Investigation Capability of EEG-Based Non-Linear Features in Depression Detection
Abstract
Purpose: The purpose of this study is to investigate the potential of non-linear electroencephalography-based features in depression detection.
Materials and Methods: First, the Electroencephalography (EEG) signal was recorded from 25 normal and 25 depressed subjects. After preprocessing these signals, non-linear features including the Sum of Logarithmic (SL) and second-order spectral Moment (2M) of the amplitudes of diagonal elements in the bispectrum, the normalized Entropy (En) of bispectrum in beta and gamma frequency bands, Katz Fractal Dimension (KFD), and Lempel-Ziv Complexity (LZC) are extracted from them. Then, the ability of these features in depression detection was investigated using Mann-Whitney statistical test. Also, the classification performance of significant features was evaluated using a support vector machine (SVM) classifier.
Results: The results of the statistical analysis demonstrate that bispectral 2M, SL, and KFD features show significant differences between depressed and healthy groups in the Eyes-Closed (EC) condition. Also, bispectral 2M and SL in the gamma frequency band show significant differences between the two groups in parietal and temporal regions in the EC condition and only in the temporal region in the Eyes-Open (EO) condition. Bispectral En does not show a significant difference in the whole 19 channel, but it shows significant differences in the frontal region and beta frequency band. Between these features, gamma bispectral 2M in the temporal region and EO condition shows the highest classification result with 78.6±7.2% accuracy.
Conclusion: Findings confirm that bispectral 2M in the gamma frequency band and EO condition can classify depression and healthy subjects.