Prediction of high-resolution portal images for treatment verification in radiotherapy by coupling Monte Carlo simulations to non-parametric Bayesian denoising
Résumé
In order to ensure safety and efficiency in the delivery of Intensity Modulated Radiotherapy (IMRT) treatments, amorphous silicon (a-Si) electronic portal imaging devices (EPIDs) are now routinely used for dosimetric verifications. A straightforward way to do this is to compare the measured EPID image with a reference image which can be calculated in the treatment planning system (TPS).
Among the various prediction models proposed to compute the reference image, Monte Carlo (MC) simulation is highly attractive due to its ability to predict accurately and directly the dose to the detector in a wide range of configurations, without requiring the conversion to dose in water. However, MC remains to date so time consuming that MC computation of portal images with meaningful statistical uncertainty is only feasible for pixel sizes around 2 mm, which could lead to misinterpretations due to the loss of image resolution.
To overcome this problem, we developed a new method to compute high resolution reference EPID images in reasonable computing times. This method is based on the denoising of MC calculated images with a non parametric Bayesian algorithm called DPGLM (for Dirichlet Process Generalized Linear Model), particularly suited to very noisy images. In this study, an accurate model of an a-Si EPID was first developed and validated. Then, the performances of the denoising algorithm were assessed and compared to those obtained with IRON (Iterative Reduction Of Noise), a denoising algorithm usually employed in radiotherapy. Finally, MC images of a head-and-neck treatment plan were computed and denoised, and then compared to acquired EPID images.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...