Second order total generalized variation for low-dose computed tomography image reconstruction

High radiation dose during x-ray computed tomography (CT) examinations can increase the risk of cancer and has become major concerns to patient. Accordingly, minimizing the radiation exposure without sacrificing image quality is a meaningful research topic. In this work, with the aim to reduce radiation during data acquisition, we propose a penalized weighted least-squares (PWLS) scheme to retain the image quality by incorporating a total generalized variation (TGV) regularization, which is referred to as “PWLS-TGV”. Specifically, the TGV regularization utilizes second-order derivatives of the desired image with imposing some higher order smoothness in regions away from the edges and the weighted least-squares term considers a data-dependent variance estimation serving for improvement of image reconstruction from low-dose CT measurement. Subsequently, an alternating minimization algorithm was adopted to optimize the associative objective function. The experimental results on digital phantom and real patient data show that the present PWLS-TGV method can achieve significant gains over the existing similar methods in noise and artifacts suppression.