Effects of electrocardiogram qrs detection algorithms in heart rate variability metrics

Figures

• 

  Fig. 1. HRV metric computation pipeline: The data acquisition process records the raw ECG from a subject. After detecting the QRS complexes in the ECG, the time intervals \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_0, T_1, \ldots$$\end{document} between consecutive R-peaks are determined. These sequences of intervals are used to build the interval tachogram. A given HRV metric is then computed from the interval tachogram.

• 

  Fig. 2. Raw ECG signals. (a) Example of raw ECG signal during the jogging condition. Upper: In black, the noisy ECG recording from the loose cables setup does not contain annotations for R-peaks. Lower: The same ECG recording, but now from the chest strap, including annotations for R-peaks. (b) ECG signal of subject 16 during the sitting condition recorded with the chest strap setup. Orange plus signs mark the annotated peak, the red cross marks the only peak detected by engzee, and the green dots mark the peaks detected by the two_average method.

• 

  Fig. 3. ECG signal detection cases ranked by JF score. (a–c) Best three cases for the chest strap setup. (d–f) Worst three cases for the chest strap setup. (g-i) Best three cases for the loose cables setup. (j-l) Worst three cases for the loose cables setup. Orange plus signs mark the annotated peak, and the green dots mark the detected QRS complex.

• 

  Fig. 4. CCC for HRV metrics MeanNN, TINN, LF, HF, SD2, and SD1SD2, and the chest strap setup. The dotted lines show the concordance thresholds: “very strong” above 0.8, “moderately strong” between 0.5 and 0.8, “fair” between 0.3 and 0.5, and “poor” for less than 0.3.

• 

  Fig. 5. Interpretation of CCC. Each plot corresponds to the chest strap setup, a QRS detector, and HRV metrics (a) MeanNN, (b) LFHF, and (c) TINN. Each point represents a subject, with horizontal and vertical coordinates corresponding to HRV values from annotated and QRS detector-derived R-peak locations, respectively. A linear regression line is included, with annotated and detected HRV values as independent and dependent variables. Conditions are color-coded with CCC values indicated in the legend. A 45-degree dotted line indicates perfect agreement.

• 

  Fig. 6. Correlation between JF and CCC. Each panel shows a scatter and regression plot for each HRV metric. The plot includes one point per condition and QRS detector. Blue and orange points correspond to the chest strap and loose cables setup, respectively. The lines represent a simple regression model for each setup. The corresponding \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R^2$$\end{document} value for the chest strap and loose cables setup is shown in parentheses in the title.