Automated measurement of left ventricular ejection time via contactless suprasternal notch laser vibrometry

Figures

• 

  Fig. 1. Experimental setup for recording signals from the suprasternal notch. The vibrometer laser is mounted on the roof, positioned directly above the patient to target the laser beam onto the suprasternal notch for signal acquisition.

• 

  Fig. 2. (Left) Sample annotations on the electrocardiogram (ECG, green) and the pulse wave displacement signal from the suprasternal notch (SSN, orange), along with the SSN acceleration signal (turquoise). Left ventricular ejection time (LVET) is computed from the difference between the dicrotic notch and onset time intervals. (Right) Pulsed-wave Doppler recording with time-synchronized ECG signal (green) of the left ventricular outflow tract illustrating the estimation of LVET. LVET is determined by measuring the interval between the onset of systolic flow and the return to baseline (red).

• 

  Fig. 3. Bland–Altman plot comparing left ventricular ejection time (LVET) estimates derived from the SSN signal and the echocardiogram exam. The horizontal axis represents the average of the two measurements, while the vertical axis shows their difference (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$lvet_{ssn}$$\end{document} - \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$lvet_{echo}$$\end{document}). The green dashed line indicates the mean difference, and the red dashed lines represent the 95% limits of agreement (mean ± 1.96 SD).

• 

  Fig. 4. Bland–Altman plots comparing left ventricular ejection time index (LVETI) estimates derived from the SSN signal and the echocardiogram exam for both sexes. The horizontal axis represents the average of the two measurements, while the vertical axis shows their difference (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$lvet_{ssn}$$\end{document} - \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$lvet_{echo}$$\end{document}). The green dashed line indicates the mean difference, and the red dashed lines represent the 95% limits of agreement (mean ± 1.96 SD).

• 

  Fig. 5. Our ConvNeXts architecture adapted for 1D signal, with a probabilistic head that outputs the parameters of a Laplace distribution for the onset and dicrotic notch, the input is the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ssn_{disp}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ssn_{acc}$$\end{document} signals.

• 

  Fig. 6. Grad-CAM visualisation highlighting regions of the input signals most influential for predictions of the onset and dicrotic notch. Left panels: Activation maps for onset prediction using the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ssn_{\textrm{disp}}$$\end{document} (top) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ssn_{\textrm{acc}}$$\end{document} (bottom) signals. Right panels: Activation maps for dicrotic notch prediction using the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ssn_{\textrm{disp}}$$\end{document} (top) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts}…

• 

  Fig. 7. Flowchart illustrating the study population selection and categorization process. Patients are filtered based on clinical data, cardiovascular risk factors (hypertension, diabetes, cardiovascular disease), and clinical characteristics, with further subgroups defined by SCORE2, BMI, cholesterol levels. The final cohort includes patients at low risk for cardiovascular disease, as determined by multiple clinical criteria

• 

  Algorithm 1. find_intervals(sorted_peaks, prior_length).

• 

  Algorithm 2. predict_ssn_with_prior(peaks, signal, prior_length, prior_window_analysis).

• 

  Fig. 8. Scatter plot depicting LVET measured by SSN versus (x-axis) LVET measured by Echo (y-axis). The overlaid ordinary least squares regression line illustrates a statistically significant positive correlation between the two modalities (Pearson’s \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r = 0.67, p = 1.01e-31$$\end{document}), supporting the concordance between SSN and Echo measurements