Modelling the impact of treatment uncertainties in radiotherapy

Abstract

Uncertainties are inevitably part of the radiotherapy process. Uncertainty in the dose deposited in the tumour exists due to organ motion, patient positioning errors, fluctuations in machine output, delineation of regions of interest, the modality of imaging used, and treatment planning algorithm assumptions among others; there is uncertainty in the dose required to eradicate a tumour due to interpatient variations in patient-specific variables such as their sensitivity to radiation; and there is uncertainty in the dose-volume restraints that limit dose to normal tissue.

This thesis involves three major streams of research including investigation of the actual dose delivered to target and normal tissue, the effect of dose uncertainty on radiobiological indices, and techniques to display the dose uncertainty in a treatment planning system. All of the analyses are performed with the dose distribution from a four-field box treatment using 6 MV photons. The treatment fields include uniform margins between the clinical target volume and planning target volume of 0.5 cm, 1.0 cm, and 1.5 cm. The major work is preceded by a thorough literature review on the size of setup and organ motion errors for various organs and setup techniques used in radiotherapy.

A Monte Carlo (MC) code was written to simulate both the treatment planning and delivery phases of the radiotherapy treatment. Using MC, the mean and the variation in treatment dose are calculated for both an individual patient and across a population of patients. In particular, the possible discrepancy in tumour position located from a single CT scan and the magnitude of reduction in dose variation following multiple CT scans is investigated. A novel convolution kernel to include multiple pretreatment CT scans in the calculation of mean treatment dose is derived. Variations in dose deposited to prostate and rectal wall are assessed for each of the margins and for various magnitudes of systematic and random error, and penumbra gradients.

The linear quadratic model is used to calculate prostate Tumour Control Probability (TCP) incorporating an actual (modelled) delivered prostate dose. The Kallman s-model is used to calculate the normal tissue complication probability (NTCP), incorporating actual (modelled) fraction dose in the deforming rectal wall. The impact of each treatment uncertainty on the variation in the radiobiological index is calculated for the margin sizes.