Accelerated statistical iterative reconstruction
A method for reconstructing a volume image of a subject, executed at least in part by a computer, accesses projection x-ray images of the subject and performs a volume image reconstruction using the x-ray images by iteratively performing alternating steps of a tomographic update, where a gradient based algorithm having a preconditioner is used to update the volume image reconstruction and a spatially varying denoising that is a function of the preconditioner. The method displays, stores, or transmits the volume image reconstruction.
1. A method for reconstructing a volume image of a subject, the method executed at least in part by a computer, comprising:
accessing a plurality of projection x-ray images of the subject;
performing a volume image reconstruction using the plurality of projection x-ray images by iteratively performing:
(1) a tomographic update step, where a gradient based algorithm having a preconditioner is used to update the volume image reconstruction; and
(2) a spatially varying denoising step, where the preconditioner is used to spatially vary the strength of a denoising algorithm; and
displaying, storing, or transmitting the volume image reconstruction.
2. The method of claim 1 wherein the spatially varying denoising step uses a bilateral filter, non-local mean, or a Gaussian denoising method.
3. The method of claim 1 wherein the tomographic update step uses a simultaneous algebraic reconstruction technique (SART).
4. The method of claim 1 wherein the preconditioner comprises a diagonal matrix.
5. The method of claim 4 wherein a ith diagonal element of the diagonal matrix D is the average of a statistical weights that are back projected into a ith voxel by the volume image reconstruction.
6. The method of claim 1 further comprising using momentum function to accelerate the tomographic update step.
7. The method of claim 1 wherein the preconditioner is used to determine the number of iterations of the spatially varying denoising step.
8. The method of claim 1 wherein performing the spatially varying denoising step comprises using an optimization function.
9. The method of claim 1 , wherein an ordered subset method is applied to the tomographic update step, the denoising step, or both the tomographic update and denoising steps.
10. The method of claim 1 wherein the tomographic update step uses an algorithm taken from the group consisting of gradient descent, conjugate gradient, coordinate descent, algebraic reconstruction technique, simultaneous algebraic reconstruction technique, Cimmino's method for iterative reconstruction, and component averaging.
11. The method of claim 1 further comprising using sequences of sub-iterations with different ordered subsets of the plurality of projection x-ray images.
12. The method of claim 1 , wherein the denoising step includes using a spatially varying regularizer to denoise the image reconstructed in the tomographic update step.
13. The method of claim 12 , wherein the regularizer is at least one of a total variation, Huber, Tikhonov, q-GGMRF, FAIR potential, generalized total variation, higher order derivatives, PICCS, K-SVD, non-local total variation.
14. The method of claim 1 , wherein the denoising step includes using a spatially varying denoising filter to denoise the image reconstructed in the tomographic update step.
15. The method of claim 14 wherein the denoising filter is at least one of a bilateral filter, non-local mean, BM3D, median, or a Gaussian denoising method.
16. The method of claim 1 , wherein the reconstruction comprises a Bregman iteration step or a momentum step.
17. The method of claim 1 , wherein the tomographic update step uses a statistical noise model incorporating captured photon statistics.
18. The method of claim 1 , wherein the tomographic update step uses preprocessing steps for one or more of detector response and normalization, scatter and beam hardening correction, and denoising of the measured projections.
19. A method for reconstructing a volume image of a subject, the method executed at least in part by a computer, comprising:
accessing a plurality of projection x-ray images of the subject;
performing a volume image reconstruction using the plurality of projection x-ray images by iteratively performing:
(1) a tomographic update step, where a simultaneous algebraic reconstruction technique and a preconditioner is used to update the volume image reconstruction; and
(2) a spatially varying denoising step, where a spatially varying denoising algorithm is a function of the preconditioner; and
displaying, storing, or transmitting the volume image reconstruction.
20. A method for reconstructing a volume image of a subject, the method executed at least in part by a computer, comprising:
accessing a plurality of projection x-ray images of the subject;
performing a volume image reconstruction using the plurality of projection x-ray images by iteratively performing:
(1) a tomographic update step, where a gradient based algorithm having a preconditioner is used to update the volume image reconstruction;
(2) an acceleration step, where a momentum function is used to accelerate the tomographic update step; and
(3) a spatially varying denoising step, where a spatially varying denoising algorithm is a function of the preconditioner; and
displaying, storing, or transmitting the volume image reconstruction.