Investigation of non-negativity constraint on basis images in half-rotation data reconstruction in spectral CT

Abstract

In conventional computed tomography (CT) reconstruction, a non-negativity constraint on the image values is often included. Such constraint reduces the feasible set in a constrained optimization problem and often results in faster search. In basis-material-based spectral CT reconstruction, the physical grounding for the non-negativity constraint of basis image values is no longer present. It is either not used [1] or relaxed to allow only small-magnitude negative values [2]. However, such constraint can still be applied as part of the decomposition model [3] and to reduce the feasible set and search space. We have developed an optimizationbased reconstruction method that integrates decomposition into reconstruction and allows for flexible spectral data acquisition [4], [5]. The non-negativity constraint was included in the previous reconstructions as it was part of the decomposition model for the phantom used. As we recently applied the method to half-plus-half rotation scans with a dual energy phantom containing iodine and calcium inserts, streak artifacts appeared in the images surrounding calcium inserts which also showed HU value biases. This has led us to re-evaluate the need for the non-negativity constraint in this particular problem and investigate its impact on the reconstruction in terms of visual comparison and bias evaluation.