A joint registration and extrapolation method for cone beam computed tomography volume of interest imaging

The high patient dose applied during acquisition of C-arm based CBCT scans limits the applicability of this imaging technique. Acquisition of only a region-of-interest can decrease the dose significantly. However, reconstruction from such truncated projection data can lead to severe artifacts in the tomographic image. For typical usages of C-arm based CBCT however, there often exists a prior CT or CBCT scan of the patient. The proposed method enables the incorporation of such priors into the reconstruction process by jointly registering and extrapolating the prior to the limited data acquisition scenario by maximizing consistency conditions across thusly extended projections. Results show improved image quality over simple heuristic extrapolation methods typically used in practice. The proposed method can be regarded as a novel hybrid registration and extrapolation algorithm similar to established DRR-based registration methods, but unlike these, working on mutually disjoint regions of the projection data.