Abstract
Cancer is the number two killer after cardiovascular diseases according to the World Health Organisation. It is generally accepted that about 50-55 % of all cancer patients benefit from radiotherapy treatment in which high-energy photons from medical linear accelerators (linacs) are commonly used. It is the goal of 3D conformal radiotherapy and intensity modulated radiotherapy (IMRT) to maximise the radiation dose to the tumour site while minimising the dose to the surrounding normal tissues. Thus the radiation beam is shaped to conform to the tumour outline by the multileaf collimator (MLC). Fast and accurate dose calculation is essential to the success of the treatment. The current method of choice is the superposition/convolution method for its computation efficiency but the complexity of the algorithm grows as the treatment moves into complicated regimes. The Monte Carlo method, on the other hand, uses one algorithm for different treatment regimes and its accuracy has been well proven. The drawback of the Monte Carlo method is in its computationally intensive and time-consuming nature. In a Monte Carlo simulation of a linac, it is common practice to divide the process into steps so that duplicate simulation of the patient-independent components can be avoided. Furthermore, the data of all particles emerging from any linac component form a phase space. A summary of these data allows, in principle, the generation of unlimited number of particles for simulations downstream. This summary is known as phase space model. This thesis examines different phase space models generated from the simulation of the patient-independent components. Under investigation is the 6 MV beam from the Elekta SLi linac. Two well-known phase space models, the point source model (PSM) and the multiple source model (MSM), were successfully implemented with MCNPX version 2.4.0. A new model termed the directional spectrum model (DSM) was proposed. In contrast to the PSM and the MSM which loosely relate the particle energy to its direction, the DSM couples the energy spectrum directly to the flight direction so that the scattering properties in the linac head are well accounted for. The DSM calculated dose distributions compare favourably with measurements in water phantom. It performs well inside and outside the 5x5, 10x10 and 20x20 cm2 fields. The confidence limits are generally within the American Association of Physicists in Medicine recommended tolerance of 2 % on central axis (CAX) beyond the depth of maximum dose (dmax) and 3 % in other low dose gradient regions. The shifts in the high dose gradient regions are also within the recommended tolerance of 2 mm. These shifts were measured in the dose build-up regions before dmax and in the isodose curves in a diamond-shaped field. The DSM also performs satisfactorily in the dose profiles formed by a single leaf of the MLC in a large field. After convolution with a Gaussian kernel, near perfect matches were obtained between the DSM calculated profiles and the RK, chamber measured ones. Since statistical fluctuations are unavoidable in any Monte Carlo calculations, denoising techniques from the image processing community could be invaluable tools in smoothing out the statistical noise in the dose distributions. The two digital filters assessed in this work are a Gaussian filter and a median filter. The median filter preserves the beam edges better than the Gaussian one. The smoothed isodose curves also have shifts within the recommended tolerance. This study indicates that the DSM, possibly together with denoising techniques, is a good candidate for IMRT calculations. Further studies should be carried out to confirm the DSM performance over a wider range of assessments including the modelling of higher energy linacs.