EEG-based seizure prediction with machine learning
Loading...
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
UMT.Lahore
Abstract
Epilepsy is one of the most recognized neurological illnesses, affecting millionsof individuals worldwide. The illness has long been important in the biomedicalfield because of the threats it poses to human life. The research purposeis to develop a methodology that combines signal processing and machinelearning to predict patient-specific seizure attack so that it can be medicatedbefore the actual seizure attack. A novel method for seizure predictionis proposed that combines support vector machine and wavelet packet decomposition.The essential characteristic of this proposed method is thatonly one hour of data for processing and one to two channels for training andtesting were used, resulting in a computationally efficient technique. Firstly,raw data is being segmented of patient-specific, then discrete wavelet decompositionis performed on this segmented data which is decomposed usingdiscrete wavelet transform into four bands i.e. delta, theta, alpha and beta.Secondly, four features are extracted from this decomposed data. Thirdly,the feature matrix extracted from this decomposition is fed into the classifierto categorize seizure phase i.e. (pre-ictal and inter-ictal). In the end, oncethe pre-ictal state is detected by the support vector machine(SVM) classifier,using the Kalman filtering an alarm is generated. False-positive rate, sensitivity,and accuracy were measured as performance indicators. An averagedaccuracy of 94.9%, 97.43% of sensitivity, and 0.138 false positive rate wereachieved.
Keywords: EEG, Seizure, SVM, Wavelet Packet Decomposition,Pre/Inter-Ictal