Software-guided Operative Planning of MitraClip Placement

ABSTRACT Mitral regurgitation (MR) is the most common type of valvular heart disease in patients over age 75. Almost half of the patients identified with moderate-severe MR are not candidates for open-heart surgery due to frailty and co-morbidities. MitraClip (MC) is a recent percutaneous approach whereby a clip is placed in the center of the MR “jet” to reduce MR. Currently, when clinicians prepare to place the MC on the mitral valve (MV), they have data on the degree of MR, and the size and shape of the MV and LV measured using real-time 3D transesophageal echocardiography (RT3D-TEE). Myocardial and leaflet stresses, however, are not considered in the current MC placement strategy. Additionally, mean left atrial pressure (MLAP) has been introduced to assess long-term MC outcomes. The objective of this Fast-Track proposal is to develop and validate a machine learning (ML)-based MC placement software tool (MCP-ST) for finding the optimal MC scenarios and real-time predictions of MR, MLAP, MV leaflet stresses, and myocardial stresses, which are known to affect, respectively, the safety of device placements and the cardiac function. Accordingly, Specific Aims are proposed: Phase I) Generate a dataset of MR, MLAP, MV, and LV stress. AIM 1: Finite element (FE) structural simulation of the MV + LV of patients from a retrospective data base. To add additional retrospective patient data to our current dataset. Milestone 1: A dataset of MV geometrical parameters, MR, MLAP, MV stress, and LV stress from over 5,000 FE models obtained from over 1,000 patient images. Timeframe: 6 months from the award date. Phase II) Automating computing MR, MLAP, MV, and LV stress from DICOM data and testing the results using animal experiments. To automate the prediction of MC therapy outcomes by approximating MR, MLAP, MV, and LV stresses from echo images. Milestone 2: Manually process echo images to create a dataset of echo images. Milestone 3: ML model development to process echo images and integrate this ML model into the ML workflow that predicts MR, MLAP, MV, and LV stress. Timeframe: 12 months from the award date. Milestone 3: Assessing the performance of the respective ML workflow in swine. Timeframe: 24 months from the award date. AIM 2: Automation of echo image processing for ML predictions and Validation of Animal Studies. To develop an ML platform that predicts MR, MLAP, MV, and LV stresses from echo images and to assess its performance using swine experiments. Timeframe: 24 months from the award date. AIM 3: Prepare submission package for software as a medical device. To submit a package to FDA that includes good laboratory practices (GLP) animal study, verification and validation of software, as well as full quality documentation. Milestone 4: Assess the performance of the respective ML workflow in swine. Timeframe: 30 months from the award date. Moreover, the proposed research provides a template for developing and validating ML algorithms concerned with other cardiac conditions in line with precision/physics-based medicine.