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

Figures

• 

  Figure 1. Distribution ranges of SD values of RR intervals for NSR and CHF groups when (A) N = 300 and (B) N = 1000.

• 

  Figure 2. The distribution ranges of SampEn between NSR and CHF groups at different setting of tolerance threshold when (A) N = 300, m = 2 and (B) N = 1000, m = 2. The symbol ‘*’means statistical significance p < 0.05 and ‘**’means statistical significance p < 0.01, using t-test.

• 

  Figure 3. Presentation of traditional thresholds and physically meaningful thresholds in time difference of RR intervals in a CHF subject.

• 

  Figure 4. Percentage of valid RR segments at m = 1, 2, 3 and 4 respectively when using rt for (A) NSR group and (B) CHF group. The length of RR time series is 300.

• 

  Figure 5. Block diagram of the proposed evaluation process for CHF analysis. NSR: normal sinus rhythm, CHF: congestive heart failure, CI: confidence interval.

• 

  Figure 6. Block diagram of the proposed evaluation process for AF analysis. AF: atrial fibrillation, NSR: normal sinus rhythm, BWL: beat window length.

• 

  Figure 7. Entropy calculation and statistical analysis contrasting rt (left) with rp (right) between NSR and CHF groups at N = 300 and m = 1 for (A) SampEn results and (B) statistical significance.

• 

  Figure 8. Entropy calculation and statistical analysis contrasting rt (left) with rp (right) between NSR and CHF groups at N = 300 and m = 2 for (A) SampEn results and (B) statistical significance.

• 

  Figure 9. Entropy calculation and statistical analysis contrasting rt (left) with rp (right) between NSR and CHF groups at N = 1000 and m = 1 for (A) SampEn results and (B) statistical significance.

• 

  Figure 10. Entropy calculation and statistical analysis contrasting rt (left) with rp (right) between NSR and CHF groups at N = 1000 and m = 2 for (A) SampEn results and (B) statistical significance.

• 

  Figure 11. ROC curve plots with AUC values for the four values of rt and rp in the RR Interval Databases for classifying NSR and CHF subjects. The top two sub-figures (A1–A2) show SampEn results using rt, while bottom sub-figures (B1–B2) show SampEn results using rp. Different combinations of m and N were used: (A1) and (B1) for m = 1 and N = 300, and (A2) and (B2) for m = 1 and N = 1000.

• 

  Figure 12. Percentage of invalid RR segments for the four values of rt in the RR Interval Databases for classifying AF and non-AF subjects. Different combinations of m and BWL were used: (A1) For m = 1 and BWL = 30, (A2) m = 1 and BWL = 60, (A3) for m = 2 and BWL = 30, and (A4) for m = 2 and BWL = 60.