Modelling geometric uncertainties in prostate radiotherapy

Image guided radiotherapy (IGRT) presents a range of techniques to enhance thegeometric accuracy of radiotherapy treatment. A full evaluation of such techniques,against increased cost, workload, or unnecessary radiation dose, would require theincorporation of geometric errors into existing methods for modelling theeffectiveness of radiotherapy treatment. In this work the impact of geometric errors onthe tumour control probability (TCP) for prostate radiotherapy is investigated.Positional measurement data are analysed in a manner which differentiates betweengenuine measured variation and measurement uncertainty. A deformable model of theprostate is developed which encompasses population variation, interfractionalvariation and observer variability. Ultimately a simulation tool is produced to testdifferent schemes for prostate radiotherapy in the presence of different levels ofgeometric uncertainty.Measurement error is not generally considered in studies of geometric variation inradiotherapy. Maximum likelihood estimation (MLE) is used to separately determinethe measurement error and the random patient position error in a series of externalposition measurement for 112 prostate patients. Measurement error can be asignificant component of the measured variation, with serious implications for IGRTwhere measurements are used to correct the patient position.Measurements are made of internal and external errors for 44 patients, comparingtechniques for rectal and bladder preparation. Internal marker seeds are used todetermine prostate position from portal images. Seed based portal imaging is found tobe an effective method for IGRT for patients in all arms of the trial.Two techniques for modelling prostate shape are investigated. To analyse patterns ofobserver variability a method is developed based on the inclusion of image pixels, orvoxels, in the shape volume, using principle component analysis (PCA) of binarymasks. Inclusion probability is found to be a useful concept but the results of the PCAare poor, demonstrating a high level of dependence between modes. A moreconventional approach, the PCA of boundary vectors, is used to generate a 3Ddeformable vector model of the prostate from correlated 2D orthogonal views. The model was trained from data for 28 patients with 2 CT scans and 2 independentobservations of prostate shape per scan. MLE was used to separate the components ofshape variation. Of the total variance observed 35% was due to population variation,17% due to day to day variation and 48% due to observer variation.The final simulation tool allows dose to individual tissue elements to be accumulatedover a simulated course of treatment for a treatment population. The tool is used tocalculate TCP for a wide range of situations; investigating the impact of observererror, target motion, deformation, margin size, IGRT technique, dose prescription andthe use of a high dose boost. The reduction in calculated TCP with margin size isshown to be continuous, with no clear cut-off point. The reduction in TCP with'inadequate' margins can be modest and easily offset by small increments in dose.This clearly validates the recent trend towards small volume, high dose techniques,even where dose coverage might be compromised. In order to truly determine theoptimum technique the simulation would need to be expanded to incorporate the riskof damage to the rectum. Even in its current form, the simulation tool still provides aunique insight into the relationship between the dose-volume prescription, the IGRTtechnique and the population TCP; allowing an objective assessment of the relativebenefits of different approaches to prostate radiotherapy.

Level dc:type.qualificationlevel

Doctoral (Level 8)

Year dc:date.issued

2011

Author dc:creator

• Sage, J

Language dc:language

en

Identifier

oai:salford-repository.worktribe.com:1338116

OAI identifier oai:identifier

oai:salford-repository.worktribe.com:1338116