Estimation of Left Ventricular Ejection Fraction Using Cardiovascular Hemodynamic Parameters and Pulse Morphological Characteristics with Machine Learning Algorithms

Figures

• 

  Figure 1. The framework of estimating LVEF in this study includes collecting 33 parameters, extracting optimal features by RFE, and estimating LVEF by ML regression.

• 

  Figure 2. (a) The cuff pressure has three phases, inflating duration (compliance measurement), deflating duration (oscillometric measurement), and holding duration (pulse contour analysis, PCA). (b) The oscillometric signal was used to measure the blood pressure, and the pulse signal was used to measure the SV.

• 

  Figure 3. The flowchart of pulse quality analysis. Phase 1 is the pulse segmentation and the four characteristics determination. Phase 2 is to apply the decision rules for evaluating the pulse quality. Phase 3 is to mark the quality of each pulse.

• 

  Figure 4. (a) The four characteristics of pulse, (b) the pulse wave amplitude (PWA), (c) the pulse wave duration (PWD), (d) the systolic duration (SD) of pulse wave, and (e) the ratio of SD and diastolic duration (DD).

• 

  Figure 5. The decision rule (I). If any one rule is true, the pulse is poor quality.

• 

  Figure 6. The decision rule (II). If any one rule is true, the pulse is poor quality.

• 

  Figure 7. A signal segment in the duration of PCA has seven pulses. When the quality of the pulse wave is good, a high level is marked in the corresponding cycle. Otherwise, a low level is marked.

• 

  Figure 8. (a) The left ventricular ejection time (LVET) is defined at the systolic ending time (Tinst), the integral area of which is A1, (b) the ejection relaxation time (ER) is defined at the dicrotic notch time (Tdic), the integral area of which is A2, and (c) the total area is defined as A.

• 

  Figure 9. (a) The training (blue) and validation (orange) curves of ERMS, (b) the training (blue) and validation (orange) curves of R2 with SoNFIN.

• 

  Figure 10. (a) The training (blue) and validation (orange) curves of ERMS, (b) the training (blue) and validation (orange) curves of R2 with XGBoost.

• 

  Figure 11. Bland–Altman plots for (a) SoNFIN, and (b) XGBoost.

• 

  Figure 12. The estimated LVEF for the lowest and highest ERMS values by SoNFIN, the blue points are the LVEF measured by the echocardiography, green points are the estimated LVEF of training model, and red points are the estimated LVEF of the testing model, (a) patient 5, and (b) patient 9.