CHAPTER 1 INTRODUCTION

Transcription

1 1 CHAPTER 1 INTRODUCTION 1.1 INTRODUCTION Epilepsy, a disease known from ancient times, is now considered the most common disorders of the nervous system (Victor and Ropper, 2001). The Greek physician Hippocrates was the first one to recognize that it was a disease of the brain and tried to treat it as such. Religious beliefs avoided systematic, scientific investigations in epilepsy until the 1800s. Epilepsy is now regarded as a window to the brain s function and thus, has become an increasingly active, interdisciplinary field of research (Lockard and Ward 1992). It is second, only to stroke, and affects approximately 1% of the world s population (Engel 1989, Engel and Pedley 1997). While epilepsy occurs in all age groups, the highest incidences occur in infants and in the aged (Iasemidis 2003, Niedermeyer and Lopes de Silva 1993). The high incidence of epilepsy occurs as a result of a large number of causes, including genetic abnormalities, developmental irregularity, febrile convulsion, as well as brain insults such as craniofacial distress, infections of central nervous system, hypoxia, ischemia and tumours. The hallmark symptoms of epilepsy are recurrent seizures. The seizures are due to sudden development of synchronous neuronal firing in the cerebral cortex and are recorded by electrodes on or inside the brain. Electroencephalography (EEG) is the recording of the electrical activity of the brain. To study the brain s electrical activity, through the

2 2 electroencephalographic records, is one of the most important tools, which are simple and inexpensive for the diagnosis of neurological diseases. EEG evaluation of brain function plays a significant role in the diagnosis, discrimination and management of brain diseases such as epilepsy, brain tumours and brain disorders. Epileptics can be affected by one or more types of seizures. Partial seizures begin in a localized area, while generalized seizures develop over a prevalent area on the cortex of the brain. Partial seizures can be further subdivided into simple and complex, where only complex seizures can cause loss of consciousness. Generalized seizures are grouped into six major categories. They are Absence seizures (also known as petit mal) are characterized by a partial loss of consciousness when the individual briefly appears vacant and unresponsive and in addition, involuntary muscle twitches, particularly in the face, are often seen. Myoclonic seizures consist of very brief and irregular arrhythmic movements. Tonic seizures consist of sudden stiffening movements involving the head, body, or extremities that often occur during sleep. Clonic seizures are characterized by repeated, rhythmic motor movements, often involving a large portion of the body as well as causing unconsciousness. Tonic-clonic seizures (also called grand mal) begin with the tonic phase of sudden stiffening movements when the individual may experience symptoms such as loss of orofacial motor control resulting in tongue biting or clenched teeth

3 3 and/or urinary incontinence. This is followed by the clonic phase of rhythmic body movements. After the seizure, the individual may be emotionally distraught, feeling confused or sleepy. Atonic seizures consist of a sudden loss of muscle tone. A brief atonic seizure may elicit mild symptoms such as drooping of the head, but often the seizure is prolonged and the individual falls down from loss of postural tone. Status epilepticus is the term given to describe the lifethreatening condition when an individual experiences extended or successive seizures with no recovery time. Depending on the medical professional, seizure activity can be considered status epilepticus if it lasts a minimum of five minutes up to a more conservative 30 minutes. Seizures come and go, in a seemingly unpredictable way. In some patients, seizures can occur hundreds of times per day; in rare instances, they occur only once every few years (Guyton 1991). Approximately 33% of patients with epilepsy have seizures that are refractory to medical therapy. For these patients, surgical treatment may be an option. Surgical treatment can be effective in carefully selected cases, which usually represent 8% of the total epileptic patients. The seizures of the remaining 25% of the population cannot be controlled (Iasemidis 2003). The detection of epileptiform discharges occurring in the EEG between seizures is an important component in the diagnosis of epilepsy. (Adeli et al., 2003, Subasi et al., 2006). There are various techniques for seizure detection. The techniques that is used to address this problem, such as

4 4 the analysis of EEG signals for epileptic seizure detection is by using the autocorrelation function, frequency domain features, time frequency analysis, and wavelet transform (WT) (Guler et al., 2001, Adeli et al., 2003, Subasi et al., 2006). The results of the studies in literature demonstrated that the WT is the most promising method to extract features from the EEG signals (Hazarika et al., 1997, Adeli et al., 2003, Khan and Gotman 2003, Kiymik et al., 2004, Subasi et al., 2007). In this aspect, in the present study for epileptic seizure detection in patients with seizures, the WT is used for feature extraction from the EEG signals belonging to the normal and the subjects with seizure. Wavelet is an effective time frequency analysis tool for analyzing transient signals. Its feature extraction and representation properties can be used to analyze various transient events in biological signals. Through wavelet decomposition of the EEG records, transient features are accurately captured and localized in both time and frequency context. Conventional methods of diagnosing epileptic seizure rely on detecting the presence of particular signal features by a human observer. The classification of EEG patterns is based on features that are used to describe the EEG. This consideration must provide the necessary differentiation between the different EEG pattern types and must be such that features are clinically meaningful. In an attempt to describe the temporal and spatial characteristics of the EEG, many features have been discussed in the literature (Agarwal and Gotman 2001). Feature extraction with few parameters was used as inputs to the network for classification. Principal component analysis, or PCA, is a technique that is widely used for applications such as dimensionality reduction, lossy data compression, feature extraction, and data visualization. PCA is a statistical method used to transform the input space into a new lower dimensional space

5 5 and has been used in respiratory, EEG measurements for identifying the most representative features (Ferrigno and Carnevali 1998, Anandan Kavitha et al., 2009, Nazari et al., 2009). PCA has been widely used to identify and summarize many inter-relationships that exist among individual variables. In this, inter-correlated variables are combined into a smaller number of new variables called Principal Components. The first Principal Component accounts for much of the variability in the data and each succeeding component accounts for the remaining variability. The uncorrelated variables are linear combinations of the original variables can be removed with minimum loss of real data to identify new meaningful underlying variables. PCA technique has been investigated before by researchers for signal and image processing (Salaffi et al., 2000, Marek et al., 2003, Arnaz and Robert 2004). Such analyses were employed to estimate driver s drowsiness level, classification of alcoholics and to reduce data dimensionality (Pari Jahankhani et al., 2007, Mu Li et al., 2008). The objective of this study is to extract features out of the different wavelets, choose the suitable wavelet using PCA and analyze the interdependency of various features using Principal Component Analysis. Artificial neural networks (ANNs) have been used successfully in prediction and classification of signals, images and data (Jesu and Ramakrishnan 2007). An ANN is trained from the presented input parameters and the trained ANN can be employed for the classification of a set of information, including the training examples. The advantage of neural networks is that they can be used to predict one or more output types through a flexible network of weights, transfer functions and input variables (Sachin et al., 2007, Sujatha and Ramakrishnan 2008). They have been used in a great number of medical diagnostic decision support systems (Benardos and Vosniakos 2007). Artificial neural networks (ANNs) may offer a potentially superior method of EEG signal analysis to the spectral analysis methods. In

6 6 contrast to the conventional spectral analysis methods, ANNs not only model the signal, but also make a decision as to the class of signal (Guler and Ubeyli, 2005). ANNs have been used as computational tools for pattern classification including epileptic seizure detection (Qu and Gotman 1997, Kiymik et al., 2004, Subasi and Erçelebi 2005). Radial basis function (RBF) neural networks are good at modelling nonlinear data and can be trained in one stage rather than using an iterative process as in Multilayer Layer Perceptron and also learn the given application quickly. A Radial Basis Function Network, a highly versatile and easily implementable classifier was chosen to facilitate the selection of decisive features. Radial basis function networks train rapidly, usually orders of magnitude faster than Back Propagation Network BPN. The PSO algorithm is a population based search algorithm based on social behaviour of birds within a flock. PSO requires only primitive mathematical operators and is computationally inexpensive in terms of both memory requirements and speed. The features that drive PSO are social interaction. Individuals (particles) within the swarm learn from each other and based on the knowledge obtained move to become more similar to their better neighbours. Each individual in PSO flies in the search space with a velocity that is dynamically adjusted according to its own flying experience and its companions flying experience. Compared with other evolutionary algorithms, such as genetic algorithm, PSO algorithm possesses attractive properties such as memory and constructive cooperation between individuals, so it has more chance to fly into the better solution areas more quickly and discover reasonable quality solution much faster. In recent years, the integration of neural networks and fuzzy logic has given birth to new research into neuro-fuzzy systems. Neuro-fuzzy systems have the potential to capture the benefits of both these fields in a

7 7 single framework. Neuro-fuzzy systems eliminate the basic problem in fuzzy system design (obtaining a set of fuzzy if then rules) by effectively using the learning capability of an ANN for automatic fuzzy if then rule generation and parameter optimization. As a result, those systems can utilize linguistic information from the human expert as well as measured data during modelling. Such applications have been developed for signal processing, automatic control (Bhuvaneswari et al., 2009), information retrieval, database management, computer vision and data classification (Iasemidis et al., 2003, Subasi 2006, 2007). 1.2 OBJECTIVE OF THE THESIS The objectives of this research work are To identify the significant wavelet for Seizure Detection using Principal Component Analysis Technique, To identify distinct features in EEG using Wavelet transform decomposition method and to correlate the results with those of conventional methods, To derive significant and useful features from the Principal Component Analysis from normal and seizure EEG To classify the seizure EEG as normal and abnormal using Back propagation algorithm, Radial Basis Function, Adaptive neuro fuzzy inference system and Particle swarm Optimisation Neural network. Figure 1.1 shows the flow diagram of the proposed study.

8 8 EEG SIGNAL (Collection of clinical data) WAVELET TRANSFORM TECHNIQUE HAAR, DB2, DB4, DB5,DB8, BIOR 4.4, QUADRATIC SPLINE, COIFLET4, SYMLET4 FEATURE EXTRACTION PRINCIPAL COMPONENT ANALYSIS PCA BASED CHOICE OF WAVELET AND FEATURE REDUCTION INTELLIGENT SYSTEMS BPN ANFIS RBFNN PSONN NORMAL, COMPLEX PARTIAL, TONIC CLONIC (CPS) (GTCS) Figure 1.1 Flow diagram of the proposed study 1.3 ORGANISATION OF THE THESIS The work reported in the thesis is organized into 5 Chapters: Chapter 2 discusses a critical review of the literature on methods for detection of epileptic seizures, Wavelet Transform method, Principal Component Analysis, Radial basis function neural network, Particle swarm optimisation neural network and Adaptive Neuro Fuzzy Inference System. Chapter 3 describes the methods and protocols for detection of the various types of seizures using Wavelet transformation and Principal Component Analysis methods and explains the feature extraction and classification of features based on BPN, RBFNN, PSONN and ANFIS. Chapter 4 focuses on the results obtained through the analysis and the conclusions drawn from the analysis are presented in Chapter 5. The scope of future work is discussed in Chapter 6.