Quality of statistical reconstruction in medical CT

Abstract

CT images can be either reconstructed analytically or iteratively. The analytic methods, for example filtered backprojection (FBP), are known to be computationally inexpensive and highly accurate. Iterative reconstruction has the advantage that arbitrary constraints and beam profiles can be incorporated. For medical computed tomography (CT) the subgroup of statistical reconstruction algorithms seems of high interest since, similar to positron emission tomography (PET) reconstruction, better dose usage is expected. However, iterative methods are computationally extremely expensive and therefore have been applied to modalities with low amounts of data (e.g. PET) only. Recently, a highly promising study of ordered subset convex (OSC) reconstruction for medical CT applications has been published. The publication lacks a comparison to the gold standard FBP, however. Therefore, we have implemented OSC and compared it to filtered backprojection. Image quality was evaluated qualitatively and quantitatively. Simulations of a head, a thorax and a low-contrast phantom as well as the reconstruction of patient data showed a strong dependence of OSC image quality on the object shape for the suggested number of iterations. Highly eccentric objects, such as the shoulder, yield distortions and CT number deviations whereas the more circular cross-sections (as used in the original paper) can be reconstructed quite well. Analysis of the image noise levels (at equal MTF and therefore equal resolution) show advantages of OSC over FBP. Noise reductions of 20 to 30% were observed. Since medical CT images must be of high fidelity for all cases and since the reconstruction times are increased by several orders of magnitude for OSC, the statistical algorithm does not appear suitable for use in clinical CT scanners, yet.