Radiation therapy is often offered as the primary treatment for head and neck cancer(HNC). Accurate target delineation is essential for the success of radiation therapy. The current target definition technique - manual delineation using Computed Tomography(CT) - is subject to high observer variability. Functional imaging modalities such as 2-[18F]-fluoro-2-deoxy-D-glucose Positron Emission Tomography(FDG-PET) can greatly improve the visualization of tumor. FDG-PET co-registered with CT has shown potential to improve the accuracy of target localization and reduce observer variability. Unfortunately, due to the limitation of PET, the degree of improvement obtained by qualitative and simple quantitative (e.g. thresholding) use of FDG-PET is not ideal. However, both PET and CT images contain a wealth of texture information that could be used to improve the accuracy of target definition.
This thesis has investigated using texture analysis techniques to automatically delineate radiation targets. Firstly, PET and CT texture features with high discrimination ability were identified and a texture analysis technique- a decision tree based K Nearest Neighbour(DTKNN) classifier – was developed. DTKNN could accurately classify head and neck tissue with an area under curve(AUC) of a Receiver Operator Characteristic(ROC) of 0.95. Subsequently, an automated target delineation technique - CO-registered Multi-modality Pattern Analysis Segmentation System(COMPASS) - was developed that can delineate tumor on a voxel-by-voxel basis. COMPASS was found to accurately delineate HNC with 84% sensitivity and 95% specificity on a voxel basis per patient. To accurately evaluate the utility of the COMPASS in radiation targeting, a validation method which can combine biased observers' contours to generate a probabilistic reference for validation was developed. The method was based on maximum likelihood analysis using a simulated annealing(SA) algorithm.
The results from this thesis show that texture features of both PET and CT images can enhance the discrimination between HNC and normal tissue, and an automated delineation method of HNC using texture analysis of PET and CT images can accurately and consistently define radiation targets in head and neck. This suggests that automated segmentation of radiation targets based on texture analysis techniques may significantly reduce observer variability and improve the accuracy of radiation targeting.
| Identifer | oai:union.ndltd.org:LACETR/oai:collectionscanada.gc.ca:OTU.1807/24920 |
| Date | 02 September 2010 |
| Creators | Yu, Huan |
| Contributors | Caldwell, Curtis |
| Source Sets | Library and Archives Canada ETDs Repository / Centre d'archives des thèses électroniques de Bibliothèque et Archives Canada |
| Language | en_ca |
| Detected Language | English |
| Type | Thesis |
Page generated in 0.0023 seconds