Accurate quantification of left ventricular (LV) regional deformation has been shown to be important for understanding management of patients with ischemic heart disease. Quantitative, noninvasive measurement of 3D strain properties from images has been primarily limited to special forms of magnetic resonance acquisitions, especially MR tagging, and to a lesser extent MR phase contrast velocity. In previous years of this project, we have been working on developing an image analysis strategy capable of measuring strain from any one of a variety of imaging modalities. While our approach is capable of utilizing the specialized MR tag or velocity data mentioned above, we have been able to show using a strategy based on biomechanical modeling, differential geometric features and mathematical optimization, that dense 3D LV strain information can be recovered from standard (magnitude) cine-gradient echo MR (cine- MRI) acquisitions, as well as from 3D cine-computed tomography and 3D echocardiography (3DE) at the patient's bedside. However, for our system to be fully robust to variations in image data across the subject population, as well as to reduce the time and expertise required to initialize the system (primarily in the LV segmentation step), further effort it needed. To address these issues, we intend to extend our approach to incorporate temporal continuity constraints, provide feedback and integration between the segmentation/deformation algorithm modules, integrate a notion of active contraction into our biomechanical model and pose our entire approach in a probabilistic framework, where certain models and prior-learned-information from subject populations can e more readily incorporated. We are confident that these improvements to out system will make it more robust to image variation, reduce user- interaction to a minimum and ultimately provide more accurate results within a clinically- tolerable amount of time. We will confirm these incremental improvements, further validate our approach and verify the utility of in vivo strains for distinguishing transmural injury and its extend using acute and sub-acute (3 days post injury) canine models of infarction and two imaging modalities (cine-MRI and 3DE). Finally, we plan to verify that strain values using our approach are reproducibly derivable from a set of normal and abnormal human image datasets, acquired from 3 different imaging modalities.
