Robust 2D/3D-registration is a challenging in clinical practice. One of the difficulties is the large number of possible data sources in 3D (DynaCT, 3DDSA, CT, MR, and etc.) as well as 2D (acquisition, fluoroscopy, DSA, and etc.). Nowadays, if the physician detects a misalignment, the automatic registration can be started by pressing a button. The ideal system of the future would automatically detect a patient movement and correct the misalignment in the background. The physician would not be interrupted in his current workflow step, but still see a motion corrected overlay. However, some challenges need to be addressed for developing such a real-time, dynamic 2D/3D-registration, which is the focus of this work.

In this project aims at a general framework of dynamic 2D/3D registration for patient motion compensation, including

• Analysis of 2D image sequence and different motion

sources (e.g. patient or external devices);

• Development of algorithms for automatic patient

motion detection;

• Analysis and development of algorithms for 2D/3D

motion correction;

• Optimization of the algorithms for dynamic and real-

time motion compensation.