The Research Of Automatic Diagnosis Of Cardiovascular Diseases Based On PCG And ECG

Electrocardiogram(ECG)and Phonocardiogram(PCG)play an important role in early prevention and diagnosis of heart diseases.As the fast development of machine learning,automatic detection of cardiovascular diseases based on ECG and PCG has been paid much attention.However,because of the complexity and individual variation of heart activity,there are still enormous challenges in the field of automatic detection and diagnosis of cardiovascular diseases based on ECG or PCG.In previous relative studies,researchers mostly did signal analysis and model development based on single ECG data source or single PCG data source.But in clinic,cardiologist combines a variety of diagnostic data source to make a final diagnosis of cardiovascular disease,such as PCG,ECG,biomarkers,cardiac color ultrasound and so on.Therefore,it is meaningful and desirable to develop an efficient multi-modal method to automatically predict and diagnose cardiovascular diseases.In 2016,PhysioNet released a dataset about cardiovascular disease that includes both PCG and ECG.Benefit from this,in this study,we proposed a novel multi-modal method for predicting cardiovascular diseases based on machine learning method.Firstly,we utilized supervised learning method combining convolutional neural Network and recurrent neural network,which we called CL-ECG-Net,to extract ECG deep-learned features.And we used supervised method of modified CL-ECG-Net by adding multiple input channels,which we called CL-PCG-Net,to extract PCG deep-learned features.After these,genetic algorithm was used to screen the two modal features and obtain the optimal feature set.Finally,a classifier of SVM is used to make the final prediction of labels.It shows that the accuracy and AUC(Area Under the ROC Curve)reaches up to 0.880,0.936 when using the proposing multi-modal method,which are better than those when using single data resource(ECG:0.872,0.916,PCG:0.704,0.748).In addition,we make the comparison experiment based on the feature dimension reduction method of PCA to evaluate the performance of genetic algorithm.The results show that the genetic algorithm shows better performance in solving the problem of feature dimension reduction.