Abstract

Info

Title:: A random walk method with adaptive regression-kernel for delineation of biological target volumes of tumors

Author(s):: LIU Guocai1; GUAN Wenjing1; TIAN Juanxiu1; ZHU Suyu2; JU Zhongjian3; 1. College of Electrical and Information Engineering, Hunan University, Changsha 410082, China; 2. Department of Radiotherapy, Hunan Cancer Hospital, Xiangya School of Medicine, Central South University, Changsha 410013, China; 3. Department of Radiotherapy, Chinese PLA General Hospital, Beijing 100853, China

Keywords:: ; Keywords: medical image segmentation; adaptive kernel regression; random walk; target volume delineation; intensity-modulated radiotherapy; nasopharyngeal carcinoma

Abstract:: Abstract: Objective To access a novel method for the delineation of biological target volumes (BTV) of tumors in positron emission computed tomography (PET) images. Methods In order to effectively discriminate between normal tissues and tumors with similar standard uptake value (SUV) in PET images, the adaptive regression kernels of every voxel in PET images of tumors were calculated, and the adaptive regression kernels were integrated into the weight function of edges of undirected graph for random walk (RW). Then the seeds of RW were automatically selected by three-dimensional adaptive region growing method to realize the automatic delineation of BTV in PET images. Results The PET images of 7 patients with nasopharyngeal carcinoma were used to evaluate the performances of the proposed method. The gross target volumes delineated manually by radiation oncologists were taken as a surrogate of the ground truth. The mean value of DICE similarities of BTV delineated by the proposed method in 7 cases of nasopharyngeal carcinoma was 0.837 6, which was increased by 4.31% as compared with RW only based on PET SUV and increased by 3.34% as compared with RW based on PET SUV and contrast. Conclusion The proposed RW algorithm with adaptive regression-kernel can delineate the BTV of tumors in PET images more accurately.