Smooth intensity maps for IMRT

The optimization process in intensity-modulated radiation therapy (IMRT) is closely related to an inverse problem and is therefore ill-conditioned. Besides, it is a computationally complex and intensive task, leading to numerical instabilities and solutions sensitive to high-frequency perturbations. For these reasons, a regularizer is needed for smoothing the intensity maps generated, thus removing the unwanted noise. However, regularization has to be performed taking into account the physical meaning of the solution, i.e., the delivery of radiation. In this work, we propose a new term within the objective function, which uses a discrete Laplacian filter adapted to each position for excluding the inactive neighbours that do not radiate a tumour. The term was tested in six cancer cases and the results show that the final decomposition of the maps for delivering the treatment are improved in terms of the number of segments and total number of monitor units, thus reducing the treatment time, while the quality of the solution is equivalent to the original.