Motion correction in CBCT imaging using gate-less model-based reconstruction of non-rigid motion and images

BACKGROUND: Cone beam computed tomography (CBCT) plays a crucial role in verifying patient positioning during radiotherapy. However, CBCT images are often blurred by motion originating from breathing, bowel motion, and/or repositioning. Conventional methods often employ gating techniques to mitigate motion artifacts by assuming periodicity, which is restricted to respiratory motion. However, this does not resolve a-periodic motion such as irregular breathing. PURPOSE: The purpose of this study is to estimate and correct for both periodic motion and irregular motion. We present a novel method for gate-less reconstruction of motion and images (CBCT-MOTUS), thereby estimating and correcting for all non-rigid 3D motion at high temporal frequency (per-projection temporal resolution of 182 ms). METHODS: Starting from a motion-corrupted image, we perform the reconstruction process by alternating between a motion estimation step and an image correction step. Motion estimation is model-based, that is, it is performed directly in projection space by comparing acquired projections to simulated projections that take motion-fields into account. The optimization process benefits from assumptions, including (i) compressing the motion-fields using B-spline parameterization to reduce the number of parameters to estimate, (ii) exploiting spatio-temporal correlation of motion using a low-rank motion model, and (iii) enforcing smoothness using spatial regularization to include a priori knowledge on motion-fields. High temporal resolution (182 ms for this scanner) is achieved as motion-fields are estimated for each acquired gantry angle. The method is validated in silico, on phantoms and on clinical in vivo acquisitions. RESULTS: The framework can estimate and correct irregular and periodic motion with high temporal resolution (per-projection temporal resolution of 182 ms). The motion-corrected images show improved image features for all experiments with reduction of motion artifacts, such as deblurring on organ interfaces. CONCLUSIONS: We have developed and validated a gate-less reconstruction method (CBCT-MOTUS) for model-based motion estimation and correction in CBCT imaging for any kind of motion. High temporal resolution (per-projection temporal resolution of 182 ms) is achieved by exploiting the smoothness and compressibility of motion using a low-rank B-spline motion model, enabling the correction for non-rigid irregular and periodic motion. These proof-of-principle results warrant further clinical validation.