Low-Dose Cone-Beam CT Iterative Reconstruction via Total Variation and Gradient Total Variation [post]

Junlong Cui, Gang Yu
2021 unpublished
The compressed sensing (CS) technique has been utilized to reconstruct Cone-beam computed tomography (CBCT) images via limited projection from under-sampled measurements. However, the condition of limited projection is an ill-posed problem. Since the CBCT image itself doesn't have sparse features, the total variation (TV) transform has been widely adopted in CBCT reconstruction. This method, which penalizes the weight of each voxel at a constant rate regardless of different spatial gradient,
more » ... not recover qualified CBCT images from ill-posed projection data. This work presents a new strategy to deal with the deficits stated above by utilizing non-uniform weighting penalization in CBCT reconstruction. The proposed new strategy combines TV and gradient total variation (GTV) for reconstruction in a hybrid weighting penalization way, where the total variation is penalized by the gradient total variation in advance. The proposed penalty not only retains the benefits of TV, including artifact and noise suppression, but also maintains the structures in regions with gradual gradient intensity transition more effectively. This study tested the proposed method by under-sampled projections of 2 objects and 2 experiments (2 digital phantom). We assessed its performance against the OS-SART method, FDK method, conventional TV method and TV+GTV method in the tissue contrast, reconstruction accuracy, and imaging resolution by comparing the root mean squared error (RMSE), the correlation coefficient (CC), the structural similarity (SSIM), and profiles intensity of the reconstructed images. The proposed method produced the reconstructed image with the lowest RMSEs and the highest CCs and SSIMs for each experiment.
doi:10.21203/rs.3.rs-1162727/v1 fatcat:fjemvbruazgl5bylbhrlekehpm