Department of Anesthesiology, Kyoto Prefectural University of Medicine, Kyoto, Japan.
Department of Anesthesia, Yodogawa Christian Hospital, Osaka, Japan.
PeerJ. 2024 Nov 15;12:e18518. doi: 10.7717/peerj.18518. eCollection 2024.
Mode decomposition methods are used to extract the characteristic intrinsic mode function (IMF) from various multidimensional time series signals. We analyzed an electroencephalogram (EEG) dataset for sevoflurane anesthesia using two wavelet transform-based mode decomposition methods, comprising the empirical wavelet transform (EWT) and wavelet mode decomposition (WMD) methods, and compared the results with those from the previously reported variational mode decomposition (VMD) method.
To acquire the EEG data, we used the software application EEG Analyzer, which enabled the recording of raw EEG signals the serial interface of a bispectral index (BIS) monitor. We also created EEG mode decomposition software to perform empirical mode decomposition (EMD), VMD, EWT, and WMD operations.
When decomposed into six IMFs, the EWT enables narrow band separation of the low-frequency bands IMF-1 to IMF-3, in which all central frequencies are less than 10 Hz. However, in the upper IMF of the high-frequency band, which has a center frequency of ≥ 10 Hz, the dispersion within the frequency band covered was widespread among the individual patients. In WMD, a narrow band of clinical interest is specified using a bandpass filter in a Meyer wavelet filter bank within a specific mode-decomposition discipline. When compared with the VMD and EWT methods, the IMF that was decomposed WMD was accommodated in a narrow band with only a small variance for each patient. Multiple linear regression analyses demonstrated that the frequency characteristics of the IMFs obtained from WMD best tracked the changes in the BIS upon emergence from general anesthesia.
The WMD can be used to extract subtle frequency characteristics of EEGs that have been affected by general anesthesia, thus potentially providing better parameters for use in assessing the depth of general anesthesia.
模式分解方法用于从各种多维时间序列信号中提取特征固有模式函数 (IMF)。我们使用两种基于小波变换的模式分解方法，即经验小波变换 (EWT)和小波模式分解 (WMD)，对七氟醚麻醉的脑电图 (EEG) 数据集进行了分析，并将结果与先前报道的变分模态分解 (VMD) 方法进行了比较。
为了获取 EEG 数据，我们使用了 EEG 分析器软件，该软件能够记录原始 EEG 信号和 bispectral index (BIS) 监视器的串行接口。我们还创建了 EEG 模式分解软件，以执行经验模式分解 (EMD)、VMD、EWT 和 WMD 操作。
当分解为六个 IMF 时，EWT 能够将低频带 IMF-1 到 IMF-3 进行窄带分离，其中所有中心频率都小于 10 Hz。然而，在高频带的较高 IMF 中，中心频率≥10 Hz，各个患者的频带内的频带分散较为广泛。在 WMD 中，使用特定模式分解学科中的 Meyer 小波滤波器组中的带通滤波器指定感兴趣的临床窄带。与 VMD 和 EWT 方法相比，通过 WMD 分解的 IMF 被容纳在一个窄带内，每个患者的方差都很小。多元线性回归分析表明，WMD 获得的 IMF 的频率特征最能跟踪全身麻醉苏醒时 BIS 的变化。
WMD 可用于提取受全身麻醉影响的 EEG 的细微频率特征，从而为评估全身麻醉深度提供更好的参数。
