Pablo Martinez-Legazpi

Echocardiography is the most used imaging technique in cardiovascular medicine. Despite extensive research during decades, echocardiography is limited to predict accurately critical hemodynamic variables such as pulmonary mean and capillary pressures and cardiac output. Thus, invasive right-heart catheterization is still routinely performed to obtain these measurements. We hypothesize that an appropriately designed and trained deep-learning algorithm will improve the accuracy of current hemodynamic estimations derived from ultrasound. The main objective of this proposal is to settle the bases for a deep neural network for providing clinicians with reliable noninvasive estimators of cardiac hemodynamic data.

The overarching aim of this study was to implement, validate and transfer to the clinical practice new post-processing methods of cardiac-imaging flow data, in order to widen the characterization and understanding of cardiac physiology while targeting their clinical application. These methods are based on the measurement and quantification of the blood velocity fields inside healthy and diseased human left ventricles (LV), as well as in the indices derived from them.