Non-Invasive Estimation of Right Atrial Pressure Using a Semi-Automated Echocardiographic Tool for Inferior Vena Cava Edge-Tracking

Figures

• 

  Figure A1. Bland–Altman plots comparing estimated (by LM) and measured RAP on our dataset split into different groups of patients.

• 

  Figure A2. Bland–Altman plots comparing estimated (by LM) and measured RAP on our dataset split into patients with sinus rhythm or with irregular pacing.

• 

  Figure 1. Examples of US video frames of three subjects (A–I), IVC borders and midline highlighted in red and white, respectively, (B,E,H) and vessel diameters over time (C,F,I) obtained by the semi-automated algorithm, together with the values of CCI, mean diameter, and RAP. The three subjects belong to different RAP classes: low (A–C), intermediate (D–F), or high (G–I).

• 

  Figure 2. Comparison of IVC properties measured by the operator or estimated by the semi-automated algorithm. (A) Scatter plot and (B) Bland–Altman representation of IVC size. (C) Scatter plot and (D) Bland–Altman representation of IVC pulsatility (in terms of caval index after sniff measured by the operator and CCI obtained from short US videos).

• 

  Figure 3. Box and whisker plots of the different features, dividing the dataset in the 3 RAP classes (1 = low, 2 = intermediate, 3 = high). The average of Fisher ratios (FR) is also indicated. CI: cardiac index; CCI: cardiac caval index; eGFR: estimate glomerular filtration rate; hs-TnT: high sensitivity troponin T; D: inferior vena cava diameter; LAVI: left atria volume index; LVEF: left ventricular ejection fraction; LVMI: left ventricular mass index; NT-proBNP: amino terminal pro brain natriuretic peptide; PAWP: precapillary artery wedge pressure; PVR: pulmonary vascular resistances; RAd: right atrial diameter; RVd: right ventricular diameter; RV-FAC: right ventricular fractional area change; TAPSE: tricuspid annular plane excursion; TVR: tricuspid regurgitation velocity; TRd: tricuspid regurgitation degree.

• 

  Figure 4. Classification of the subjects obtained by (A) guidelines, (B) LM, and (C) SVM. To avoid superposition of different cases, the estimated values are randomly shifted around the categorical output provided by guidelines and SVM (on the other hand, the continuous output of the linear model is shown in (B) as a scatter plot). RAP classes are: 1 = low, 2 = intermediate, 3 = high. LM: linear model; RAP: right atrial pressure; RHC right heart catheterization; SVM: support vector machine.

• 

  Figure 5. Confusion matrix (CM) of leave-one-out cross validation of (A) guidelines, LM (B), and SVM model (C). RAP classes are: 1 = low, 2 = intermediate, 3 = high. The following average performance indexes were obtained. Guidelines: sensitivity 54.0%, specificity 76.4%, F1 score 49.3%, precision 55.4%. LM: sensitivity 63.7%, specificity 82.2%, F1 score 62.7%, precision 72.6%. SVM: sensitivity 62.3%, specificity 82.3%, F1 score 60.0%, precision 62.2%.