Research On Identification Of Patients With Pulmonary Hypertension Based On Entropy Measure Of Heart Sound

Pulmonary hypertension(PH)is a hemodynamic and pathophysiological condition in which pulmonary artery pressure rises above a certain threshold.The cause of PH is complicated and it is difficult to diagnose in the early stages.Patients with PH from the early to the end of the disease are accompanied by a series of cardiac hemodynamic changes.Heart sounds carry a large amount of information related to cardiac hemodynamics and have important application value for non-invasive diagnosis of PH.Entropy is an efficient method to quantify the complexity of short-term and noisy physiological time series and it has been widely used in biomedical signals.This paper introduced the entropy measures to analyze the heart sound signal of subjects with and without PH and then accurately identified PH patients based on the entropy measures.The work of this paper was mainly divided into the following two parts.(1)Propose and screen the entropy measures of heart sound features.A total of 104 subjects participated in this study,with 50 PH patients and 54 healthy subjects.16 heart sound features were extracted from time,frequency,energy and amplitude domains.The study used sample entropy(SampEn),fuzzy entropy(FuzzyEn),fuzzy measure entropy(FuzzyMEn)to get 48 entropy measures of heart sound features.Mann-Whitney U test was used to screen out the entropy measures with significant difference between PH patients and healthy subjects.This paper screened out two types of entropy measures according to the effect of the age confounding factor.The first type of entropy measures neglected the influence of age factor and select the first nine entropy measures with higher significant difference.The second type of entropy measures excluded the strong age-related entropy measures by correlation analysis and selected the first nine entropy measures from the remaining entropy measures.(2)Propose a method to identify PH patients based on the selected entropy measures.The significance analysis showed that the entropy measures of PH patients and healthy subjects came from different groups,and the probability density functions(pdf)of PH patients and healthy subjects were different.In this paper,the non-parametric probability density function estimation method was used to estimate the pdf of one-dimensional entropy measures and the joint pdf of two-dimensional,multi-dimensional entropy measures for both PH group and health group.According to Bayesian theory,the probability of entropy measure occurrence was compared to distinguish PH patients from healthy subjects and the performance was evaluated by leave-one cross-validation and five-fold cross-validation.For the first type of entropy measures,the results showed that the best identification accuracy of the one-dimensional,two-dimensional,multi-dimensional entropy measures was about 0.83,0.85,0.93,respectively.For the second type of entropy measures,the best identification accuracy of the one-dimensional,multi-dimensional entropy measures was about 0.68,0.80,respectively.The analysis showed that for each kind of entropy measures,the identification results of the two cross-validation methods were highly consistent,and the identification performance of the high-dimensional joint entropy measures were better than that of the low-dimensional ones.The research in this paper showed that entropy was an effective tool and could provide important application value for early diagnosis of PH patients.