RETRACTED ARTICLE: Noninvasive method of epileptic detection using DWT and generalized regression neural network

• Published in: Soft computing (Berlin, Germany) Germany), 2019-04, Vol.23 (8), p.2645-2653
• Main Authors: Vijay Anand, S., Shantha Selvakumari, R.
• Format: Article
• Language: English
• Subjects: Accuracy; Anomalies; Approximation; Artificial Intelligence; Artificial neural networks; Classification; Computational Intelligence; Control; Convulsions & seizures; Datasets; Decomposition; Discrete Wavelet Transform; Discriminant analysis; Electroencephalography; Engineering; Entropy; Epilepsy; Error analysis; Focus; Fourier transforms; Mathematical Logic and Foundations; Mechatronics; Neural networks; Neurological disorders; Parameter sensitivity; Robotics; Seizures; Signal classification; Signal processing; Statistical analysis; Support vector machines; Wavelet transforms
• ISSN: 1432-7643; 1433-7479
• DOI: 10.1007/s00500-018-3630-y

Epilepsy is a continual disorder, the characteristic of which is recurrent, motiveless seizures. Many people with epilepsy have more than one type of seizure and may have other symptoms of neurological problems as well. In this paper, a noninvasive method using discrete wavelet transform (DWT) and neural network is projected for automatic detection of epilepsy from EEG signals. DWT of the EEG signals is carried out using Haar Wavelets. Statistical features of approximate and detailed coefficients are extracted from the transformed signal. The entropy, as well as approximate entropy of the transformed signals, is determined. The features extracted from the transformed signal are used as the training set for the artificial neural network (ANN). Two types of ANNs viz. feedforward neural network (FFNN) and generalized regression neural network (GRNN) are trained. Three types of subjects viz. healthy, seizure-free period of an epileptic patient and epileptic patients are considered. The signals are classified accordingly as normal, seizure-free epileptic and abnormal. The results are compared on the basis of the confusion matrix, error histogram, and error plot. The quality measures used for comparison are sensitivity, specificity, precision, and accuracy. On all the evaluation parameters, GRNN is found to be best suited for anomaly detection in EEG signals.