Recently, some computer aided diagnoses (CAD) systems has been introduced until now. Temporal subtraction technique, as one of the CAD systems, can emphasize temporal changes of interested regions by subtracting a previous image from a current one, which are images of a same patient. A previous image is deformed to suit a current image considering a variation of pose and inhalation differences when a patient accepts CT inspection at different times and subtracted from a current image. However, accurate registration is difficult because organizations such as a vessel are distributed three-dimensionally in a lung region. It is caused that each shift vector (shift amount and direction of a pixel) is less consistency to surrounding shift vectors in some area of a lung region by mis-registration and many artifacts are remained on temporal images. In this paper, we propose a method of registration based on GGVF considering information of structures in a lung regions and smoothing of shift vectors, and aim to improve registration and reduce artifacts in a subtraction image.
