Linearized Bregman Iterations for Low-dose CT Reconstruction

A new low dose CT reconstruction model is proposed under the condition of low signal-to-noise ratio measured data, which are caused by reducing the X-ray source tube current in order to avoid the excessive radiation dose. In the objective function of the model, the logarithm likelihood function under Poisson noise is used as the fidelity functional, and sparse prior of image transform domain coefficients is used as the regularization functional. The fidelity functional is more effective than the additive Gaussian noise model, while the regularization the functional can overcome the ill posed problem of image reconstruction expecially in the low-dose situation. By using the linearized Bregman iteration, the sum minimization scheme is split into one step of quadratic programming with variable coefficient and the other step of the denoising issue. It can accelerate the convergence speed through the variable coefficient calculation in the quadratic programming to approximate the original fidelity term. Experimental results show that this proposed approach can be successfully applied to low-dose fan-beam CT reconstruction and it outperforms some existing algorithms including filter back projection algorithm, maximum likelihood algorithm and classical weighted l2 norm reconstruction algorithm.