Research On Multiple Classification Of Heart Diseases And Severities Based On ECG Signals

In today’s society,there are many diseases that pose a serious threat to human life and can even lead to death.These diseases are still difficult to detect and treat in a timely manner with current technology.Among them are the numerous cardiovascular diseases.It is well known that the incidence and mortality rates of cardiovascular disease are increasing.In recent years,they have become the leading cause of death among all diseases.Typical cardiovascular diseases such as coronary artery disease(CAD),myocardial infarction(MI)and congestive heart failure(CHF)are sometimes difficult to distinguish as some of the symptoms of diseases are similar at the time of presentation.Failure to accurately identify the disease and determine its severity in a timely manner can delay optimal treatment and lead to serious consequences.Therefore,timely and accurate diagnosis of cardiovascular disease is essential.The electrocardiogram is an intuitive and quick diagnostic tool for cardiovascular diseases,but doctors have limited energy and experience to accurately identify diseases.The rapid development of computer technology has enabled rapid and accurate diagnosis through high accuracy calculations,thus providing faster and more effective assistance to physicians.This paper proposes the design of a computer-aided diagnosis system based on deep learning algorithms using ECG signal data,taking into account the actual medical clinical situation,to achieve accurate diagnosis of various cardiovascular diseases such as heart failure and the classification of the severity level of heart failure.The main research content of this paper is summarized as follows.(1)This paper constructs a multiclassification diagnostic model for cardiovascular diseases such as heart failure based on common two-lead ECG signals.Most studies have performed intra-patient experiments for categorical diagnosis,whereas the actual reality of the conditions is that the tests are basically performed between patients.Most studies ignore this part.Inter-patient experiments can be more closely applied in a practical way.Therefore,this paper proposes a deep learning network model with dual feature extraction.The network model first adapts to subsequent feature extraction by applying a modest dimensionality reduction to one-dimensional convolutional layers,and continues to apply Resnet residual network block layers and Transformer encoder layers in turn to extract key and relevant abstract features.In inter-patient experiments,the model was able to quickly distinguish between normal subjects,patients with coronary artery disease,patients with myocardial infarction,and patients with heart failure,with an average accuracy,sensitivity,positive predictive value,and specificity of 97.48%,93.54%,96.30%,and 97.89%,respectively,for diagnosing heart failure disease.Inter-patient experiments can be more closely applied in a practical way.Therefore,this paper proposes a deep learning network model with dual feature extraction.The network model first adapts to subsequent feature extraction by applying a modest dimensionality reduction to one-dimensional convolutional layers,and continues to apply Resnet residual network block layers and Transformer encoder layers in turn to extract key and relevant abstract features.In inter-patient experiments,the model was able to quickly distinguish between normal subjects,patients with coronary artery disease,patients with myocardial infarction,and patients with heart failure.The mean accuracy,sensitivity,positive predictive value,and specificity for diagnosing heart failure disease are 97.48%,93.54%,96.30%,and 97.89%,respectively.In addition,the model performs well in unbalanced datasets and has good noise robustness.(2)In this paper,deep neural network(DNN)is constructed and this DNN is trained using 3 kinds of heart rate variability(HRV)segments without manually extracting features.The performance of the model is evaluated in two ways.The best performance is attained by segmenting HRV signals into 2048 sampling numbers.The accuracy,sensitivity and specificity are 92.65%,91.62%,91.84% respectively using 10-fold cross-validation and the accuracy,sensitivity and specificity are 87.09%,75.02%,91.19% respectively using grouping validation.Finally,the model can divide CHF patients into 4 levels based on the New York Heart Association(NYHA)classification rules: NYHA class Ⅰ,NYHA class Ⅱ,NYHA class Ⅲ,NYHA class Ⅳ.