The derivation of margins for use in external beam radiotherapy involves a complex balance between ensuring adequate tumour dose coverage that will lead to cure of the cancer whilst sufficiently sparing the surrounding organs at risk (OARs). The treatment of cancer using ionising radiation is currently witnessing unprecedented levels of new treatment techniques and equipment being introduced. These new treatment strategies, with improved imaging during treatment, are aimed at improved radiation dose conformity to dynamic targets and better sparing of the healthy tissues. However, with the adoption of these new techniques for radiotherapy, the validity of the continued use of recommended statistical model based margin formulations to calculate the treatment margins is now being questioned more than ever before. To derive margins for use in treatment planning which address present shortcomings, this study utilised novel applications of fuzzy logic and neural network techniques to the PTV margin problem. As an extension of this work a new hybrid fuzzy network technique was also adopted for use in margin derivation, a novel application of this technique which required new rule formulations and rule base manipulations. The new margin models developed in this study utilised a novel combination of the radiotherapy errors and their radiobiological effects which was previously difficult to establish using mathematical methods. This was achieved using fuzzy rules and neural network input layers. An advantage of the neural network procedure was that fewer computational steps were needed to calculate the final result whereas the fuzzy based techniques required a significant number of iterative computational steps including the definition of the fuzzy rules and membership functions prior to computation of the final result. An advantage of the fuzzy techniques was their ability to use fewer data points to deduce the relationship between the output and input parameters. In contrast the neural network model requires a large amount of training data. The previously stated limitations of currently recommended statistical techniques were addressed by application of the fuzzy and neural network models. A major advantage of the computational intelligence methods from this study is that they allow the calculation of patient-specific margins. Radiotherapy planning currently relies on the use of ‘one size fits all’ class solutions for margins for each tumour site and with the large variability in patient physiology these margins may not be suitable for use in some cases. The models from this study can be applied to other treatment sites, including brain, lung and gastric tumours.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:536122 |
Date | January 2011 |
Creators | Mzenda, Bongile |
Contributors | Brown, David ; Liu, Honghai ; Gegov, Alexander Emilov |
Publisher | University of Portsmouth |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | https://researchportal.port.ac.uk/portal/en/theses/computational-intelligence-margin-models-for-radiotherapeutic-cancer-treatment(c91a87fd-a7af-416d-9336-ada908feda33).html |
Page generated in 0.0011 seconds