Research On The Method Of Patients' Heart Beat Feature Generation And Recognition Based On Deep Learning

Recently,there is a high incidence of cardiovascular diseases in our country.As the basis of diagnosis,the ECG not only has a huge amount of data,but also has a various kind of heart beat,which makes that become a difficult task for the artificial analysis of ECG.Especially in the clinical monitoring or wearable health monitoring environment,the real-time ECG diagnosis is faced with multiple concurrent,high-frequency,relatively limited medical staff and other practical problems,which is an impossible task for medical staff.In addition,due to the sudden and infrequent occurrence of some abnormal heart beats,it is difficult for cardiologists to capture some important ECG changes of emergency conditions in time,which will directly threaten the life safety of patients.Therefore,how to automatically and timely recognize abnormal heartbeats from a large number of ECG and improve the accuracy and timeliness of ECG diagnosis has become a hot research topic in recent years.This paper mainly focuses on the feature generation and recognition of heart beat.The main contents and innovations are as follows:(1)A method for feature generation of deep stack Auto-Encoder based on cosine distance is proposed.The signal of heart beat usually contains additive environmental noise,which will seriously disturb or destroy the key information of the state of heart beat,so it is very difficult to extract the feature information of heart beat effectively.In addition,part of the time-domain signals in the beat signal change violently,and the rate of change does not conform to the Gaussian distribution.In view of these findings,the above feature generation method first extracts the sparse effective heart beat feature information from the noisy heart beat signal by the Contractive Auto-Encoder and the Sparse Auto-Encoder.Then the cosine distance is used to measure the difference between the input and output heart beat.Finally,the output of this method is used as the input feature of the subsequent heart beat recognition model.The experimental results show that the features generated by this method have high robustness and discrimination.(2)A heart beat recognition based on convolutional neural network for different patients is proposed.When the existing methods of heart beat recognition are faced with serious data imbalance between different classes of heart beat and new patients,the recognition performance will be reduced to a certain extent.In order to suppress the adverse effects caused by the above problems,the recognition method proposed in this paper firstly uses the general training set to train the general background model based on the convolutional neural network,and learns the common information in different patients' ECG.After that,the model coefficient obtained by learning is taken as the initial value of the corresponding parameters in the specific patient recognition model,and the recognition model is fine-tuned with the first five minutes of the patient's heart beat,so that the model could learn the difference characteristics of the specific patient's heart beat,and enhance the recognition performance of the model for the specific patient's heart beat.In addition,the method adopts the improved objective function,which sets the penalty coefficient to modify the weight of loss,so as to improve the model's recognition accuracy for the class of low sample number.The experimental results on MIT-BIH data set show that this method has high recognition performance.(3)Based on the above algorithms and software engineering design ideas,an ECG Housekeeper system for online ECG diagnosis is developed.The system integrates the functions of ECG collection,heart beat segmentation,feature extraction,ECG diagnosis and labeling,monitoring and reminding,etc.In addition,the diagnosis system also has the functions of routine user registration,login,ECG data import,output diagnosis report,risk level determination,and treatment plan,etc.,which makes up for the limitation that information collection,ECG diagnosis and analysis cannot be integrated in the traditional ECG management system.The developed ECG diagnosis system has a friendly interface,simple operation and good practical value.