Assisted Diagnosis Of Congestive Heart Failure Based On Short-Term RR Sequence

With the improvement of science and technology,people pay more attention to their physical health except material needs.Among many diseases,heart disease seriously threatens human life.This article took congestive heart failure(CHF)as the research object,and the purpose was to build a diagnostic model of CHF for clinical application based on short-term(500 consecutive heartbeat intervals)RR sequences and artificial intelligence classification algorithms.The main work of this study was as follows:(1)The data set of CHF and normal RR sequences was constructed.Downloaded 116 RR sequences with a duration of 24 hours from the public Physionet database,including 72 healthy subjects and 44 CHF patients.After removing the abnormal beat interval in each long-term RR sequence,divided it to multiple short-time RR sequence fragments with 500 consecutive heartbeat intervals,forming two types of data sets,namely,the unbalanced data set and the balanced data set.(2)A set of physiological characteristics reflecting autonomic nerve activity of CHF were extracted,and a key feature subset was selected.This study extracted 10 timedomain features,8 frequency-domain features,10 time-frequency domain features,and 7 nonlinear complexity features to form the 35-dimensional feature space.In order to improve the generalization performance of the models and reduce the operation time,this study used a sequence forward selection algorithm to reduce the dimension of the feature space.(3)Explored the impact of different balance levels of data set and validation methods on the stability and generalization of diagnostic models of CHF.This study was based on two pairs of influencing factors,namely,the unbalanced data set and the balanced data set,ten-fold cross validation and cross-subject validation.The two-by-two cross combination analyzed the classification results in four cases.(4)Based on the feature subset of the diagnostic models of CHF,the destructive effects of CHF on autonomic nerves were analyzed.The study obtained the following results and conclusions: The small number of features does not cause the models to overfit and degrade generalization performance,based on the sequence forward selection algorithm,6 best features were selected from the 35-dimensional feature space to construct a diagnosis model of CHF,namely RMSSD,INDEX,RATIO1,SD1,?2,and SampEn.Analyzing the performance of the models in the four cases,it was found that using the balanced data set and ten-fold cross validation can greatly improve the performance of the models for indicators ACC(accuracy)and AUC(area under the receiver operating characteristic curve).Based on the balanced data set,the models had the strong stability in the recognition task because it does not favor any category.However,although the models obtained a high recognition rate through tenfold cross validation,the models are often not effective in clinical applications due to poor generalization performance.In order to improve the generalization performance of the models to the greatest extent,based on cross-subject validation,this study applied 6 features and KNN classification algorithm(K = 1)to construct a diagnosis model of CHF.The recognition rate was 94.31%,AUC reached 0.94.From a physiological perspective,analyzed that due to autonomic nervous imbalance in CHF patients,and the heart system dominated by it will have functional dysfunction.Therefore,for the indicators in the time domain,frequency domain,time-frequency domain,and nonlinear complexity was very different from the healthy group.In summary,the main advantage of this study was based on the short-term RR interval time series,in the case of ensuring the recognition rate,the minimum features were used to make the diagnosis model of CHF have the best stability and generalization.This model can provide a more effective clinical diagnosis of CHF.