ROI reconstruction of motion-contaminated data with a backprojection filtration algorithm

Abstract

In order to evaluate the potential of using the backprojection filtration (BPF) algorithm for reconstructing images from motion-contaminated data, simulation studies were conducted with three virtual phantoms. The first was a uniform elliptical phantom, which underwent rotational motion during half of its temporal cycle. The second was a normal-sized modified FORBILD phantom with a dynamic insert undergoing contractile motion during 65% of its temporal cycle. This phantom was expanded to form a third phantom, whose portions extended beyond the field-of-view (FOV). For the elliptical phantom, the BPF algorithm was able to obtain an exact reconstruction of a region-of-interest (ROI) covering a portion of the ellipse, whereas the fanbeam filtered backprojection (FFBP) algorithm could not. For the normal-sized phantom, nine full-scan data sets were acquired with percents of motion-contaminated data PMCDs ranging from 17.5% to 100%. For each data set, the mean absolute difference MAD, root mean square error RMS, and correlation CORR metrics were used to assess the differences between a defined ROI reconstructed from motion-contaminated data from the same ROI reconstructed from motion-free data. The BPF algorithm using a reduced-scan interval was able to produce better MAD, RMS, and CORR metrics than both FFBP and BPF algorithms using the same short-scan interval over all PMCDs. For the expanded phantom, the presence of truncations in the data sets did not affect the overall trends of the three metrics in BPF reconstructions of eight data sets with PMCDs ranging from 15% to 100%.