Deep Neural Network to Accurately Predict Left Ventricular Systolic Function Under Mechanical Assistance

Figures

• 

  Figure 1. General framework, modified from Bonnemain et al. (12) under CC-BY license. The DNN is fed with systemic and pulmonary arterial pressures, formulated in their frequency domain, as well as LVAD RPM, to predict parameters of LV systolic function. These parameters are integrated to the 0D model to retrieve the indicated additional hemodynamic parameters. Shapes and colors significance in the 0D model are: deoxygenated blood (blue elements); oxygenated blood (red elements:); compliance (circles); inertance (rectangles); resistance (facing triangles, double arrows); cardiac valves (single white arrows in heart chambers); elements affected by autoregulation (gray contour). RPM, Rotations per minute; SAP, systemic arterial pressure; PAP, pulmonary arterial pressure; DNN, deep neural network; RA, right atrium; RV, right ventricle; LA, left atrium; LV, left ventricle. Emax,lv [mmHg/ml]: end-systolic left ventricular elastance, Emax,lv,0 [mmHg/ml]: end-systolic left ventricular elastance in absence of baroregulation, GEmax,lv [mmHg/ml/(spikes/ml)]: maximum baroreceptor gain, kE,lv [1/ml]: steepness of end-diastolic pressure-volume curve.

• 

  Figure 2. Left ventricle pressure-volume diagrams for different stages of heart failure, without LVAD (A) and with LVAD for different pump speeds (B–D).

• 

  Figure 3. Performance of architecture #2 (see Supplementary Table 2) on the test set. For each of the four parameters, the exact value is plotted against the predicted one. (A) Emax,lv [mmHg/ml], end-systolic left ventricular elastance; (B) Emax,lv,0 [mmHg/ml], end-systolic left ventricular elastance in absence of baroregulation; (C) GEmax,lv [mmHg/ml/(spikes/ml)], maximum baroreceptor gain; (D) kE,lv [1/ml], slope of end-diastolic pressure-volume curve.