures before its occurrence. It has been observed that EEG signals during the pre-seizure state are less chaotic compared to their behavior at normal state. Therefore, chaoticity measure can be used to develop seizure predictor. In this paper, we propose seizure prediction algorithm based on Largest Lyapunov Exponent (LLE) to measure the chaoticity of scalp EEG signals. The proposed algorithm makes use of LLE to define two baselines; one for the normal state and the other for the pre-state. The distance between the two baselines and the LLEs of an Electroencephalography (EEG) signal of unknown state is computed for signal classification. The two baselines are updated through a simple mechanism. The performance of proposed algorithm has been evaluated using MIT database.

Enhancing the reliability of epileptic seizure alarms for scalp EEG signals

One of the key requirement for the development of seizure prediction system is that the seizure alarms generated should be reliable, i.e. the system should have high seizure detection rate and minimum false alarm rate. In this paper, we explore the relationship between the chaotic behavior and energy ratios of the sub bands of an EEG signal. This relationship will then be used to enhance the reliability of seizure alarms generated by measuring the chaoticity of EEG signals using the Largest Lyapunov Exponent (LLE). It is shown, in this paper, that when both LLE and energy ratios of EEG signal sub bands are used to predict an incoming seizure, then the reliability of prediction system gets enhanced; hence any alarm generated in this case must be taken by the patient (caregivers) seriously and appropriate safety measures must also be taken place.

Epileptic MEG Spikes Detection Using Amplitude Thresholding and Dynamic Time Warping

Epilepsy is a brain disorder that may strike at different stages of life. Patients’ lives are extremely disturbed by the occurrence of sudden unpredictable epileptic seizures. A possible approach to diagnose epileptic patients is to analyze magnetoencephalography (MEG) signals to extract useful information about subject’s brain activities. MEG signals are less distorted than electroencephalogram signals by the intervening tissues between the neural source and the sensor (e.g., skull, scalp, and so on), which results in a better spatial accuracy of the MEG. This paper aims to develop a method to detect epileptic spikes from multi-channel MEG signals in a patient-independent setting. Amplitude thresholding is first employed to localize abnormalities and identify the channels where they exist. Then, dynamic time warping is applied to the identified abnormalities to detect the actual epileptic spikes. The sensitivity and specificity of proposed detection algorithm are 92.45% and 95.81%, respectively. These results indicate that the proposed algorithm can help neurologists to analyze MEG data in an automated manner instead of spending considerable time to detect MEG spikes by visual inspection.

Epileptic Seizure Prediction Using CSP and LDA for Scalp EEG Signals

This paper presents a patient-specific epileptic seizure predication method relying on the common spatial pattern- (CSP-) based feature extraction of scalp electroencephalogram (sEEG) signals. Multichannel EEG signals are traced and segmented into overlapping segments for both preictal and interictal intervals. The features extracted using CSP are used for training a linear discriminant analysis classifier, which is then employed in the testing phase. A leave-one-out cross-validation strategy is adopted in the experiments. The experimental results for seizure prediction obtained from the records of 24 patients from the CHB-MIT database reveal that the proposed predictor can achieve an average sensitivity of 0.89, an average false prediction rate of 0.39, and an average prediction time of 68.71 minutes using a 120-minute prediction horizon.

Epileptic MEG Spike Detection Using Statistical Features and Genetic Programming with KNN

Epilepsy is a neurological disorder that affects millions of people worldwide. Monitoring the brain activities and identifying the seizure source which starts with spike detection are important steps for epilepsy treatment. Magnetoencephalography (MEG) is an emerging epileptic diagnostic tool with high-density sensors; this makes manual analysis a challenging task due to the vast amount of MEG data. This paper explores the use of eight statistical features and genetic programing (GP) with the K-nearest neighbor (KNN) for interictal spike detection. The proposed method is comprised of three stages: preprocessing, genetic programming-based feature generation, and classification. The effectiveness of the proposed approach has been evaluated using real MEG data obtained from 28 epileptic patients. It has achieved a 91.75% average sensitivity and 92.99% average specificity.

A common spatial pattern approach for scalp EEG seizure detection

This paper presents patient-specific epileptic seizure detection approach based on Common Spatial Pattern (CSP) and its variants; Diagonal Loading Common Spatial Pattern (DLCSP), and Tikhonov Regularization Common Spatial Pattern (TRCSP). In this proposed approach, multi-channel scalp Electroencephalogram (sEEG) signals are traced and segmented into overlapping segments for both normal and epileptic seizure intervals. Features are extracted from each signal segment through projection on a CSP projection matrix. The extracted features are used for training a Support Vector Machine (SVM) classifier, which is then employed in the testing phase. A leave-one-out cross validation strategy is adopted in the experiments. The proposed approach was evaluated using 443.55 hours of sEEG including 39 seizures. The experimental results reveal that a patient-specific CSP-based algorithm is capable of detecting epileptic seizures with high accuracy. In particular, the CSP approach has achieved 100% an average sensitivity, 1.17 an average false alarm, and 7.02 s an average detection latency time.