Research On Prediction Of Mitral Regurgitation Based On PCG And ECG

Mitral regurgitation is a common heart valve disease that increases in prevalence with age and causes structural and functional changes in the heart as the disease progresses,therefore,early diagnosis and treatment of mitral regurgitation can reduce the risk of irreversible damage to the heart.The electrocardiogram(ECG)and phonocardiogram(PCG)are commonly used to record the electrical and mechanical activity of the heart and play an important role in the detection of cardiac abnormalities.When mitral regurgitation occurs,the instrument detects a heart murmur and changes in cardiomyocyte action potential.Among these,the heart sound signal,as the initial diagnostic modality for mitral regurgitation,still relies heavily on the subjective judgment of the physician.In the era of rapid development of computer technology,combining deep learning methods to jointly analyze heart sounds and ECG signals is an effective means to assist physicians in confirming the diagnosis of mitral regurgitation.In addition,no publicly available dataset containing synchronously acquired heart sounds and ECG signals for mitral regurgitation disease has been found to date.Therefore,this paper addresses these challenges with the following major research efforts:(1)In response to the lack of publicly available data sets on synchronized heart sounds and ECG signals in mitral regurgitation,this paper establishes a "Synchronized Phonocardiogram and Electrocardiogram with Mitral Regurgitation Database",which is a database of synchronized heart sounds and ECG signals in mitral regurgitation patients.All data were collected from inpatients of a tertiary care hospital from March2021 to August 2021.(2)In response to the problem of noise in the collected data,the wavelet decomposition and reconstruction algorithm and the empirical mode decomposition(EMD)algorithm were used to make the heart sounds and ECG signals smoother and lay the foundation for further prediction of mitral regurgitation.(3)In response to the current clinical practice of using physicians’ subjective judgment as the basis for initial detection of mitral regurgitation,two early prediction studies of mitral regurgitation were designed in this paper,one requiring segmentation algorithms for the heart sounds and ECG signals,and the other without segmentation algorithms.Among the algorithms for segmentation of heart sounds and ECGs,the main study is based on the Ada Boost prediction model,which is based on the normalized Shannon energy envelope for segmentation and extraction of features in the time and frequency domains,respectively.Experimental results show that the Ada Boost algorithm outperforms Support Vector Machine(SVM)and decision tree algorithms in terms of recall metrics.In the segmentation-free algorithm,the noise-reduced signal is first converted into an image by the Gramian Angular Difference Field(GADF),then the features of PCG and ECG are automatically extracted by a convolutional neural network and fused with features using the Transformer model,and finally a custom residual downsampling neural network(Residual Net-Downsample(Res Net-DS)was used for prediction.The results of the study showed that Res Net-DS obtained accuracy,precision,recall,and F1 metrics of 96.90%,97.10%,97.10%,and 97.10%,respectively,in the SPE-MRDB dataset.In addition,the accuracy,precision,recall and F1 metrics of 94.34%,91.13%,97.69% and 94.30%,respectively,were obtained for the prediction results on the Physio Bank2016 dataset,further demonstrating the good robustness of the Res Net-DS model.In this paper,mitral regurgitation prediction algorithms based on heart sounds and ECG signals are investigated,including both segmented and unsegmented methods.The results showed that the two algorithms achieved 91.30% and 97.10% recall rates,respectively,proving that the algorithm proposed in this paper can effectively predict potential patients.This is important for early prediction of mitral regurgitation,helping physicians in diagnosis,and improving medical care.