Parametric Model for Compensation of Partial Volume Effect in CT Iterative Reconstruction

Iterative reconstruction methods relying on linearization of the relationship between images and projections have to cope with the inherent representation of the scanned object by voxel elements. When the voxel size becomes large enough to reduce the overall computational requirements, partial volume can become an issue for overall image quality. We propose in this paper a linear parametric model with the purpose of providing a better representation of the object in regions of rapid variations in local density. We show that this approach is effective at retaining the quality of thin slice reconstructions without explicitly modeling thinner slices. Good image quality is obtained at a significantly reduced computation cost relative to that of using finer sampling of the image volume.