Research On Personalized Ectopic Heartbeat Detection Method Based On Improved U-net

Dynamic electrocardiography can effectively capture sudden and insidious ectopic heartbeats by recording 24-hour ECG signals without interruptions,enabling real-time monitoring of subjects’ cardiac conditions.Ectopic heartbeat detection methods based on dynamic ECG signals can be automated.Still,most of them cannot locate ectopic heartbeats and face the problems of lack of labels and significant differences in signal characteristics between patients.Therefore,this paper carried out research on the multi-target detection of ectopic heart beat,screening and labeling of patients with ectopic heart beat and personalized multi-target detection of ectopic heart beat for ectopic heartbeats from ectopic heartbeats the practical need of designing a personalized multi-target detection method for patients.The main work contents and innovations are as follows:(1)To realize the multi-target detection of ectopic heartbeats,the U-net-Bi LSTM-CRF based ectopic heartbeat multi-target detection method was proposed.This method can also solve the problems of U-net’s insufficient learning ability of long time sequence correlation and difficult modeling of correlation between labels.Based on the improvement of U-net,the feature extraction is improved by introducing Bi-directional Long Short-Term Memory(Bi LSTM);the relationship between labels is modeled using Conditional Random Field(CRF)to optimize the classification results.This method can simultaneously detect the classification and location of target arrhythmias.A total of 85,609 heartbeat records in the MIT-BIH arrhythmia database were classified according to the ANSI/AAMI EC57:2012 heartbeat classification standard.The accuracy of the improved method for heartbeat classification reached 99.11%,the specificity was 99.76%,the sensitivity was 97.21%,and the accuracy for ventricular premature beats.In addition,the accuracy of the improved method for Premature Ventricular Contraction(PVC)and Premature Atrial Contraction(PAC)increased by 9% and 47%,respectively.The experimental results show that the method can fully learn the long time series dependence and correlation between labels,which is better than the traditional U-net-based method.(2)Clinically,it is difficult for physicians to label each of the massive dynamic ECG data.Therefore,an ectopic heartbeat screening method based on RR interval and QRS wave group characteristics is proposed.As many abnormal samples of patients as possible were marked in the initial screening according to the time limit characteristics of RR interval and QRS wave group maximum amplitude and then reviewed and marked by cardiologists.The average accuracy of the proposed method in identifying abnormal ECG signal segments could reach 99.82% when 28 patients with representative ECG data from the MIT-BIH arrhythmia database were selected for the experiment.The experimental results show that the method can quickly and accurately screen abnormal ECG signals in the first place,improving the efficiency of fine labeling by physicians and laying a data foundation for the subsequent custom design of ectopic heartbeat detection models.(3)To address the problems of limited ECG data of a single person,weak generalization ability,and long training time of the generic model,a personalized ectopic heartbeat multi-objective detection method based on deep migration learning is proposed.The method fully considers the personalized characteristics of patients and uses model-based migration to apply the knowledge learned from external ECG data to a specific patient and uses the ECG data of that patient to train a personalized detection model on this basis.The short-term database(28 patients)and the long-term database(7 patients)of MIT-BIH were selected for the comparison experiments.The average accuracy of the models after migration increased by 13.82% and 4.82%,respectively and the training time was shortened to 5 minutes The results showed that this method can save the time cost of training and realize the multi-target detection of ectopic heartbeat in specific patients,which has practical application value.With the progress of living standards and drastic changes in lifestyles,bad habits such as insufficient exercise,staying up late,and being sedentary are highly likely to trigger heart pathologies.Ectopic heartbeats may occur even in a currently healthy population.When ectopic and normal heartbeats accumulate to a certain level in subjects,a highly efficient and personalized monitoring method will be developed using the method proposed in this paper.As the personalization of the model continues to escalate,the physician’s workload will be reduced,and the efficiency of the diagnostic aid will be improved while meeting the targeted diagnostic needs of the subjects.