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

Area of Science:

• Cardiovascular Physiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Assessing left ventricular (LV) systolic function with a left ventricular assist device (LVAD) presents significant challenges.
• Accurate hemodynamic assessment is crucial for managing heart failure patients with LVADs.

Purpose of the Study:

• To develop and validate a computational framework for predicting LV systolic function parameters in the presence of an LVAD.
• To provide an innovative tool for better understanding LV-LVAD interactions and optimizing therapeutic strategies.

Main Methods:

• A deep neural network (DNN) was integrated with a 0D cardiovascular model.
• The DNN was trained using systemic/pulmonary arterial pressures and LVAD rotation speeds.
• The model incorporated diverse LVAD settings and heart failure conditions to generate training data.

Main Results:

• The DNN accurately predicted end-systolic maximal elastance (E ) with a mean relative error of 10.1%.
• Other LV function parameters were predicted with a mean relative error below 13%.
• LV physiological variables (pressures, volumes, ejection fraction) were retrieved with a mean relative error under 5%.

Conclusions:

• The developed framework offers a reliable method for assessing LV hemodynamics in LVAD patients.
• This innovative tool can enhance the understanding of LV-LVAD interplay.
• The findings support improved therapeutic optimization for heart failure management.