PET Based Textural Features And Intelligent Delineations Of Biological Target Volumes Corresponding To Head And Neck Tumors

The positron emission tomography (PET) functional molecular biological imaging system has been widely used to delineation of the biological target volume (BTV) and biological intensity-modulated radiotherapy (BIMRT) of tumor. Highly precise delineation of the BTV is the key of the radiation treatment plan of BIMRT, which can be done by PET image segmentation. However, a large amount of clinical cases have demonstrated that it is very difficult and need the intelligent segmentation methods. In fact, the segmentation technique based on the threshold of FDG standardized uptake value (SUV) cannot effectively delineate the tumor biological target. The reason is that high SUV values may be present in normal tissues, such as tonsil, salivary glands and muscles. High SUV values can also be existed in the areas of post-operative inflammatory reaction and radiation reaction. Moreover, the SUVs also depend on the resolution and partial volume effect of a PET scaner, and biological characteristics of malignant tumors. On the other hand, threshold based segmentation techniques use only the grayscale information of PET images and not the spatial information, so it may result in "empty region" in the segmented BTV due to strong noise in PET images.To overcome the limits of the segmentation technique based on the threshold of PET SUV, we proposed an adaptive method for the delineation of the biological target volume via an improved two-stage adaptive three-dimensional volume-growing method based on the PET SUV and its texture features of head and neck cancer. First, we exacted three-dimensional molecular biological textures based on the gray-level co-occurrence matrix (GLCM) of the PET SUV image of tumors. Then, the exacted biological texture, namely variance(VAR) and the SUV feature were simultaneously used to automatically delineate the target volume via an improved two-stage adaptive region-growing algorithm.The results demonstrated that the proposed method can more accurately and precisely delineate a BTV than our previous two-stage adaptive region-growing algorithm, which is due to the combined VAR and SUV features of tumors.