Mitral regurgitation (MR) is the most common type of valvular heart disease in patients over the age of 75 years in the US. Despite the prevalence of MR in the elderly population, however, almost half of patients identified with moderate- severe MR are turned down for traditional open-heart surgery due to co-morbidities. MitraClip (MC) is a recent percutaneous approach to treat MR by placement of MC in the center of the mitral valve (MV) to reduce MR. Despite the positive short-term outcomes of the MC procedure in reducing MR, the long-term outcome can be further improved if the effects of MC on both the fluid and solid mechanics of the MV and left ventricle (LV) were available at the time the clip is placed. Recently, we developed a physics-based human cardiac function simulator for the optimal design of a novel annuloplasty ring with a sub-MV element for correction of MR, as well as physics-based simulations of MC placement. The problem with these simulations, as far as clinical applications is concerned, is they are extremely time consuming (3 days to complete simulations on 96-processor cluster). One way to make these time consuming simulations clinically applicable is to run them in advance for a wide range of patient characteristics (e.g., degree of MR, size and shape of the MV and LV, etc.) and MC placements. Currently, when clinicians are ready to place the MC on the MV, they have at their fingertips real-time data on degree of MR, and size and shape of the MV and LV measured using 3D transesophageal echocardiography (RT3D-TEE). We propose the development and validation of a searchable virtual patient atlas (SVPA) that will provide the clinician with detailed predictions of patient outcomes in real time that are based on MC placement and the RT3D-TEE patient-specific data. The first 50 models in our SVPA will be created from existing RT3D-TEE datasets provided by National Heart Centre Singapore, NHCS. Then, we will use our novel-shape dictionary learning models to automatically generate 150 additional models for our SVPA. Machine learning models will be trained with the simulation data in order to create machine learning-FE (ML-FE) surrogates that can predict FE outputs directly from the model geometry. This would enable real-time prediction of patient-specific MC device outcomes. Our preliminary studies using 3D heart simulations clearly show that the main advantage with the ML model over the 3D FE model is speed (i.e., ML runs in 1 CPU second versus 3D FE model runs in 1100 CPU hours!). We will validate the outcome predictions of our SVPA using an additional 50 existing RT3D-TEE datasets with known MC patient outcomes provided by NHCS. After the outcome prediction using SVPA for each case, we will use the dataset and the measured outcome to train the original SVPA further and validate a new dataset with the original and the updated SVPA. We will select the more accurate SVPA (the original or the updated) to determine possible correlations between the primary geometrical parameters and other patient overall biometric information with the MR and optimal MC placement.
A leaky inlet valve of the major pumping chamber of the heart is the most common valvular heart disease in elderly patients. Recently, the US Food and Drug Administration approved a device that can be inserted into the patient?s heart using a catheter. The purpose of this Phase-I proposal is to develop and validate a software tool for predicting patient-specific outcomes in order to optimize this therapy for patients.
