On the Value of the Non-negativity Constraint in CT

Haase V, Hahn K, Schoendube H, Stierstorfer K, Maier A, Noo F (2019)

Publication Language: English

Publication Type: Conference contribution, Abstract of lecture

Publication year: 2019

Conference Proceedings Title: 2019 IEEE Nuclear Science Symposium and Medical Imaging Conference Proceedings (NSS/MIC)

Event location: Manchester GB

DOI: 10.1109/nss/mic42101.2019.9059667

Abstract

Model based iterative reconstruction (MBIR) has attracted a lot of attention in X-ray computed tomography (CT). A strength of MBIR over classical filtered backprojection is its ability to apply constraints over the voxel values, which can be critical to improve image quality. Given that the linear attenuation coefficient of X-rays is non-negative, applying a non-negativity constraint appears very natural. Indeed, most MBIR-related publications in CT invoke it. However, there is little to no information in the literature on the intrinsic value of the non-negativity constraint. In this work, we shed light on this question in the context of two challenging imaging scenarios: (i) heavy truncation, (ii) photon starvation due to a metal implant. Real CT data sets are used, and the effect of the constraint is examined in terms of image similarity and closeness to a preferred ground truth. Additionally, convergence properties are examined. The reconstruction is performed using a provably converging algorithm applied with a large number of iterations to nearly reach convergence, and is also performed using ordered subsets to obtain a result in a manner that is more practical for clinical routine applications. Our results show that the non-negativity constraint can be both beneficial and detrimental depending on the imaging scenario. However, the observed differences tend to be much smaller than the overall level of inaccuracy in the image. We also find that the non-negativity constraint can prevent divergence when using ordered subsets, but this gain does not translate into a satisfactory reconstruction. Altogether, we conclude that strong value for the non-negativity constraint is difficult to demonstrate. This constraint could thus be discarded in favor of other constraints or utilization of algorithms that cannot handle it.

Authors with CRIS profile

Viktor Haase Lehrstuhl für Informatik 14 (Bild- und Sprachverarbeitung) (LME) Andreas Maier Lehrstuhl für Informatik 5 (Mustererkennung)

Involved external institutions

Siemens AG, Healthcare Sector DE Germany (DE) University of Utah US United States (USA) (US)

How to cite

APA:

Haase, V., Hahn, K., Schoendube, H., Stierstorfer, K., Maier, A., & Noo, F. (2019). On the Value of the Non-negativity Constraint in CT. Paper presentation at 2019 IEEE Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC), Manchester, GB.

MLA:

Haase, Viktor, et al. "On the Value of the Non-negativity Constraint in CT." Presented at 2019 IEEE Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC), Manchester 2019.

BibTeX: Download