Adaptive multi-parent crossover GA for feature optimization in epileptic seizure identification

Highlights

• EEG analysis involves multi-frequency non-stationary brain waves from many channels.

• A challenging situation when involving multiple EEG channels in ES identification.

• The feature vectors which include data from both the time and frequency domain.

• An adaptive multi-parent crossover GA is used to classify epileptic seizures.

• The Algorithm is evaluated using data from the machine learning of the UCI center.

Abstract

EEG signal analysis involves multi-frequency non-stationary brain waves from multiple channels. Segmenting these signals, extracting features to obtain the important properties of the signal and classification are key aspects of detecting epileptic seizures. Despite the introduction of several techniques, it is very challenging when multiple EEG channels are involved. When many channels exist, a spatial filter is required to eliminate noise and extract relevant information. This adds a new dimension of complexity to the frequency feature space. In order to stabilize the classifier of the channels, feature selection is very important. Furthermore, and to improve the performance of a classifier, more data is required from EEG channels for complex problems. The increase of such data poses some challenges as it becomes difficult to identify the subject dependent bands when the channels increase. Hence, an automated process is required for such identification.

The proposed approach in this work tends to tackle the multiple EEG channels problem by segmenting the EEG signals in the frequency domain based on changing spikes rather than the traditional time based windowing approach. While to reduce the overall dimensionality and preserve the class-dependent features an optimization approach is used. This process of selecting an optimal feature subset is an optimization problem. Thus, we propose an adaptive multi-parent crossover Genetic Algorithm (GA) for optimizing the features used in classifying epileptic seizures. The GA-based approach is used to optimize the various features obtained. It encodes the temporal and spatial filter estimates and optimize the feature selection with respect to the classification error. The classification was done using a Support Vector Machine (SVM).

The proposed technique was evaluated using the publicly available epileptic seizure data from the machine learning repository of the UCI center for machine learning and intelligent systems. The proposed approach outperforms other ones and achieved a high level of accuracy. These results, indicate the ability of a multi-parent crossover GA in optimizing the feature selection process in EEG classification.

Introduction

There are various types of neurological diseases such as neurogenetic diseases, degenerative diseases and convulsive disorders [1]. Most of the convulsive disorders occur as a result of an uneven electrical conductivity in the brain which results to uncontrollable quivering of the body. Despite the fact that seizures are part of epilepsy, not all seizures are as a result of epilepsy [2]. Epilepsy affects about 1% of the world population [3]. It is one of the most common neurological diseases, which is incurable, occurs recurrently and unpredictable [4], [5]. However, epilepsy can be controlled with medications, and surgery when patient does not respond to medication [6]. Electroencephalography (EEG) is a brain signal processing technique that provides an insight to the complex inner mechanism of normal and abnormal brain waves, which is used to diagnose brain disorder [7]. Practically, epileptic seizures can be seen as an abnormal, automatic movement or alteration of consciousness related to abnormal EEG changes [8]. However, the EEG signal characteristics differ among patients and between different seizures in the same patient. Since epilepsy show abnormal EEG signal changes, clinically intra-cranial EEGs are used to diagnose, differentiate and classify epileptic seizures [9]. These signals can appear in different forms such as spikes, poly-spikes, or spike and waves. EEG recordings are gathered during seizures and between seizures. These huge data are currently interpreted manually by medical personals, making the process of visual analysis cumbersome. The main aim of a computer-aided EEG classification is not only to reduce the time and effort for medical personals but also to be able to predict future occurrences of epileptic seizures.

Relevant information regarding different brain disorder and states are provided by the EEG signals. These signals are collected in form of datasets and transferred to tools for plotting and visualization of the time–frequency plots [7]. The challenges of automating this process for seizure detection includes:

• The variety of signal characteristics that vary between patients (inter-patient variability), along with different seizures in the same patient (intra-patient variability).

• Abnormalities seen during epileptic seizures can also be observed during some non-epileptic seizures.

• Addressing the high complexity and nonstationarity heterogeneous types seizures [10].

• The EEG includes massive amounts of data and handling them is a costly process [11].

The aim of this work is to propose a technique for epileptic seizure identification in order to efficiently analyze EEG signals for identifying epileptic and non-epileptic seizures. This complex problem of classifying aberrant EEG signals warrants the use of improved feature extraction and reduction techniques, and classification algorithms. The proposed technique adopts a GA, with an adaptive multi-parent crossover, to encode the temporal and spatial filter estimates and optimize the feature selection of EEG.

Genetic algorithms, as a member of the evolutionary algorithms, are search algorithms that imitates the natural evolution process. It defines how a population can regenerate a modified version of its individuals in order to satisfy the required goal based on the natural concept; survival of the fittest. The GA works on coded population by an initializing process. Then a selection process is applied to select some solution that are evaluated to find their fitness value. After that, a reproduction process starts to generate new offspring using crossover and mutation operators. Finally, the replacement process replaces the old solutions by new ones. For more information the reader may refer to [12].

The rest of the paper is organized as follows: In Section 2, a brief survey and a review of the different implementations of epileptic seizure classification is presented. Section 3 presents the proposed framework that comprises of four phases; namely, the windowing phase, feature extraction phase, optimization phase, and classification. The materials and experimental methods are highlighted in Section 4. In Section 5, the results are presented and discussed in Section 6. Finally, Section 7 concludes the research with the findings and future work.