CT image noise reduction using rotational-invariant feature in Stockwell transform

Iterative reconstruction and other noise reduction methods have been employed in CT to improve image quality and to reduce radiation dose. The non-local means (NLM) filter emerges as a popular choice for image-based noise reduction in CT. However, the original NLM method cannot incorporate similar structures if they are in a rotational format, resulting in ineffective denoising in some locations of the image and non-uniform noise reduction across the image. We have developed a novel rotational-invariant image texture feature derived from the multiresolutional Stockwell-transform (ST), and applied it to CT image noise reduction so that similar structures can be identified and fully utilized even when they are in different orientations. We performed a computer simulation study in CT to demonstrate better efficiency in terms of utilizing redundant information in the image and more uniform noise reduction achieved by ST than by NLM.