22 March 2016 Texture-preserved penalized weighted least-squares reconstruction of low-dose CT image via image segmentation and high-order MRF modeling
Author Affiliations +
Abstract
In this paper, we proposed a low-dose computed tomography (LdCT) image reconstruction method with the help of prior knowledge learning from previous high-quality or normal-dose CT (NdCT) scans. The well-established statistical penalized weighted least squares (PWLS) algorithm was adopted for image reconstruction, where the penalty term was formulated by a texture-based Gaussian Markov random field (gMRF) model. The NdCT scan was firstly segmented into different tissue types by a feature vector quantization (FVQ) approach. Then for each tissue type, a set of tissue-specific coefficients for the gMRF penalty was statistically learnt from the NdCT image via multiple-linear regression analysis. We also proposed a scheme to adaptively select the order of gMRF model for coefficients prediction. The tissue-specific gMRF patterns learnt from the NdCT image were finally used to form an adaptive MRF penalty for the PWLS reconstruction of LdCT image. The proposed texture-adaptive PWLS image reconstruction algorithm was shown to be more effective to preserve image textures than the conventional PWLS image reconstruction algorithm, and we further demonstrated the gain of high-order MRF modeling for texture-preserved LdCT PWLS image reconstruction.
© (2016) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Hao Han, Hao Zhang, Xinzhou Wei, William Moore, Zhengrong Liang, "Texture-preserved penalized weighted least-squares reconstruction of low-dose CT image via image segmentation and high-order MRF modeling", Proc. SPIE 9783, Medical Imaging 2016: Physics of Medical Imaging, 97834F (22 March 2016); doi: 10.1117/12.2216428; https://doi.org/10.1117/12.2216428
PROCEEDINGS
6 PAGES


SHARE
Back to Top