Development of a portal dose image prediction algorithm for arbitrary detector systems

Abstract

Portal imaging was originally developed for geometric treatment verification of photon beams used in cancer radiotherapy. More recently, portal imaging systems have been successfully used in dosimetric treatment verification applications. Many of the proposed dosimetric applications involve the accurate calculation of a predicted portal dose image, including both primary and scatter dose components emerging from the patient. This thesis presents the development of a two step model that predicts dose deposition in arbitrary portal image detectors. The algorithm requires patient computed tomographic data, source-detector distance, and knowledge of the incident photon beam fluence. The first step predicts the photon fluence entering a portal imaging detector located behind the patient. Primary fluence is obtained through simple ray tracing techniques, while scatter fluence prediction requires a library of scatter fluence kernels generated by Monte Carlo simulation. These kernels allow prediction of basic radiation transport parameters haracterizing the scattered photons, including fluence and energy. The second step of the algorithm involves a superposition of Monte Carlo-generated pencil beam kernels, describing dose deposition in a specific detector, with the predicted incident fluence of primary and scattered photons. This process is performed eparately for primary and scatter fluence at high and low spatial resolutions respectively, and yields a predicted planar dose image. This algorithm is tested on a variety of phantoms including simple slab phantoms and anthropomorphic phantoms. Other clinical parameters were varied over a wide range of interest, including 6, 18, 23 MV photon beam spectra and 10-80 cm air gap between phantom and portal imaging detector. Both low and high atomic number detectors were used to verify the algorithm, including a linear array of fluid ionization chambers and a solid state, amorphous silicon detector. Agreement between predicted and measured portal dose is better than 5% in areas of low dose gradient (<30%/cm) and better than 5 mm in areas of high dose gradient (>30%/cm) for the variety of situations tested here. It is concluded that this portal dose prediction algorithm is fast, accurate, allows separation of primary and scatter dose, and can model dose image formation in arbitrary detector systems.