Running prior for patient motion correction in low-dose 3D+time interventional flat detector CT

4D (=3D+time) interventional image guidance requires tomographic data acquisition during the whole intervention. However, the method, which is also called tomographic fluoroscopy or CT fluoroscopy, will be routinely accepted if the patient dose level can be kept as low as in 2D+time fluoroscopic guidance, which is the standard image guidance technique in today's interventions. To achieve this goal of enabling tomographic fluoroscopy at the same dose as projective fluoroscopy a high quality prior volume acquired before intervention is necessary to guarantee high quality update volumes during the intervention, despite the fact that these update volumes are based on a very sparse angular sampling at very low dose. Depending on the type and duration of an intervention patient motion can become a problem. Consequently, the prior volume needs to be continuously updated. We propose a running prior that adapts itself' by the combination of the two concepts registration and substitution. In the registration step a combination of affine and deformable registration adapts the prior to the current situation. In the subsequent substitution step a forward projection of the registered prior yields virtual rawdata that are densely sampled in the angular direction. The latest measured projections are used to substitute the corresponding virtual projections. A reconstruction of these substituted data yields the adapted prior. For the evaluation of the method we used an in vivo head scan of a pig acquired by a prototype volume CT system since as of today no dedicated 3D+temporal interventional CT system exists. This CT system consists of a flat detector mounted on a continuously rotating clinical CT gantry. By the running prior technique it is possible to correct for motion of the pig's head of up to 30 mm between the prior scan and intervention scan. The resulting running prior images show high image quality without introducing new artifacts. In consequence the temporal updates show less inconsistency artifacts when using the running prior technique compared to using the static prior image. We conclude that the running prior technique is superior to the static prior in case of patient motion. The temporal updates then show less artifacts without additional patient dose.