Atlas-based mosaicing of 3D transesophageal echocardiography images of the left atrium

3D transesophageal echocardiography (TEE) is routinely used for planning and guidance of cardiac interventions. However, the limited field-of-view dictates the compounding of multiple images for visualization of large structures, e.g. the left atrium (LA). Previously, we developed a TEE image acquisition protocol to capture the LA in six views. This paper proposes the use of fused TEE data of other patients as atlases to guide the registration of these views in a novel method, denoted atlas-based mosaicing (ABM). In this method, each TEE view is rigidly registered to a fused TEE atlas image of another patient, followed by pairwise registration of neighboring TEE views. Experiments on ten patient data sets and five atlas sets reveal the superior performance of ABM compared to regular pairwise registration. Similar to atlas-based segmentation approaches, ABM has the power of combining the outcome of multiple atlases resulting in accurate registration of images with little overlap.