Comparison of Non-Linear Filtering Methods for Positron Emission Tomography

Abstract

As a result of the limited radiotracer dose, acquisition time and scanner sensitivity, positron emission tomography (PET) images suffer from high noise. In the current clinical practice, post-reconstruction filtering has become one of the most common noise reduction techniques. However, the range of existing filters is very wide, and choosing the most suitable filter for a given measurement is far from simple. This paper aims to provide assistance in this choice by comparing the most powerful image denoising filters, covering both image quality and execution time. Emphasis is placed on non-linear techniques due to their ability to preserve edges and fine details more accurately than linear filters. The compared methods include the Gaussian, the bilateral, the guided, the anisotropic diffusion and the non-local means filters, which are examined in both static and dynamic PET reconstructions.