A framework for automatic model-driven 2D echocardiography acquisition for robust respiratory motion estimation in image-guided cardiac interventions

Respiratory motion limits the range of clinical applications for image-guided cardiac interventions, causing misalignments between the static pre-procedural information used for guidance and the moving intra-procedural cardiac anatomy. We propose a novel framework for automatic model-driven 2D echocardiography (echo) acquisition for robust respiratory motion estimation. The framework uses a MR-derived respiratory motion model to automatically control the 2D echo acquisition, for example by using a robotic arm. The motion model is also used to estimate the uncertainty associated with respiratory motion variability, which is resolved by using the model-driven 2D echo imaging in the Bayesian estimation technique previously proposed by our group for use with 3D echo [1]. For a thorough accuracy validation of the proposed framework, MR-derived gold standard motion fields and echo images of 4 volunteer datasets are employed. In this work, two clinically relevant 2D echo imaging planes were considered and compared. We show that the Bayesian estimation technique is accurate and robust when using 2D echo images, with improvements in motion estimation for each dataset of 41.5%, 46.5%, 38.1% and 18.9% over the best comparative method. Accuracy improvements due to the model-driven echo acquisition were achieved for each dataset, with improvements of up to 30.5%.