Research On Pattern Recognition Of Surface Electrogastrogram Based On Nonlinear Feature Fusion

Surface electrogastrogram(EGG)is a non-invasive method for the evaluation of gastric function.Because the frequency of EGG changes very slowly,approximately three cycles per minute,and the nonlinear features of EGG are more prominent than those of electroencephalogram(EEG)and electrocardiogram(ECG).Therefore,it has important clinical significance and practical value to apply nonlinear features to the extraction and analysis of EGG features in patients with functional dyspepsia(FD)and other gastropathy.Based on the above background,the Laplacian electrode with concentric ring structure was designed according to the principle of minimum error and maximum potential ratio,the flexible circuit board processing technology was adopted to make it into a flexible electrode and applied to EGG signal recording of 10 healthy subjects.The results showed that the Laplacian electrode could effectively suppress ECG and other interference compared with the conventional disk electrode,and the average signal-to-noise ratio(SNR)increased by 5.15 d B.Besides,it was preliminarily proved that the signals recorded by the Laplacian electrode could represent the electrical activity of the stomach.On this basis,the complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)was applied to the preprocessing of simulated EGG signals.The results showed that CEEMDAN could effectively solve the problem of mode mixing existing in traditional empirical mode decomposition(EMD).Compared with FIR bandpass filtering and EMD,the signals preprocessed by CEEMDAN was closer to the simulated real EGG signals,and the maximum correlation coefficient could reach 0.64.Meanwhile,it was proved that CEEMDAN had the best denoising effect which based on the verification experiment of the Laplacian electrode,and compared with FIR bandpass filtering and EMD,the average SNR increased by 7.21 d B and 1.69 d B respectively.Finally,this paper proposed a nonlinear feature extraction method for EGG signal,and used the support vector machine(SVM)to classify the EGG data of before / after acupuncture and patients with FD / normal subjects.The results showed that the nonlinear feature could effectively improve the recognition rate of the EGG signal,especially the feature vector after the fusion of the nonlinear feature and the traditional feature,which namely the traditional-nonlinear feature vector,had the best recognition effect.Among them,the features extracted based on the draft of evaluation standard of EGG in 1999 was called traditional feature in this paper.In acupuncture data,the average recognition rate of traditional-nonlinear eigenvectors in polynomial kernel function SVM and radial basis kernel function SVM were 81.6% and 89.8%respectively.And in FD patients’ data,the average recognition rate of traditionalnonlinear eigenvectors in polynomial kernel function SVM and radial basis kernel function SVM were 82.0% and 87.3% respectively.To sum up,this paper provides a more comprehensive technology solution for the detection and analysis of EGG in clinic,which from the aspects of the Laplacian electrode of EGG detection,preprocessing method of EGG signal based on the CEEMDAN and nonlinear feature extraction method of EGG signal,which lays a foundation for the research of EGG in the future.