Nonrigid registration methods for measuring regional left atrial motion using retrospective gated CT images

Atrial fibrosis in patients suffering from atrial fibrillation (AF) is known to decrease atrial contractility and increase atrial wall stiffness. Regional atrial deformation can be measured using strain and strain rate. MRI feature tracking (FT) and echocardiography speckle tracking are used to measure atrial strain, and both modalities' strain measurements have been found to indicate local fibrosis in the literature. However, echo-derived strain measurements are often performed using 2D imaging, which fails to capture full atrial motion and compromises the strain measurements. MR images typically have in-plane spatial resolution of the same order as left atrial wall thickness, which may reduce confidence in imaging and tracking the left atrial endocardial border.

In this project we propose to use high resolution contrast-enhanced retrospective gated CT images to measure atrial wall deformation in 3D. We plan to optimise the Temporal Sparse Free Form Deformation (TSFFD) and Voxel Morph methods to track left atrial (LA) motion in gated CT images. In order to create registration errors for the LA anatomy, we will create LA segmentations and surface meshes at the initial cardiac phase, and deform these to the proceeding phases. Segmentations and meshes at subsequent phases provide a 'ground truth' to the deformed outputs. Global metrics (average surface distance, dice score coefficient and Hausdorff distance) and local metrics (point-to-point errors between PV ostia centre of mass) will be used to quantify the registration accuracy. 5-fold cross validation will be used to fine-tune the registration method parameters to avoid overfitting to data and achieve generalised performance. We will calculate regional LA strains (area and fibre strains) using the Universal Atrial Coordinate system, and compare strains between heart failure patients with and without AF.