Research On Classification And Identification Of Cardiovascular Diseases Based On Deep Learning

With the gradual increase in the rhythm and pressure of people's lives,cardiovascular disease has become one of the major killers that threaten people's lives and health.Electrocardiogram(ECG)can reflect the health status of human heart and is widely used clinically for cardiovascular diseases.However,with the dramatic increase in the volume of ECG data,doctors spend most of their time on distinguishing abnormal ECGs rather than focusing on analyzing abnormal ECGs.At present,more and more computer-aided diagnosis technology has been integrated into the medical field,resulting in a variety of algorithms and models that are suitable for ECG data.This is of great significance for achieving hierarchical diagnosis and treatment and rational allocation of medical resources.The analysis and diagnosis of ECG signals involves a lot of research content.At present,there are still many deficiencies,mainly including the following difficulties:1.At present,many research methods only obtain good evaluation results on the standard ECG public database.However,the features of the model mining are more of the characteristics of the data set itself,and are not the essential characteristics of the ECG.At that time,the effect is significantly reduced,and the generalization ability of the algorithm is difficult to guarantee.2.The classification performance of traditional algorithms usually depends on the manual extraction of ECG features,and then fed into the designed model for classification and identification.However,due to the large differences in ECG waveform characteristics of different people,the extracted features cannot truly reflect the intrinsic properties of the ECG,and feature selection is difficult.The ultimate goal of the cardiovascular disease classification model is to fit into a highly complex nonlinear decision function.In this paper,the powerful learning ability of deep learning is used to extract the spatial and temporal features of the original ECG data layer by layer.The feature extraction and classification process are fused together.Finally,the decision function is used to automatically classify the ECG data.And when using large,annotated ECG datasets,ECG classification models based on deep learning can approach and possibly exceed human diagnostic capabilities.For the current difficulties,the main research work of this paper is divided into the following sections:1.This article is based on wavelet transform feature extraction algorithm theory,using bior 2.6 mother wavelet and 8-level scale factor,ECG signal denoising,get a higher signal to noise ratio,and by extracting ECG of various dimensions The features were characterized and the support vector machine was used as a classifier.The ECG classification and recognition model was established to realize the classification and recognition of five types of ECG signals.This simulation experiment explored the feature expression mechanism.The ECG classification and recognition model with robustness needs multi-dimensional extraction to the essential characteristics of ECG.2.This paper proposes a 50-layer Convolutional Neural Network classification model based on the deep residuals framework.After more than 150,000 clinical ECG evaluations in the Chinese cardiovascular disease database,accurate classification of accurate abnormal cardiovascular disease is achieved.The rate reached 89.43%.The network model can significantly increase the depth to improve the classification accuracy of the model,and the model can autonomously learn the essential characteristics of the ECG,which can effectively replace the features of traditional manual selection by experts and time-consuming experts.3.Since the ECG signal is a continuous sequence,this paper proposes a fine-tuned variant long-term and short-term memory hybrid network that combines the excellent receptive fields of convolutional neural networks and the memory advantages of long and short-term memory networks to automatically learn ECGs.The signature of the signal is expressed,and the extracted feature sequence is sent to the long-term and short-term memory network.The valuable memory stored in the past and the contextual state at the current moment can be integrated,and the strong correlation between ECG signal points can be further explored.Handle ECG beats of various lengths.After more than 150,000 standard 12-lead short-term clinical ECG records were evaluated,the accuracy of model classification was 93.39%.In the clinical application,it has achieved highly efficient and accurate classification of cardiovascular diseases and has a very practical significance for the clinical diagnosis of cardiovascular diseases.