Optimization of Euclidean distance threshold in the application of recurrence quantification analysis to heart rate variability studies

Abstract

An integrated approach is proposed to solve the optimization problem of the Euclidean distance threshold ε in recurrence quantification analysis (RQA), which is increasingly applied in the study of heart rate variability (HRV). In this paper, ε is inversely computed from a given recurrence rate (REC), the percentage of recurrence points. From the inversely computed ε, two other RQA output variables: determinism (DET), the percentage of recurrence points forming diagonal line structures, and laminarity (LAM), the percentage of recurrence points forming vertical and horizontal structures, are computed out as well. The trend of DET, LAM values at different REC levels (DLR trend) is introduced to comprehensively represent the dynamic properties of a time series. Based on the DLR trend, the variation of discrimination power, represented by the average loss (or Bayes risk), of DET and LAM, at different REC values is analyzed. Surrogate techniques are used to generate reliable test data sets for the discrimination evaluation. In particular, the results show that (1) the optimal REC can be much higher than the widely used 1% REC, and (2) after the optimization, the average loss can be reduced compared to 1% REC. It is also demonstrated that the optimal ε depends on the dynamic source and RQA variables, and the DLR trend based ε optimization method can improve RQA discrimination analysis especially for the short term HRV analysis.