Level Set based Real-time Anatomy Tracking

Real-time motion estimation is always challenging especially under MR imaging modality due to its low SNR. This study proposes a novel level set based method to estimate the anatomical boundary motion in real time. We construct a correspondence map on the spatial-temporal domain and advect it with the underlying dynamic level set function that delineates the organ of interest. In real-time, such correspondence map is estimated and moving trajectories for the anatomy boundaries are evolved. Unlike conventional level set-based registration, where velocity is assumed to be normal, we solve for the velocity with tangential component by minimizing an elasticity energy. The proposed method was tested with renal MR image sequences under both synthetic and physiological motion: the former generated by artificially translating a static reference MR image and the latter acquired with EPI pulse sequence under heavy breathing. With the new scheme to handle tangential velocity component, the proposed method is capable to estimate motion with good accuracy and/or physiological implication. In the synthetic motion test where ground-truth velocity is accessible, it significantly improved the accuracy of the motion estimation; in the real-time MR sequence test, the reconstructed motion field is smooth and exhibits periodic temporal behavior in accordance with respiratory motion and spatial variation in agreement with dynamics for renal physiology. In conclusion, the proposed scheme improves the estimation accuracy and provides insights about the spatial-temporal characteristics of anatomical and physiological motion. Such information will be incorporated into real-time motion adaptive radiotherapy to improve cancer target coverage and normal tissue sparing.