Inverse radiation treatment planning using the Dynamically Penalized Likelihood method

In this paper we present a new method of solving the inverse radiation treatment planning problem. The method is based on a Maximum Likelihood Estimator with dynamically changing penalization terms. The resulting Dynamically Penalized Likelihood (DPL) algorithm achieves a dose distribution of excellent uniformity in a tumor volume and a much lower dose in regions containing sensitive volumes. A simple model of a patient and of energy deposition has been used for the initial results presented: a two-dimensional computer generated phantom and monochromatic x rays, without scattering. Three two-dimensional problems are solved with the DPL algorithm, corresponding to different size and spatial relationships between the tumor and sensitive tissue volumes. The results show that the DPL algorithm is robust and flexible; it only requires moderate computation times and leads to promising solutions, even in rather difficult problems. The results encourage the extension of the present work to more realistic therapy situations.