Molecular biological characteristics based Hierarchical Mumford-Shah Vector-Model for the delineation of biological target volumes corresponding to head and neck tumors

To more accurately and precisely delineate a biologic target volume (BTV) for the positron emission tomography (PET)-guided radiotherapy treatment planning, we proposed a novel Hierarchical Mumford-Shah Vector Model (HMSMv), where the image-vector was composed of the molecular biological characteristics such as contrast, busyness and the standard uptake value (SUV) of tumors. The BTV was delineated via the propagation of the level set flow of the proposed HMSMv. The propagation took place inside a ring-volume of interest (VOI) which was defined by an adaptive volume-growing algorithm based on the PET SUV, contrast and busyness of a tumor. Four patient studies were assessed and visually inspected by two radiation oncologists (Suyu Zhu and Zaijie Huang). Compared the resulting BTV with the gross target volume (GTV) of one patient study, the sensitivity, specificity and similarity were 92.28%, 87.29%, 79.28% respectively. Moreover, all of four BTVs were between the corresponding GTV and planning target volume (PTV). Most of the BTVs were between the GTV and the clinical target volume (CTV). The results demonstrated that the proposed method can delineate the BTVs of nasopharyngeal carcinomas more accurately and precisely than the manual delineation and also the threshold segmentation method with SUV of 2.5 as the threshold. The GTV, CTV, and PTV were manually delineated by the two radiation oncologists based on PET, CT and MRI images for the intensity modulated radiotherapy (IMRT) planning of the four patients.