Are uniform phantoms sufficient to characterize the performance of iterative reconstruction in CT?

Abstract

The evaluation of dose reduction potential from iterative reconstruction (IR) algorithms is an area of ongoing research in CT. The non-linearity of IR algorithms poses challenges to using traditional image quality metrics. Past attempts to evaluate iterative algorithms have relied on measurements taken from uniform background phantoms. In this study, noise is evaluated in CT images with no texture (water), fine texture (sponge + water), and gross texture (acrylic spheres + water. Images were reconstructed with a commercially available IR algorithm (SAFIRE 5) and filtered back projection (FBP). Noise was characterized in terms of its magnitude (pixel standard deviation) and stationarity across reconstruction algorithms and background types using an image subtraction technique. The IR algorithm reduced noise magnitude across all dose levels by 66 ±1%, 47 ±3%, and 29 ±4% in uniform, finely textured, and grossly textured backgrounds respectively. Noise was reasonably stationary in uniform FBP and IR images. For IR images with gross texture, pixel noise was 29 ±4% lower in acrylic sphere regions compared to water regions in the same slice. For FBP images, there were negligible differences between acrylic sphere and water regions in terms of pixel noise. This object-dependent noise is a feature of SAFIRE reconstruction that has not been previously reported.