I. Introduction
In recent years, the use of medical images to guide interventions has been growing in popularity. Such image guidance systems typically make registered preprocedure images available during interventions. The images are static, so the organ's position and shape is also assumed to be static. For this reason, high accuracy was initially only achieved in rigid, bony regions such as the head. Organs in the chest and abdomen, such as the heart, lungs, and liver, move significantly during respiration. Therefore the accuracy of guidance information during image-guided interventions on these organs is reduced. To overcome this problem, motion models have been proposed that can predict and correct for breathing motion. However, our previous work has suggested that the accuracy of such models can be reduced during nonstandard breathing patterns, such as fast or deep breathing [1]. In this paper, we propose a novel predictive and adaptive motion model that will address this limitation.