A New Physically Meaningful Threshold of Sample Entropy for Detecting Cardiovascular Diseases

Sample Entropy (SampEn) is a popular method for assessing the regularity of physiological signals. Prior to the entropy calculation, certain common parameters need to be initialized: Embedding dimension m, tolerance threshold r and time series length N. Nevertheless, the determination of these parameters is usually based on expert experience. Improper assignments of these parameters tend to bring invalid values, inconsistency and low statistical significance in entropy calculation. In this study, we proposed a new tolerance threshold with physical meaning ([Formula: see text]), which was based on the sampling resolution of physiological signals. Statistical significance, percentage of invalid entropy values and ROC curve were used to evaluate the proposed [Formula: see text] against the traditional threshold ([Formula: see text]). Normal sinus rhythm (NSR), congestive heart failure (CHF) as well as atrial fibrillation (AF) RR interval recordings from Physionet were used as the test data. The results demonstrated that the proposed [Formula: see text] had better stability than [Formula: see text] , hence more adaptive to detect cardiovascular diseases of CHF and AF.