Prognostic Model Of Chronic Systolic Heart Failure Based On Neural Network And Echocardiography

Heart failure is one of the reasons for poor prognosis of various cardiovascular diseases.It usually occurs in the terminal stage of cardiovascular diseases such as valvular disease,coronary artery related disease,etc.the occurs of heart failure can result in significant morbidity and mortality rate.As a stable,objective and exact index,echocardiographic parameters can provide abundant cardiac information for patients with HF,which has a unique value in the prognosis evaluation of heart failure.Objective:the objective of this research is for patients who have the chronic systolic heart failure(CSHF),using the BP neural network learning algorithm to analyze their echocardiographic parameters.And on the basis,we establish a model of patients with CSHF which can predict the prognosis.The model is useful to assist clinicians in formulating individualized monitoring standards and treatment measures for patients with HF,which can enhance the value of echocardiography in clinical work.Methods:In this study,from January 2007 to December 2008,all 313 patients with CSHF in Qianfoshan Hospital were analyzed retrospectively,they were followed up for 4-7 years.Collect the data of every patients about their 1-year readmission and survival time.The echocardiographic indexes of all patients were analyzed,left ventricular end-diastolic diameter,left ventricular ejection fraction,pulmonary arterial systolic pressure,tricuspid regurgitation,mitral regurgitation,pericardial effusion and pleural effusion confirmed by chest X-ray were collected.According to the ratio of 7:1.5:1.5,all data were divided into 3 groups:training set,test set and verification set.The program was written by MATLAB,BP neural network learning algorithm was used to build a prediction model of 1-year readmission and 3-year survival through data training.The precision and generalization capacity of the prediction model were evaluated.Results:The predict model which built through BP neural network learning algorithm can judge the precision and generalization capacity of this model according to the distribution of performance(MSE),R index and error range.After 15 iterations,each sample set can achieve the lowest mean square error(MSE).At this time,the mean square error of the validation set is very small,only 0.078321,which is below the order of 10-1.R index indicates the degree of fitting,which can evaluate the relevance between the output of the model and the real information.Each set of this model has a high degree of fitting,in which R=0.96456 for training set,R=0.96007 for verification set,R=0.91943 for test set,and R=0.95723 for the whole.The error distribution is obey normal distribution,which means the proportion of the predictand is approach to the real value.Conclusion and significance:the prognosis prediction model based on neural network learning algorithm and echocardiography can show good performance in evaluating the risk of readmission and long-term death of patients with CSHF.It is suggested that neural network learning algorithm can be used to help clinicians precisely estimate the prognosis of patients with HF and give an effective direction for the next treatment plan,and has a great clinical using prospect.