PET Image Reconstruction Using Nonlocal Means Regularization Based on Structural Similarity

Abstract

In this paper, we propose an improved algorithm for PET image reconstruction using nonlocal means regularization based on structural similarity which is developed from the previously proposed algorithm LMROS-NLM and is named LMROS-NLM-SSIM. The structural similarity index (SSIM) is used to describe the similarity between the neighborhoods of two pixels of the image in nonlocal means (NLM) regularization rather than the Gaussian weighted Euclidean distance as the latter weights calculation method may lead to wrong value in noisy PET images. The performance of the proposed algorithm was compared with LMROS-NLM through reconstruct 50% of the data of human subject collected with an All-digital Brain PET system. Visual inspection reveals that the proposed algorithm can yield images with enhanced spatial resolution and suppressed image noise.