An Early Warning Model Of Heart Failure Risk Based On ECG Signals

Today,a series of social problems caused by cardiovascular diseases are becoming more and more serious.Heart failure,as the end point of many cardiovascular and cerebrovascular diseases,has gradually attracted the attention of related researchers.The research on automatic early warning model of heart failure risk based on ECG signals is of positive significance for alleviating the imbalance of medical resources and comprehensively improving people’s quality of life.The main research contents of this thesis include:(1)An automatic analysis neural network structure for the most common12-lead electrocardiogram is proposed to realize the multi-label classification of nine typical arrhythmias.The average score of F1 in three different test sets was above 0.886,and the highest score was 0.919.In the case of more test samples,its stability is better than other methods.(2)The classification activation map with gradient weight was used to visualize the key areas of concern in the model.It was proved that the bands of interest of the model in ECG signals were similar to the bands of interest of professional doctors to some extent.Then,the model was applied to the heart failure dichotomization task using transfer learning method,and the accuracy,sensitivity and specificity scores on the test set were 98.35%,98.20% and98.50%,respectively.(3)Three basic classifiers,namely support vector machine,logistic regression and decision tree,were constructed,and a simple feature fusion method was used for heart failure classification and classification experiments.On the one hand,it shows that multimodal features are effective for improving model accuracy,and on the other hand,it shows that the importance of abstract features changes in the classification task,which has potential research value for the prediction of heart failure.The F1 scores of the three algorithms in the heart failure typing experiment 1 were 94.74%,94.24% and 92.99%,respectively.The F1 scores of the three algorithms in the heart failure typing experiment 2 were87.00%,86.50% and 85.77%,respectively.The experimental results show that deep learning method has outstanding advantages for intelligent implementation of heart failure risk early warning.Attention mechanism,transfer learning,multi-mode fusion and other training strategies are helpful to improve the accuracy of heart failure risk early warning model.The interpretability of the model paves the way for automatic diagnosis technology to be widely recognized by the medical industry,and promotes the pioneering and innovative research in many fields,such as heart failure prediction,heart failure typing,heart failure prognosis and heart failure monitoring.