Comparison of post-filtering and filtering between iterations for SPECT reconstruction

Iteratively reconstructed Single Photon Emission Computed Tomography (SPECT) images are known to become noisy and exhibit edge-artifacts at high iteration numbers. In order to suppress these undesirable image distortions some method of regularization has to be applied. One of the frequently applied regularization methods is low-pass filtering of the reconstructed image. In this paper SPECT images which have been filtered after a complete iterative reconstruction (post-filtering) are compared to SPECT images which have been filtered in-between each iteration step of the reconstruction (in-between filtering). The comparison of post-filtering and in-between filtering is made for two different kind of filters, a Gaussian filter and an edge-preserving diffusion filter. In order to make a fair comparison of the best possible performance of each method, the specific filters are automatically selected by varying the filter parameters until the difference between a phantom and corresponding filtered SPECT images is minimized (optimal filtering). The resulting minimum of the difference is used as the standard to compare the performance of the filter processes. The difference between optimized post-filtering and optimized in-between filtering is found to be small. In most the cases which were investigated, smoothing in-between iterations with optimized fixed kernels gives slightly less accurate results than optimized post-filtering or optimized in-between filtering. Since post-filtering is much easier to optimize than in-between filtering, post-filtering may be preferred to use in practice.