Fifty-eighth annual meeting of the american association of physicists in medicine
SU-C-206-06: Shift-Invariant Projection and Backprojection for Helical CT Based On A Self-Consistent Coordinate
Despite its promising utility, iterative CT image reconstruction is often time-consuming, especially for helical CT, for which the X-ray projections and backprojections are computationally dominant. To overcome the insufficient memory storage for projection and backprojection matrices, they are usually computed on-the-fly with parallel implementation via GPU. In this work, we develop an efficient method to compute projections and backprojections without repeated on-the-fly computations for helical CT.
We design the shift-invariant projections and backprojections based on a self-consistent coordinate, so that the size of projection and backprojection matrices can be significantly reduced in order to be practically stored on GPU. For reconstruction accuracy, the projections and backprojections are based on two different strategies, i.e., ray-driven and pixel-driven algorithm respectively. Moreover the proposed method can be readily incorporated into standard iterative reconstruction algorithm. Rigorous mathematical analysis is carried out to establish the shift-invariance for ray-driven projections and pixel-driven backprojections.
Numerical results using GPU parallelization suggest that the proposed shift-invariant method has much improved computational efficiency from the conventional on-the-fly method, i.e.,by 3–6 fold acceleration for the projection and 3–16 fold acceleration for the backprojection using standard configuration of helical CT.
We have proposed a shift-invariant projection and backprojection method based on a self-consistent coordinate so that the projection and backprojection matrices can be stored instead of being repeatedly computed on-the-fly. The proposed method can be conveniently utilized in standard iterative reconstruction method with minimal modification. Numerical results demonstrate that the proposed method can speed up the projection by 3 to 6 folds and the backprojection by 3 to 16 folds using standard configuration of helical CT.
The authors were partially supported by the NSFC (#11405105), the 973 Program (#2015CB856000), and the Shanghai Pujiang Talent Program (#14PJ1404500).