DFAspike: A new computational proposition for efficient recognition of epileptic spike in EEG
Introduction
The electrical activity along the scalp occurs due to the firing of neurons. The recording of these electrical activities is known as electroencephalogram (EEG), which aids in the diagnosis of a variety of neuro-physiological disorders such as epilepsy, coma, encephalopathy, brain death and other focal brain disorders. Epilepsy is a chronic neurological disorder, which is characterized by symptoms such as abnormal synchronous activities in the brain. Such symptoms are termed as seizures. These are manifested as sharp abrupt changes in the EEG amplitude in a very small time interval (Fig. 1). Epilepsy cannot be totally cured, but it can be controlled. However, for the treatment of epilepsy, its occurrence must be efficiently identified. Because of its non-stationary and nonlinear properties, it is very hard to predict the time of occurrence of epileptic events. Thus, a smart automated system with minimum complexity is required for its diagnosis.
Various techniques are available to the researchers and clinicians for the recognition of epileptic spikes. Some of the important methods by which one can proceed for diagnosis of epilepsy are using artificial neural networks [1], [2], [3], [4], [5], fuzzy logic [6] and various other time-frequency approaches [7], [8], [9], [10], [11], [12]. Most of the previous techniques for automated spike recognition need initial training as well as the optimization of the system. Moreover, to the best of authors' knowledge, none of these methods provide accurate information about the time of initiation, total number and the frequency of occurrences of the epileptic spikes.
To overcome the mentioned demerits in spike recognition techniques, deterministic finite automata (DFA) based spike identification algorithms have been developed with both sequential [13] as well as parallel processing [14] approaches. In these approaches the epileptic spike patterns present in the EEG data were defined with a specific combination of occurrences of a set of predefined symbols, which was having only two elements. The average recognition rate of the epileptic spike pattern achieved was 95.68%. Though, in the parallel processing based system, the average recognition rate of epileptic spike pattern was similar to the sequential approach; the use of data parallelism improved the speed of processing of enormous data. Similar DFA based approach for detection of onset of epileptic seizures in noisy domain has also been proposed recently [15]. However, the recognition accuracy achieved by these DFA based methods was not the best as reported with many of the automated system approaches, where the spike recognition rate was claimed with an accuracy of more than 99%. Therefore, to improve the spike recognition rate, in the current work the epileptic spike has been redefined with a set of predefined symbols having three elements. Based on this new definition of epileptic spike and the concept of DFA, a new system named DFASpike has been purposed.
Section snippets
Data recording and pre-processing
The methodology for EEG spike generation, data recording and processing was used as described earlier [16]. The pre-recorded data taken from Sinha et al. [5] was used in this study. These EEG data files were recorded from male Charles Foster rats weighing between 200 and 250 g. EEG recording electrodes were implanted on the rat's head under urethane (Sigma, USA) anesthesia (1.5 g/kg i.p). Three stainless steel screw electrodes of 1 mm diameter soldered with flexible radio wires were implanted
Result
The DFAspike that has been designed for recognition of epileptic spikes scans the input pre-recorded and pre-processed EEG data and finds the total number of occurrences of epileptic spikes in it and also records the start and end time, time duration and maximum amplitude of each spike. After pre-processing, the EEG signal data is converted into a string of symbols −1, 0 and 1 and hence an epileptic spike is represented with a combination of symbols −1, 0 and 1. Altogether 324 different
Discussion
The recording of EEG signal is generally done for a longer duration. This signal data contains many characteristics, which can be used for various clinical researches ultimately helping doctors in the treatment of diseases of a patient. Since the sampling frequency of the recording used to be generally very high (256 Hz or more), recording of EEG signal produces enormous data. These data used to be so huge in size that the manual processing of it becomes almost impossible in real time scenario.
Summary
Classification of epileptic spike has been addressed and demonstrated with the help of various signal processing as well as computational methodologies by various authors. Although, recognition pf epileptic spikes compared to normal electroencephalograhic (EEG) patterns was achieved to the very high percentage of accuracy, the techniques and algorithms were generally found failed in analyzing the common characteristics of EEG spikes like its width, frequency, initiation and end time. In the
Conflict of interest statement
This is to certify that the article submitted for publication in ‘Computers in Biology and Medicine’ has not been published, nor is being considered for publication, elsewhere. There is no ‘Conflict of Interest’ in the publication of the manuscript “DFAspike: A new computational proposition for efficient recognition of epileptic spike in EEG”.
References (24)
- et al.
An approach to seizure detection using an artificial neural network (ANN)
Electroencephalogr. Clin. Neurophysiol.
(1996) - et al.
Recurrent neural network based prediction of epileptic seizures in intra- and extracranial EEG
Neurocomputing
(2000) - et al.
Automated recognition of alertness level by using wavelet transform and artificial neural network
J. Neurosci. Methods
(2004) - et al.
A study on fuzzy C – means clustering – based systems in automatic spike detection
Comput. Biol. Med.
(2007) - et al.
Automatic seizure detection in the newborn: methods and initial evaluation
Electroencephalogr. Clin. Neurophysiol.
(1997) - et al.
Prediction of epileptic seizures
Lancet Neurol.
(2002) - et al.
Analysis of EEG records in an epileptic patient using wavelet transform
J. Neurosci. Methods
(2003) EEG signal classification using wavelet feature extraction and a mixture of expert model
Expert Syst. Appl.
(2007)Automatic detection of epileptic seizures in EEG using discrete wavelet transforms and approximate entropy
Expert Syst. Appl.
(2009)A neural network confirms that physical exercise reverses EEG changes in depressed rats
Med. Eng. Phy
(1995)