Research And Application Of Arrhythmia Classification Based On Hybrid Model

Cardiovascular disease is one of the important diseases threatening human life and health,which has the characteristics of high morbidity,high disability rate and high mortality.Arrhythmia is a common cardiovascular disease.Electrocardiogram(ECG)is the main way to diagnose arrhythmia.In the time-varying / dynamic mode,ECG is the performance process of complex nonlinear dynamic system of cardiac electrical activity.ECG contains the rhythm characteristics of continuous heartbeat and the morphological characteristics of ECG waveform.In addition,multi-channel ECG data has upper and lower edge correlation.According to the potential characteristics of ECG,ECG automatic analysis can effectively improve the efficiency of medical diagnosis and shorten the diagnosis time,which has a reliable clinical application value.Based on the large-scale ECG data set and the target of remote ECG aided diagnosis,a multi classification aided diagnosis model is established by using recursive inference network(RIN),which can solve the problem that feedforward neural network is difficult to learn ECG time series characteristics.Further use data augmentation technology to improve the number of samples in the classification boundary area,solve the problem of unbalanced classification data,and improve the accuracy of auxiliary diagnosis model.Finally,the remote ECG monitoring system is designed to realize the remote ECG auxiliary diagnosis function.The main research work includes:(1)According to the temporal and morphological characteristics of multi-channel ECG data,a recursive inference network is constructed.By combining the context information processing mechanism of recursive network and the reasoning feature of invariable translation of convolutional network,the internal spatiotemporal characteristics of multi lead ECG data are extracted.Firstly,ECG is preprocessed by high pass filter and band stop filter,and then ECG time series data is input into Rin recursion layer in time slice mode to learn the time correlation of ECGdata.In the recursive layer,the adjacent shared convolution unit performs reasoning operation on the adjacent lead data,and constructs the network layer form to train the model in the multi-layer recursive way,so as to fit six types of arrhythmia ECG dynamic performance process.The experimental results show that the accuracy of RIN classification algorithm is 86.37%.(2)In view of the imbalance of the number of arrhythmia cases,the boundary sample synthesis algorithm(BSSA)is designed.The imbalance of sample number of different cases will reduce the classification accuracy of the model.BSSA interpolates a few samples closer to the boundary,generates composite data to expand a few samples,and improves the effectiveness of classification.The experimental results show that the accuracy of the algorithm is 91.8%,which improves the recognition rate of small sample data set and the overall recognition.(3)Aiming at the short-term and low efficiency of traditional ECG monitoring technology,the ECG monitoring system is designed.Through the demand analysis,the overall design of the remote ECG monitoring system has been completed,which provides the functions of data real-time access,data storage and cache,user information management,ECG remote diagnosis service,doctor-patient UI display and so on.Finally,based on the above-mentioned auxiliary diagnosis model to realize the remote ECG auxiliary diagnosis,to verify the connectivity of the ECG monitoring system and the effectiveness of the auxiliary diagnosis algorithm.The test results show that the system has the function of remote ECG assisted diagnosis.