Deformable Registration for Re-Contouring and Phase Prediction in 4D CT¹

Deformable image registration is currently widely used in four-dimensional (4D) radiotherapy applications including internal anatomy motion tracking, delineating and dose tracking. In this work, two optical flow based deformable registration algorithm are studied, and we have designed two experiments to evaluate their efficiency. In the first study, deformation fields are calculated between reference phases and other phases in 4D CT images, and the contours on the reference phase are deformed to corresponding phases. In the second study, motion fields between distinct phases: end-inspiration and end-expiration are interpolated to generate images in other phases. The algorithms are validated by quantitative comparison, and visual inspections of the positions and shapes of these deformed contours agree well with human judgment. Furthermore, the deformable registration methods have significant potential for phase prediction to reduce the amount of cine data acquisition in 4D CT.