Empirical wavelet transform and wavelet mode decomposition for frequency characteristic extraction of EEG during sevoflurane general anesthesia

Abstract

Purpose Mode decomposition methods are used to extract the characteristic intrinsic mode function (IMF) from various multidimensional time-series signals. Here, we applied wavelet transform-based mode decomposition to analysis of an electroencephalogram (EEG) recorded during general anesthesia. Methods An empirical wavelet transform (EWT) algorithm and a wavelet mode decomposition (WMD) algorithm with fixed frequency boundaries were added to previously reported EEG Mode Decompositor application software. Using our recently reported sevoflurane anesthesia data set, we performed EWT and WMD operations, and evaluated the significant characteristics via comparison with an existing variational mode decomposition (VMD) method. Results The EWT method, when decomposed into six IMFs, enabled narrowband separation of low-frequency bands IMF-1 to IMF-3, where all central frequencies were under 10 Hz. However, in the upper IMF of the high-frequency band with a center frequency ≥ 10 Hz, the dispersion of the frequency band covered was spread widely among the individual cases. In WMD, a narrow band of clinical interest can be specified using a band-pass filter via a Meyer wavelet filter bank within a specific mode decomposition discipline. When compared with VMD and EWT methods, the IMF decomposed using WMD is accommodated in a narrow band with a small variance for each case. Conclusion Although issues remain with the EWT, e.g., optimizing the process, the EWT can perform feature extraction similar to VMD. The Meyer filter bank used in the WMD represents an attractive technique for characteristic frequency band extraction when used as a band-pass filter in combination with the mode decomposition method.