Computed Tomography image denoising utilizing an efficient sparse coding algorithm

In this paper, the problem of reducing noise from low-dose Computed Tomography (CT) is investigated. The process is composed of: sparse coding, dictionary update and denoising; that is a time consuming process. Hence, despite the promising results reported in literature, it has not attracted much attention in medical applications. In an attempt to reduce the complexity and time consumed, we propose an efficient method for sparse coding approximation. In the proposed sparse coding approach, unlike most current methods the global search is performed only once. The potential representative atoms are identified and buffered, then only a local recursive pursuit within a few atoms is executed to find the sparse representation. Moreover, the K-SVD dictionary update method and its extension to image denoising is utilized for reducing the noise in CT scans. Our results demonstrate this approach is reliable and improves the accuracy and process time significantly, making the proposed method a suitable candidate for clinical purposes.