Inverse treatment planning and integration of segmentation procedures

The inverse treatment planning procedure in IMRT consists of two basic tasks, the fluence optimization and the segmentation into multi leaf collimator (MLC) segments. Both tasks can either be run simultaneously or separated. We focus on the latter case. When using such a separation the fluence given by MLC segments and the optimized one differ. In the same way the quality of the treatment plan decreases. In this paper we discuss two approaches to reduce the negative effect of a separation of the fluence optimization from the segmentation procedure. The fluence smoothing concept is expected to find solutions with, in a mathematical sense, smoother fluences. A mandatory segmentation should then result in less segments and less decrease of plan quality. In the second approach we take the segments and their weights as input to a second optimization cycle which readjusts the weights, so the result is as close to the original demands as possible. We demonstrate the two methods using a 7-field plan applied to the Quasimodo phantom. The results show that the number of segments can be reduced by implementing an additional term to the target function. A penalization of local extrema however does not have the expected effect. In all cases, an additional reoptimization of the segment weights improved the segmented plans significantly. (orig.)