Texture-preserved low-dose CT reconstruction using region recognizable patch-priors from previous normal-dose CT images

Abstract

Low-dose computed tomography (LdCT) has been widely used in clinical applications including image-guided biopsy needle and lung screening. With the low radiation dose in data acquiring during the scans, the image quality will degrade due to the excessive quantum noise if there is no adequate noise treatment in processing the data for image reconstruction. For conventional low-dose CT (LdCT) reconstruction, much effort has been spend on the preservation of edges while removing the noise, and not enough attention is paid to texture preserving. However, the image textures are usually recognized as imaging biomarker and play an important role for the tasks, such as lung cancer and nodule detections. In this paper, a texture-preserving method is proposed in LdCT reconstruction. The proposed method consists of the following steps: (1) decompose the normal-dose CT (NdCT) image from FBP reconstruction into overlapping patches. Then, classify these patches into different clusters by Gaussian mixture model (GMM); (2) extract the texture-preserving priors for each cluster from the NdCT image patches with the classified patches into different clusters; (3) decompose the LdCT image from FBP reconstruction into overlapping patches, and then classify these patches by the same GMM in (1); (4) apply the priors in LdCT reconstruction under penalized weighted least-squares criteria. The experimental results show that the proposed method can achieve promising results for LdCT reconstruction in terms of edge and texture preserving.