Variance stabilizing transformations in the reduction of poisson noise in 3D nuclear medicine images

Nuclear medicine (NM) is a modality of medical imaging that uses radioactive materials to provide information about the functioning of a person's specific organs. However, the NM images are characterized by presenting poor signal to noise ratio, related to low counts and to Poisson noise because of the stochastic nature of the attenuation processes (absorption and scattering). These kinds of images depict regions with high counting photons that are created by a group of cells that have a faster metabolism indicating physiological changes. This phenomenon could suggest the presence of a malignant neoplasia (tumor). Hence, before applying any technique for determining metabolic rates, first it is necessary to define the region of interest (ROI). For this purpose, segmentation techniques were used based on the Fuzzy Connectedness theory. It is possible to improve the efficiency of the segmentation stage by using appropriate filters for the treatment of the Poisson noise. This study used anthropomorphic phantoms of the left ventricle of the heart. A NM exam was simulated using GATE platform with the phantom coupled to the PET scanner recreated. Then, the projections obtained were reconstructed using the STIR platform. Several studies have shown the effectiveness of the Anscombe Transformation in the stabilization of the variance of NM images. However, in this paper other approaches like Bartlett and Freeman & Tukey were used and compared to the original method of transformation. Wiener filter was used in combination with all the types of transformations tested. Freeman & Tukey / Wiener filter improve the efficiency of the segmentation stage compared to the other two filters. In the near future, this procedure will be applied in real 3D images.