Motion Estimation and Spatiotemporal Tensor Enhanced Representation for 4D-CBCT Image Reconstruction

Four dimensional Cone-Beam Computed Tomography (4D-CBCT) has great value for motion management in radiotherapy. But high quality 4D-CBCT image reconstruction is difficult to achieve because of the lacked of projections at each phase, which result aliasing artifacts. In this work, we propose a novel joint motion estimation and spatiotemporal tensor enhanced representation (MESTER) approach for 4D-CBCT image reconstruction. To tackle the under-sampling problem of 4D-CBCT imaging, we employ spatiotemporal tensor enhanced sparse representation which tries to explore the spatiotemporal coherence of the patient anatomy in 4D data, by using inter-phase deformation vector fields (DVFs). In this approach, we divide the motion compensated 4D-CBCT images into 4D block in a non-local window and clustering similar spatiotemporal patches to a 5th-order tensor group. Then the spatiotemporal tensor dictionary sparsity representation on the tensor group is incorporated into the reconstruction cost function. The numerical simulation and preclinical data are performed to evaluate and validate the performance of this approach. Qualitative and quantitative results show that the proposed approach achieved improved image quality with less artifacts and better recovery of details.