Optical flow vs Bspline image registration for respiratory motion modeling

Respiratory motion modeling is a key approach that has the potential to improve the efficiency of radiation therapy, especially in the thorax and the abdomen areas. Such modeling may help delivering lower dose to the healthy tissues and higher, more concentrated dose to the tumor target under breathing-induced motion. These respiratory models under investigation here relate the respective organs and tumors motion to an external acquired respiratory signal. The first step for building such models consists in extracting the tumor and healthy tissues motion using a four dimensional computed tomography (4D CT) image registration algorithm. The efficacy of the registration algorithm directly affects the accuracy and the robustness of the derived respiratory motion models. In this study, we compared and evaluated the use of two registration algorithms, namely the elastic Bspline and the optical flow, for the process of building the patient specific respiratory motion model. From the same datasets of 10 4D CT patients acquisitions carried out with an external respiratory signal, two different patient specific models were built, one for each registration approach. Similar performance in accuracy was obtained using the two different registration algorithms with however an advantage for the Bspline approach with respect to the computational time involved in building the model.