Cardiac motion analysis from echocardiography

Research shows that cardiac health nay be determined to a large degree from cardiac motion, particularly the motion of the primary pumping chamber, the left ventricle (LV). For this reason, a primary goal in cardiac medical image processing is to extract LV motion from image sequences in order to diagnose and assess cardiovascular disease. Many methods have been proposed for cardiac motion analysis, but most of them are only suitable for Magnetic Resonance (MR) images. The echocardiography (ultrasound imaging of the heart) has several advantages over other image modalities in availability, portability and cost. However, the images can be difficult to interpret due to intensity inhomogeneity, distortion, and speckle noise which cause mast models to fail. In this thesis, we developed a modality independent strategy to extract, spatially-dense LV strain and strain rate maps from four-dimensional (three spatial dimensions plus time) MR and echocardiographic (4DE) images.; This thesis includes four parts. In the first part, we studied the radio-frequency (RF) image variance under different motion and deformation. The experiments were carried out on a gel cube and the acquired RF image sequences were analyzed using a phase-sensitive speckle tracking method. The tracking results are well matched with the experimental setup when the displacements or the deformation is small.; Second, we developed a LV segmentation approach termed "Fuzzy-MFFD-Active Contour", The basic idea of this approach is to deform an active contour by minimizing an energy function defined by both the fuzzy reasoning technique and a Multilevel Free Form Deformation (MFFD) model. The approach was perpormed on echocardiographic images and the results were compared to the approaches using Free Form Deformation (FFD) model and level set based technique. Our approach can successfully avoid local minimum and leak through uncompleted boundary and is with the least error by comparing to manual segmentation results.; Third, we proposed a now regularization model based on the Boundary Element Method (BEM) for cardiac motion tracking. The currently available regularization models have tradeoffs related to accuracy, lattice density, physical plausibility and computation time. The BEM-based regularization model can overcome these tradeoffs. We compared the BEM based regularization model with B-spline based regularization models: FFD and Extended Free Form Deformation (EFFD) on simulated data and the results show that the BEM based regularization model is the most accurate one. We then employed this new regularization model with the Generalized Robust Point Matching (GRPM) strategy to estimate the dense displacement fields and strains from 3D LV image sequences. This shape-based BEM-GRPM system was evaluated on in-vivo cardiac magnetic resonance image sequences. All results are compared to displacements found using implanted markers, taken to be a gold standard. The approach was also evaluated on the 4D real time echocardiographic image sequences.; Finally, we proposed an integrated image analysis system. This system will estimate both the intramural displacements from speckle tracking using RF image and shape-tracked displacements on LV boundary using B-mode image. The system will then integrate all the information using the BEM based regularization model to form a spatially and temporally dense set of strains. The experiments were performed on RF image sequences around end-diastole (ED).