3D RANDOM WALK BASED SEGMENTATION FOR LUNG TUMOR DELINEATION IN PET IMAGING

This article presents a segmentation approach based on random walk (RW) method to delineate tumors having inhomogeneous activity distributions of (18)FDG on Positron Emission Tomography (PET) images. Based on the original algorithm of RW [1], we propose an improved approach using an adaptive parameter instead of a fixed one and integrating probability densities of label into the system of linear equations used in the RW. The proposed segmentation method initializes automatically seeds in tumor voxels using Fuzzy-C Means (FCM), and then delineates the tumor volume using the improved RW. The performances of the algorithm were assessed on PET images of a physical phantom, covering a range of hot spheres simulating tumor lesions of volume [0.99 - 97.3 mL] and contrast [2 - 7.7] for a voxel size of 4.1x4.1x2.0 mm(3). A comparison study with a fixed threshold value of 40 % [2] and an adaptative thresholding algorithm [3] have been carried out. Results show that the proposed method is more effective than the other methodologies for spherical volume measurement. The good performances of the improved RW method have been also confirmed on data of three patients having heterogeneous (18)FDG uptakes.