Abstract
High quality reconstruction with interventional C-arm cone-beam computedtomography (CBCT) requires exact geometry information. If the geometryinformation is corrupted, e. g., by unexpected patient or system movement, themeasured signal is misplaced in the backprojection operation. With prolongedacquisition times of interventional C-arm CBCT the likelihood of rigid patientmotion increases. To adapt the backprojection operation accordingly, a motionestimation strategy is necessary. Recently, a novel learning-based approach wasproposed, capable of compensating motions within the acquisition plane. Weextend this method by a CBCT consistency constraint, which was proven to beefficient for motions perpendicular to the acquisition plane. By thesynergistic combination of these two measures, in and out-plane motion is welldetectable, achieving an average artifact suppression of 93 [percent]. Thisoutperforms the entropy-based state-of-the-art autofocus measure which achieveson average an artifact suppression of 54 [percent].