Research On Heartbeat Classification Based On Convolutional Neural Network

With the development of social productivity and the improvement of people’s living standards,cardiovascular disease has gradually become one of the most important health killers in our lives.Among the many types of cardiovascular diseases,arrhythmia is the most important and most common clinical manifestation.Finding its symptoms in time is of great significance for improving the patients’ health and prolonging their lives.Therefore,the accurate diagnosis of arrhythmia is highly valued in the field of cardiovascular medicine.However,cardiovascular disease experts are faced with many problems that make the accurate diagnosis of arrhythmia clinically a time-consuming and laborious task for doctors.And it is prone to cause misdiagnosis and missed diagnosis in clinical medical diagnosis.In this case,it is becoming more and more important to design an automatic diagnosis method for arrhythmia.When developing the ECG disease diagnosis model,the models based on traditional machine learning methods have many defects.Firstly,these classification models require cumbersome noise removal steps on ECG signal data;Secondly,the traditional classification models require complex feature engineering,which is based on a large amount of manual engineering and rich expert knowledge of cardiovascular diseases.In recent years,convolutional neural networks have shown excellent performance in computer vision,reinforcement learning and natural language processing with their automated feature extraction capabilities.Therefore,this thesis aims to develop an automatic classification model for cardi’ovascular disease diagnosis based on the convolutional neural networks.The main work of this thesis has the following two aspects.The first is to propose a residual network classification model based on multi-scale convolution and attention mechanism.The main innovative work is that the attention mechanism adopted has taken into account the characteristics of the ECG signal based on the work of predecessors.The second is to propose a two-stage classification model based on adaptively adjusting the convolution scale.The first stage of this model is mainly to classify ventricular ectopic beats(V)and non-ventricular ectopic beats(N&S).And the second stage of this model is carried out to distinguish between normal beats(N)and supraventricular ectopic beats(S).Finally,the model will combine the classification opinions of the two stages to give the final classification results of normal beats(N),supraventricular ectopic beats(S)and ventricular ectopic beats(V).The experiments in this thesis are carried out on the MIT-BIH database.We have verified the advanced performance of our work in two modalities:intra-patient paradigm and inter-patient paradigm.