Efficient 2D filtering for cone-beam VOI reconstruction

In some clinical applications, e.g., examination of deployed stents or coils during the intervention, only a small portion of the patient may be of diagnostic interest. For the sake of dose reduction to the patient, it is practicable to deploy a collimator to block radiation dose outside volume of interest (VOl). The resulting truncation, however, particularly in lateral direction, poses a challenge to the conventional reconstruction methods. The Approximated Truncation Robust Algorithm for Computed Tomography (ATRACT) is able to reconstruct images without the use of any explicit extrapolation schemes, even for highly truncated data. It is based on a decomposition of the standard ramp-filter into a local and a non-local filtering step, where the local step coincides with the two-dimensional (2D) Laplace operator and the non-local step is a 2D Radon-based filtering. In a practical implementation, the Radon-based filtering is not computationally efficient. In this paper, we present an improvement of the original ATRACT algorithm. The 2D Radon-based filtering step in the original algorithm is replaced by an analytical 2D convolution, resulting in a significant improvement in computational performance while retaining the image quality benefits of the VOl algorithm.