Automatic Delineation Of Tumor Target Volumes For Radiotherapy Based On PET And MRI Images

At present,malignant tumor has become one of the leading causes of death of human.The delineation of malignant tumor target volumes is a key for radiotherapy planning.With the development of medical imaging technology,computer tomography,(CT),positron emission tomography,(PET),and magnetic resonance imaging(MRI),have been used to tumor diagnosis,staging,radiotherapy planning and curative effect evaluation.PET is for functional imaging such as tissue metabolism,and MRI is for anatomy imaging with high contrast.The delineation of the tumor target volumes in PET or MRI is PET or MRI image segmentation.Based on the characteristics of tumor PET images and MRI images,two new methods for the delineation of tumor target volume were proposed in this paper,the specific methods are as follows:(1)The delineation of tumor PET target volumes based on random walk integrated with adaptive regression kernel: There are complex tissue structures in head and neck.The standard uptake value(SUV)of PET FDG in tumor is similar to the SUV in brain stem and other metabolic active brain tissue,which makes it difficult to distinguish the tumor volumes and the surrounding normal tissue volumes with similar PET SUV via classic random walk.We found that there are considerable differences between the3 D adaptive regression kernels of voxels in normal tissues and in tumors,so we proposed to integrate the vector of 3D adaptive regression kernel of voxels into the weighting function of the random walk.We selected seeds for the random walk via adaptive region growing,and the optimal adaptive regression kernel by changing the size of the kernel analysis window in the adaptive regression kernel.The proposed random walk algorithm can delineate the target volume of tumors in PET images more accurately.The resulting the mean value of the seven DICE similarities is 0.8376.Comparing the proposed method with the RW only based on PET SUV,and the RW based on PET SUV and PET contrast texture,the resulting mean DICE similarity increase by 4.40% and 3.18% respectively.(2)The delineation of tumor MRI target volumes by level set methodcombined with PET: PET can provide molecular biological function information such as tumor metabolism,proliferation,and lack of oxygen,MRI can provide anatomical information.Generally,the gray value of voxels in the tumor volumes in T1-weighted MRI images is higher than one in the neighborhood normal tissue volumes.We proposed a level set method to delineate the tumor MRI target volumes of head and neck neoplasms with the complementary information of PET images and MRI images.Firstly,the evolution region of level set method is determined by three-dimensional adaptive region growing and morphological dilation on PET SUV images.And then,according to the characteristics of MRI images,we constructed the energy function by jointing the region information with edge information of MRI images.The threshold parameter in the region term of the energy function is set as the 50% of the maximum gray value of the MRI in the evolution region.The proposed method improved the precision of the delineation of target in tumor MRI.