Automated assessment of cardiac morphology and function: An integrated B-spline framework for real-time segmentation and tracking of the left ventricle

With the increasing prevalence of cardiovascular diseases, the need for advanced diagnosis systems that are able to detect early cardiac dysfunction are now needed more than ever. Real-time 3D echocardiography has made its way into clinical practice over the last decade and is now generally accepted as a competitive alternative to cardiac magnetic resonance imaging for volumetric assessment of left ventricular morphology and function. Nonetheless, there is still a need for software tools enabling a faster, more accurate analysis, while reducing the burden to the operating physician and minimizing the intra and interobserver variability of the measured indices. This was the key medical problem driving our efforts. The fundamental methodological innovation presented directly focuses on the inheritance of desirable properties of level-set oriented algorithms, such as advanced regionbased segmentation energies and fast/robust interface evolution via B-spline filtering, while dramatically reducing the computational load associated with 3D segmentation problems. This was possible through a B-spline formulation of the original Active Geometric Functions framework, and by further exploring the mathematical link between explicit and implicit formulations for the image segmentation problem. The resulting algorithm provides a very competitive balance between accuracy and computational burden, enabling real-time 3D segmentation applications. Building on this technical breakthrough, we have sought to extensively validate its use for LV volumetric assessment in a clinical setting , while at the same time dealing with some fundamental limitations such as its initialization, the user interaction with the segmented surface and the integration of temporal information in the boundary identification and tracking problems. This results in a coordinated suite of algorithms targeting real-time, fully automatic segmentation and tracking of the LV during the cardiac cycle.