Region-of-Interest Adjoint Functions as an Initial Guess for Iterative Inverse Treatment Planning

Michael Kowalok, University of Wisconsin

Photo of Michael Kowalok

Iterative least-squares minimization is a well-established approach for optimization of beam weights for intensity-modulated radiation therapy (IMRT). Based on a convex objective function, this technique offers a global minimum but does not guarantee a unique final solution. Different initial weights may affect the number of iterations needed for convergence and the quality of the solution as evaluated by a dose-volume histogram (DVH).

This work investigates the use of region-of-interest (ROI) adjoint functions as a tool for developing initial beam weights. The adjoint function for an ROI is the sensitivity of the dose in the ROI to every possible beam position and direction. Hence, adjoint functions enable an optimizer to determine with a simple lookup what beam positions are most effective at delivering dose to a target or sensitive structure. This work will demonstrate that this information is readily available, and that tumor adjoint information may be combined with that of sensitive structures to develop an initial guess of beams and absolute beam weights. The role of adjoints in computing update factors for a least-squares iterative scheme is also illustrated. The utility of adjoint-based initial guesses for improving either the convergence of the algorithm or the final dose distribution is discussed.

Abstract Author(s): M.E. Kowalok, (Department of Medical Physics)<br />S. Yoo, (Department of Medical Physics)<br />D.L. Henderson, (Department of Nuclear Engineering and Engineering Physics)<br />University of Wisconsin-Madison