Wedge filters are routinely used in radiation therapy. At present, most radiotherapy treatment plans are generated by manually adjusting beam weights and wedge filters through a trial and error method. Such a process is both time consuming and labor intensive, especially when there are multiple fields or non-coplanar beams in the three-dimensional (3D) treatment plans. This could be automated by incorporating computer optimization into the treatment planning system. Ideally, such a system should require as little intervention by the planner as possible, yet should allow easy incorporation of any field restrictions or the planner’s clinical experience when desired. This paper reports the clinical implementation of a wedge filter optimization method. In radiotherapy treatment planning, many parameters need to be optimized, including the treatment modality, beam energy, number of beams and their directions, beam weights and wedge filters [2,3,11,15,18,20,21]. A full optimization of these adjustable parameters is out of the range of current technical capabilities. In this work, it is assumed that the treatment modality, beam energy, the number of beams and their directions have been preselected for a given patient, and our discussion will focus on the optimization of beam weights, wedge angles and wedge orientations (in terms of collimator rotation angles). We are interested in methods that can automatically select a suitable set of wedges, making full use of a patient’s 3D anatomic data. Studies have been performed previously on beam weight and wedge filter optimization based on the concept of the universal wedge [12,16], in which a wedged field with an arbitrary wedge angle is produced by the superposition of an open field and a nominal wedged field with appropriate weightings [11,25]. Assuming that there are J incident beams, the optimization of J beam weights and J wedge angles is therefore transformed into the optimization of 2J beam
weights. The technique was later extended to wedge optimization in 3D with the omni wedge method [8,26]. The algorithm, however, requires that the orientations of the nominal wedges be preselected. In practice, while this is not difficult for some simple clinical cases, it could be quite cumbersome and less intuitive when the number of incident beams in greater than three or four. For a J-field plan, one has to pre-select the orientations of the 2J nominal wedges out of 4J possibilities. This limits the clinical applications of these wedge filter optimization methods. In this paper, a significantly improved wedge filter optimization algorithm is described. The selection of the orientations of the wedges becomes an integral part of the algorithm, making it possible to fully automate the optimization of beam weights and wedge filters in the treatment planning process. The optimized plans can easily be delivered on machines using any method to create the wedged fields. The algorithm has been implemented in a clinical 3D treatment planning system (PLUNC, University of North Carolina) as an independent module. Because of the simplicity of the interface between the optimization module and the treatment planning system, it could also be implemented in any clinical treatment planning system. Application of the method to two clinical cases shows that it is able to generate optimal plans with reduced planning time and effort.