Left ventricle mesh generation for deformation analysis based on 3D echocardiographic images

Mitral regurgitation (MR) affects approximately 10% of the population and can lead to serious cardiac complications. Mitral annuloplasty, a surgical repair method involving a ring-shaped implant, is a common treatment option. However, the impact of annuloplasty on the left ventricle shape and motion remains underexplored. We introduce a deep learning-based approach that enables analyzing local shape and motion properties for the assessment of the effect of mitral valve annuloplasty. The approach is based on a template mesh that gets deformed to match the lumen of the left ventricle visualized in the image. Thereby, the original topology remains unchanged, and it is assumed that vertices move to anatomically similar locations. This enables the subsequent analysis of the deformations caused by annuloplasty. We utilized a dataset of 110 3D+t transesophageal echocardiography (TEE) images from 64 patients, suffering of MR. There are pre- and post-procedure images of most patients available, and the procedure always included a mitral annuloplasty. All images were annotated by medical experts using a self-developed tool. Our model, inspired by the voxel2mesh approach, employs a U-Net architecture for segmentation and graph neural networks for mesh deformation. To compare our approach with literature, we train it on the publicly available MITEA data and reach results comparable to those reported in the original paper. When trained and evaluated on our own data, we reached a Dice Score of 0.83 and a Mean Surface distance of 2.94 mm between annotations and predicted ventricles. The deformation analysis revealed significant alterations in ventricle geometry post-annuloplasty, especially a reduced deformation between systole and diastole. We thereby demonstrated the feasibility of using mesh generation algorithms for detailed deformation analysis in echocardiographic images, potentially aiding in better understanding and treatment of MR.