基于多参数特征融合的自动睡眠分期方法

Automatic sleep staging method based on multi-parameter feature fusion

睡眠分期是睡眠障碍等相关疾病诊断的重要依据。为实现准确且具有泛化能力的自动睡眠分期,提出了一种基于多参数特征融合的自动睡眠分期方法。首先,利用离散小波变换(DWT)对原始数据进行滤波预处理,获取脑电图(EEG)的α波、β波、θ波、δ波、肌电图(EMG)的高频成分的特征能量比,采用样本熵算法提取EEG、心电图(ECG)的非线性特征,提取呼吸信号的均值,共八个特征;然后,将特征参数输入支持向量机(SVM)分类器中进行样本训练与分类识别,并把分类结果与基于EEG的睡眠分期方法、基于EEG和ECG的睡眠分期方法结果进行对比,验证了该方法的可行性与准确性;最后,采用交叉验证方法对不同样本进行建模,验证其泛化能力。结果表明,该方法准确率可以达到92.95%,相比基于EEG睡眠分期算法平均准确率提高9.65%,交叉验证后平均准确率达85.15%,具有较好的泛化能力。

Sleep staging is the important basis of diagnosing sleep related diseases. In order to achieve accurate and generalized automatic sleep staging, a sleep staging method based on multi-parameter feature fusion was proposed. Firstly,Discrete Wavelet Transform（ DWT） was used to filter these signals, the energy ratio of α, β, θ, δ rhythm waves from Electro Encephalo Gram（ EEG） and the high frequency component from Eletro Myo Gram（ EMG） were extracted. Nonlinear characteristics of EEG and Electro Cardio Gram（ ECG） were extracted by calculating Sample Entropy（ Samp En）, and the mean time domain characteristics of Resp were calculated. Secondly, these features were input to Support Vector Machine（ SVM）classifier to train and classify, meanwhile, to test the feasibility and accuracy. The sleep staging results were compared to those based on EEG and based on EEG and ECG. Finally, different samples were modeled to verify the generalization by cross-validation method. The result shows that the accuracy of the automatic sleep staging method reaches 92. 95%, the average accuracy is 9. 65% higher than that by using the sleep staging method based on EEG. The cross-validation result with an average accuracy of 85. 15% showe good generalization.