Intersection Line Length Normalization in CT Projection Data
We present a method to improve the quality of common sinogram restoration algorithms, which are used for metal-artifact reduction in X-ray CT. The presented approach is based on the idea that the intersection length of all beams through the object of inte
- PDF / 357,841 Bytes
- 5 Pages / 439.37 x 666.142 pts Page_size
- 86 Downloads / 277 Views
bstract. We present a method to improve the quality of common sinogram restoration algorithms, which are used for metal-artifact reduction in X-ray CT. The presented approach is based on the idea that the intersection length of all beams through the object of interest can approximately be calculated from a preliminary reconstruction of the object. Incorporating this information to the reconstruction allows significant reduction of streak artifacts. Additionally, it is shown that the information about intersection length may also be used to improve the detection of metal shadows within the sinogram.
1
Introduction
As shown on the left side in Fig. 1, metal objects inside the field of view cause artifacts in the reconstructed CT image. The predominant effect for these artifacts arises from inconsistent sinogram entries, which are a direct consequence of non-linear beam-hardening. One approach to correct for these erroneous values and, therefore, to reduce metal artifacts is to replace all corrupted sinogram entries with artificially generated values. If these surrogate values are chosen for instance by incorporating a priori knowledge, the quality of the reconstructed image is improved. There are several ways to calculate these artificial values. Two methods, which are related to each other are presented in [1] and [2]. A chain of clustering, segmentation and forward/backward projection is used to reconstruct a preliminary image, which represents an approximate distribution of air, soft tissue, and bones. A forward projection of this image is then used to fill the gaps in the original sinogram. A second approach is presented in [3] and [4]. Here, corrupted data is replaced with interpolated values from adjacent detectors. This interpolation is usually done independently for each view. In [5] a method is proposed, which uses data from more than one view to generate new values by means of an directional interpolation scheme. In the present paper a method is proposed that can generally be incorporated into any interpolation based method. It is shown that a normalization of the sinogram improves both the reduction of metal artifacts and the detection of corrupted data compared to interpolation methods based on the original projection data.
78
J. M¨ uller and T.M. Buzug
Fig. 1. The uncorrected reconstruction f (left) and the image support supp(f ) (right), which is generated from an image which was corrected with a common metal artifact reduction method (Fig. 4)
2
Methods
A problem which complicates the interpolation (and also the actual detection of corrupted sinogram entries) is that the value of a particular sinogram entry does not only depend on the material of the object but also on the intersection line length of the corresponding beam through the object: For example, a beam attenuated by a metal object and having a short intersection line with the body may have a smaller attenuation value than a beam, which is only attenuated by soft tissue and bone, but having a long intersection line through the body. To reduce
Data Loading...
