Compensating for head motion in slowly-rotating cone beam CT systems with optimization transfer based motion estimation

We present an algorithm that estimates the changing pose of a rigidly moving subject, based on tomographic projections of fiducial markers. The pose estimates can then be used as input data to motion-compensated image reconstruction methods. The proposed algorithm is in the family of so-called optimization transfer algorithms, which have been applied more frequently to tomographic image reconstruction (e.g., the ML-EM algorithm for PET). The algorithm is particularly relevant to the class of slowly-rotating, point-of-care cone beam CT systems, which have recently become prominent in otolaryngological and dental clinics for head and neck scanning. These systems are more compact and inexpensive than traditional CT systems, but because of their slower gantry rotation, are more susceptible to patient head motion over the course of the scan. A virtue of our algorithm is that it does not require a priori knowledge of the marker geometry. The relative positions of the markers are co-estimated together with their 6 degree of freedom position/orientation in each projection view. This means that the marker configuration can be deformably adjusted to fit different patients, giving considerably more flexibility in the design of the marker-to-head attachment gear than with conventional fiducial-based approaches. Our algorithm is also both fast and accurate. In a highly sub-optimal MATLAB implementation, we typically achieve motion estimates yielding sub-millimeter positioning error in 15–30 seconds. We also present motion-compensated reconstructions from real CT acquisitions as evidence of motion-estimation performance.