Full Dose CT Database Induced Reconstruction With Nonlocal Means Prior for Ultra-Low-Dose Lung CT: A Preliminary Study

Abstract

Although Low-dose computed tomography (LDCT) is the most effective way for early lung cancer screening, it’s still a challenge to further reduce radiation dose on the premise of ensuring image quality. Penalized weighted least-squares (PWLS) image reconstruction with nonlocal means (NLM) prior has shown excellent performance to improve the image quality for LDCT, especially when the nonlocal weights are calculated from previous full-dose CT (FDCT) image. However, the previous FDCT image of the same patient is not readily available, and registration between the LDCT and FDCT images must be considered because of the scanning misalignment. This paper proposed a new NLM prior model to reconstruct high quality LDCT image without image registering. In order to estimate the nonlocal weights of NLM prior, a database was trained from FDCT images of different patients, from which the patch samples similar to each target patch of the LDCT were extracted. Then the nonlocal weights were determined by the patch samples, and integrated into PWLS reconstruction with the priori information of local structures from FDCT. Experiments with 10mAs LDCT data have shown its superiority in reducing noise, streaking artifacts and preserving structure detail, indicating the potential of further dose reduction in ultra-LDCT lung screening.