Previous Normal Dose Scan Induced Nonlocal Means Gibbs Prior for Low Dose Computed Tomography Maximum A Posteriori Reconstruction

摘要

Lower X-ray radiation to patient in computed tomography (CT) examination is clinically desired. Although reducing the level of milliampere-seconds (mAs) can yield lowdose CT scan, the associated image quality will be strongly degraded by excessive quantum noise. In this paper, for obtaining high-quality image in low-mAs or low-dose CT scan, we propose a previous normal dose scan induced nonlocal means Gibbs prior, named as ndiNLMGP. The ndiNLMGP was designed based on the normal-dose scan induced nonlocal means filter (ndiNLM) by exploiting the huge redundancy information from the high-quality reference image. And the associated optimization algorithm was developed based on maximum a posteriori (MAP) criterion for image reconstruction. To validate and evaluate the performance of the present ndiNLMGP, we carried out the low-dose image reconstruction from the computer simulated and real data. The results show that the present ndiNLMGP can impressively outperform the simplex nonlocal means Gibbs prior (NLMGP) model in lowering the noise, and preserving the edge and details in the desired images.