A Comparison of FDG PET-CT Tumor Segmentation for Clinical Application

Abstract

Automatic segmentation of tumor in FDG PET-CT is potentially beneficial in clinical application. However, the performances of existing methods are different. This study compares several algorithms of FDG PET-CT tumor segmentation methods that were recently proposed and which can be used for clinical radiation therapy. In our work, we have researched some methods including the gradient-based (GRAD) method, the constant threshold-based (THRESH) method, and the region growing (RG) method. For each method the same work flow is used and the tumor segmentation results are compared to the manual contouring (MC) by experienced physician which is widely used in clinical radiation therapy. The results show that the region growing is the most accurate to the MC and has the potential to play the most important role in radiation therapy.