基于固有模态分解和深度学习的抑郁症脑电信号分类分析

Empirical mode decomposition and deep learning for classifying and analyzing electroencephalography signals of depression patients

以采集到的抑郁症患者和正常人的脑电信号为基础,采用固有模态分解算法对原始信号去噪处理,通过卷积神经网络对抑郁症患者和正常人进行分类分析。首先通过脑电信号的采集实验,采集15位抑郁症患者和15位正常人对照组Fp1的静息态脑电信号;之后对采集到的静息态脑电进行去噪处理,脑电去噪处理主要包括固有模态分解算法对原始信号的分解获得不同层次的IMF分量,对IMF分量进行频域分析,通过硬阈值的方法剔除原始信号中的噪声信号;最后采用卷积神经网络对抑郁症患者和正常人对照组进行二值分类,结果相较于传统的特征提取-机器学习算法,分类准确率明显提高。

Empirical mode decomposition（EMD） was used to denoise the original electroencephalography（EEG） signal of depression patients and normal controls, and convolutional neural networks（CNN） was applied to make a classification analysis for depression patients and normal controls. The resting-state EEG signals at Fp1 collected from 15 depression patients and 15 normal controls were denoised with EMD. The original signals were decomposed with EMD method to obtain the intrinsic mode function of different layers which were analyzed with frequency-domain analysis, and the noise signals were removed with hard threshold method. Finally, CNN was applied to perform binary classification for the signals from depression patients and normal controls, and the results showed that the classification accuracy of CNN is significantly higher than that of feature extractionmachine learning algorithm.