4D-CBCT Reconstruction via Motion Compensataion Learning Induced Sparse Tensor Constraint

The challenge in 4D-CBCT reconstruction is the lacked of projections at each phase, which result in a reconstruction image full of streak artifacts and motion blurring with the conventional analytical algorithms. Aiming at improving the overall quality of 4D-CBCT images, we propose a motion compensation learning induced sparse tensor constraint reconstruction (MCL-STCR) method which tries to explore the high correlation in 4D-CBCT images for different phases. In this method, we additionally conduct motion compensation on the 4D-CBCT volume by using trained image motion compensation convolutional neural network (CNN). Then the compensated 4D-CBCT volume is viewed as a pseudostatic sequence, of which the sparse tensor constraint was imposed on. The cost function of proposed MCL-STCR is optimized by a variable splitting algorithm. Its comparison to other methods through simulated dynamic phantom and lung data experiments demonstrated that the proposed method can lead to a promising improvement of 4D-CBCT reconstruction.