GPU-based fast cone beam CT reconstruction from undersampled and noisy projection data via total variation

Abstract

Purpose: Cone-beam CT (CBCT) plays an important role in image guided radiation therapy (IGRT). However, the large radiation dose from serial CBCT 15 scans in most IGRT procedures raises a clinical concern, especially for pediatric patients who are essentially excluded from receiving IGRT for this reason. The goal of this work is to develop a fast GPU-based algorithm to reconstruct CBCT from undersampled and noisy projection data so as to lower the imaging dose. Methods: The CBCT is reconstructed by minimizing an energy functional 20 consisting of a data fidelity term and a total variation regularization term. We developed a GPU-friendly version of the forward-backward splitting algorithm to solve this model. A multi-grid technique is also employed. Results: It is found that 20~40 x-ray projections are sufficient to reconstruct images with satisfactory quality for IGRT. The reconstruction time ranges from 25 77 to 130 sec on a NVIDIA Tesla C1060 GPU card, depending on the number of projections used, which is estimated about 100 times faster than similar iterative reconstruction approaches. Moreover, phantom studies indicate that our algorithm enables the CBCT to be reconstructed under a scanning protocol with as low as 0.1 mAs/projection. Comparing with currently widely used full-fan 30 head and neck scanning protocol of ~360 projections with 0.4 mAs/projection, it is estimated that an overall 36~72 times dose reduction has been achieved in our fast CBCT reconstruction algorithm. Conclusions: This work indicates that the developed GPU-based CBCT reconstruction algorithm is capable of lowering imaging dose considerably. The 35 high computation efficiency in this algorithm makes the iterative CBCT reconstruction approach applicable in real clinical environments.

Total Variation Regularization Method for 3D Rotational Coronary Angiography

“…As a result, we only require that the image yields the projection data that are within the L2 distance of the actual projection data. || · || T V is the total variation norm, which is the L1 norm of the image gradient [6]. It is well known that the constrained problem (2) can be transformed to an easier unconstrained optimization problem…”

Medical image processing on the GPU – Past, present and future

“…Tight frame regularization (Daubechies et al, 2003), based on wavelets, is another alternative to TV but is also computationally demanding (Jia et al, 2011). Compared to the TV approach (Jia et al, 2010), the TF regularization yielded sharper edges and a slightly shorter processing time. CUBLAS was used for simple operations involving vectors while FBP methods were used for the matrix operations.…”

“…The reconstruction was performed by minimizing an energy functional, which can be written in terms of matrix algebra. To reduce the memory requirements, Jia et al (2010) reformulated the functional such that it could be evaluated without the need of large matrix operations. The work was extended by Tian et al (2011b), who developed a GPU-accelerated version of edge-preserving TV in order to minimize unwanted smoothing of edges.…”

“…The CUDA libraries CUBLAS and CUSPARSE can speedup these dense and sparse matrixmatrix multiplications and matrix-vector multiplications, but a general problem for CT data is how to fit the huge system matrix in the relatively small global memory. Jia et al (2010) implemented the total variation (TV) method (Rudin et al, 1992) on the GPU in order to more efficiently complete the reconstruction. Compared to the ordinary FBP algorithm, the TV method can handle undersampled and noisy projection data, which can be employed to lower the radiation dose.…”

EM+TV Based Reconstruction for Cone-Beam CT with Reduced Radiation

“…The assumption is that the reconstructed image cannot have a large total-variation (thus noise and reconstruction artifacts are removed). For related relevant work, we refer to [6][7][8][9][10] and [11] for issues related to compressive sensing. A preliminary version of this work was presented in [10].…”