Detection of Sleep Spindles in Sleep EEG by using the PSD Methods

Abstract

Background/Objectives: In this study, Fast Fourier Transform (FFT), Welch, Autoregressive (AR) and MUSIC methods were implemented to detect sleep spindles (SSs) in Electroencephalogram (EEG) signals by extracting features in frequency space. Methods/Statistical Analysis: A database from these signals of five subjects which were recorded at sleep laboratory of Necmettin Erbakan University in Turkey was ready for use. The database consisted of 600 EEG epochs in total. The number of epochs was 300 for both with and without SSs in this database. Comparison of the performances of these methods on SS determination process was performed by using Artificial Neural Networks (ANN) classifier. Findings: According to the test classification results, notable difference was obtained between the applied PSD methods. By using the extracted all features, maximum test classification accuracies were achieved as 84.83%, 80.67%, 80.83% and 80.33% with use of FFT, Welch, AR and MUSIC, respectively. To determine the SSs, Principal Component Analysis (PCA) also was utilized in this study. When PCA was applied, the results were 89.50%, 82.00%, 93.00% and 94.83% by use of the same PSD methods, respectively. Application/Improvements: As a result, the performance of PCA and MUSIC is better than the others. Hence, these methods can be used safely for automatic detection of SSs.