Statistical Sinogram Smoothing for Low-Dose CT With Segmentation-Based Adaptive Filtering

As is known, noises in calibrated and log-transformed projection data of low-mA (or low-dose) CT protocol follow approximately a non-stationary Gaussian distribution. In this study, we further demonstrate that some isolated noise points would not satisfy the above observation. Hence, we propose a noise reduction scheme which includes isolated data removal and segmentation-based filtering. In this scheme, an isolated data removal algorithm is first adopted to remove isolated data such that the remaining sinogram data approximately follows a Gaussian distribution. Secondly, image segmentation technique is employed for segmenting sinogram image into different segments in which pixels with similar intensities are grouped, and the segmentation-based adaptive statistical sinogram smoothing technique is proposed with different smoothness parameters applied to different segments for adaptively filtering. The effectiveness of the proposed method is validated by both computer simulations and experimental studies. The gain of the proposed approach over other methods is quantified by noise-resolution tradeoff curves.