Modelling Periodic Motion for Background Subtraction in X-Ray Imaging

摘要

Real-time guidewire segmentation in fluoroscopic images is a necessary prerequisite for automated monitoring and navigation systems to support minimally invasive en-dovascular interventions. In this paper we propose a background model that is specifically adapted to the X-ray modality and that is able to model periodic background movement, for example induced by heartbeat or breathing. The proposed background model requires about 1-2 periods of the periodic background motion to learn the motion parameters and can then be used to segment objects of interest such as guidewires or needles. We demonstrate the feasibility of the proposed method on one recorded X-ray sequence.