Effect of contrast magnitude and resolution metric on noise-resolution tradeoffs in x-ray CT imaging: a comparison of non-quadratic penalized alternating minimization and filtered backprojection algorithms

Abstract

Purpose: To assess the impact of contrast magnitude and spatial resolution metric choices on the noise-resolution tradeoff of a non-quadratic penalized statistical iterative algorithm, Alternating Minimization (AM), in x-ray transmission CT. Methods: Monoenergetic Poisson-counting CT data were simulated for a water phantom containing circular inserts of varying contrast (7% to 238%). The data was reconstructed with conventional filtered backprojection (FBP) and two non-quadratic penalty parameterizations of AM. A range of smoothing strengths is reconstructed for each algorithm to quantify the noise-resolution tradeoff curve. Modulation transfer functions (MTFs) were estimated from the circular contrast-insert edges and then integrated up to a cutoff frequency as a single-parameter measure of local spatial resolution. Two cutoff frequencies and two resolution comparison values are investigated for their effect on reported tradeoff advantage. Results: The noise-resolution tradeoff curve was always more favorable for AM than FBP. For strongly edge-preserving penalty functions, this advantage was found to be dependent upon the contrast for which resolution is quantified for comparison. The magnitude of the reported dose reduction potential of the AM algorithm was shown to be dependent on the resolution metric choices, though the general contrast-dependence was always evident. Conclusions: The penalized AM algorithm shows the potential to reconstruct images of comparable quality using a fraction of the dose required by FBP. The contrast-dependence on the tradeoff advantage implies that statistical algorithms using non-quadratic penalty functions should be assessed using contrasts relevant to the intended clinical task.