Comparative performance assessment of beam hardening correction algorithms applied on simulated data sets

Beam hardening artefacts deteriorate the reconstructed image quality in industrial computed tomography. The appearances of beam hardening artefacts can be cupping effects or streaks. They impair the image fidelity to the object being scanned. This work aims at comparing a variety of commonly used beam hardening correction algorithms in the context of industrial computed tomography metrology. We choose four beam hardening correction algorithms of different types for the comparison. They are a single-material linearization algorithm, a multimaterial linearization algorithm, a dual-energy algorithm and an iterative reconstruction algorithm. Each beam hardening correction algorithm is applied to simulated data sets of a dual-material phantom consisting of multiple rods. The comparison is performed on data sets simulated both under ideal conditions and with the addition of quantum noise. The performance of each algorithm is assessed with respect to its effect on the final image quality (contrast-to-noise ratio, spatial resolution), artefact reduction (streaks, cupping effects) and dimensional measurement deviations. The metrics have been carefully designed in order to achieve a robust and quantifiable assessment. The results suggest that the single-material linearization algorithm can reduce beam hardening artefacts in the vicinity of one material. The multimaterial linearization algorithm can further reduce beam hardening artefacts induced by the other material and improve the dimensional measurement accuracy. The dual-energy method can eliminate beam hardening artefacts, and improve the low contrast visibility and dimensional measurement accuracy. The iterative algorithm is able to eliminate beam hardening streaks. However, it induces aliasing patterns around the object edge, and its performance depends critically upon computational power. The contrast-to-noise ratio and spatial resolution are declined by noise. Noise also increases the difficulty of image segmentation and quantitative analysis. LAY DESCRIPTION: X-ray computed tomography (CT) is a major breakthrough in digital imaging technology in the late 20th century. First used as an important tool in medical imaging, CT has gradually introduced to the nonmedical areas (e.g. industrial nondestructive testing). Inherently CT is more prone to artefacts comparing to the conventional real-time X-ray image. Beam hardening artefacts caused by the polychromatic nature of X-ray spectra are known to deteriorate the reconstructed image quality in industrial CT. A number of beam hardening correction algorithms exist and are used across medical CT. However, there is a lack of research on their effectiveness on industrial CT. This study presents an in-depth beam hardening correction algorithm comparison in industrial CT. Since this study takes various factors of the algorithm performance into account, it provides insights of the advantages and disadvantages of each algorithm and assists the choice of algorithm to meet specific needs of industry. Existing beam hardening correction algorithms are divided into the following four categories: linearization, segmentation based linearization, dual-energy and iterative methods. Since the linearization method can only correct single-material objects, we did not include it in the comparative study. Among the remaining categories, we chose one from each category for comparison, for methods in one peer category share similar physical and mathematical principles. The methods are polynomial fit, Joseph segmentation, dual energy and IMPACT iterative method. This study uses a simulated polychromatic data set of a multimaterial phantom. The central slice of the corrected reconstructions is then assessed and the results are presented. In this study, we will compare beam hardening correction methods with respect to their performance on image quality, the removal of image artefacts and the influence on dimensional accuracy.