Binary and multiclass classifiers based on multitaper spectral features for epilepsy detection. (April 2021)

• Record Type: Journal Article
• Authors: Oliva, Jefferson Tales; Rosa, João Luís Garcia
• Is Part Of: Biomedical signal processing and control. Volume 66(2021)
• Publication Date: 2021-04
• Subjects: Electroencephalogram -- Epilepsy -- Signal processing -- Spectral features -- Machine learning -- Multiclass classification; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102469
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A novel method for epilepsy detection. Multitaper method applied for feature extraction from three different domains. Eight popular machine learning methods for classifier building. An experimental evaluation in a set of 400 EEG segments. Classification results evaluated through statistical tests and confusion matrices. Abstract: Epilepsy is one of the most common neurological disorders that can be diagnosed by means of electroencephalogram (EEG) analysis, in which the following epileptic events can be observed: pre-ictal, ictal, post-ictal, and interictal. In this paper, we present a novel method for epilepsy detection employing binary and multiclass classifiers. For feature extraction, a total of 105 measurements were extracted from power spectrum, spectrogram, and bispectrogram. For classifier building, widely known machine learning algorithms were used. Our method was applied in a publicly available EEG database. As a result, BP-MLP (backpropagation based on multilayer perceptron) and SMO_Pol (sequential minimal optimization supported by the polynomial kernel) algorithms reached the highest accuracy for binary (100%) and multiclass (98%) classification problems. Subsequently, statistical tests did not find a better performance model. In the evaluation based on confusion matrices, it was also impossible to identify a classifier that stands out concerning other models for EEG classification. In comparison to related words, our predictive models reached competitive results.