Upper-Bound on Dose Reduction in CT Reconstruction for Nodule Detection

Abstract

The success of X-ray computed tomography (CT) as a widespread medical diagnosis tool has led to concerns about possible harm due to the increased average radiation dose to the population. A highly effective way to reduce CT radiation dose is to use iterative image reconstruction. All major CT vendors now offer iterative reconstruction solutions and claim that they can achieve the desired image quality at only a fraction of the dose level needed with conventional filtered backprojection (FBP) reconstruction. In this paper, we empirically estimate the upper bound on the achievable dose reduction in CT reconstruction in the context of a clinical detection task. We do this based on a series of patient lung CT data sets, with and without nodules, for a large number of random noise realizations. We analyze lesion detectability in the FBP images, in which case performance is lost due to the sub-optimal reconstruction. We compare this to lesion detectability directly in the raw data domain, where by definition, all information is still available with specific location information. This, therefore, represents the best scenario any reconstruction algorithm could achieve from raw data without additional prior information. The results indicate that a threefold dose reduction is feasible relative to FBP reconstruction: specifically, the upper bound on dose reduction was estimated at 62.5% for 90% detection accuracy and 66.2% for 80% detection accuracy. Statistical weighting is responsible for approximately half of that benefit.