SU‐E‐T‐845: Benefits from Achillesˈ Heel of Intensity‐Modulated Proton Therapy via Worst‐Case Robust Optimization

Abstract

Purpose: to minimize the influence of ranger and patient setup uncertainties in intensity‐modulated proton therapy (IMPT), thus achieving the goal of “what you see is what you get” results in IMPT. Methods: A new efficient optimization scheme, which uses memory‐distributed parallelization on a multi‐processor system, is devised. Our method minimizes the possible hot spots in target volumes by penalizing not only underdosing but also overdosing in the target volumes in the objective function. Range and setup uncertainties are considered simultaneously with some simplifications. The optimization is done using the L‐BFGS method and parallelized with beamlet domain decomposition. Our approach is essentially not restrained by the memory available at one computing node and can easily be expanded to address more demanding problems in the future. Results: Compared with conventional treatment plans optimized based on planning target volumes (PTVs), our method yields plans that are considerably less sensitive to range and setup uncertainties and provides better protection for organs at risk. After robust optimization, the dose distribution is static to the CTV regardless of range and/or setup uncertainties, therefore reducing the sensitivity of the CTV coverage. This unique characteristic allows that the dose distribution is not necessarily conformal inside the whole volume of PTV, but rather a much smaller volume enclosing the CTV. Better normal tissue protection is thus accomplished by underdosing the margin of the PTV while not compromising the CTV coverage. Conclusions: Our results demonstrate the importance of robust optimization and challenge the effectiveness of the use of PTV to account for uncertainties in IMPT planning. Actually we can get some benefits rather than impediments from the sensitive dose distribution to uncertainties of IMPT. Thus, ideally all IMPT plans should be robustly optimized and our approach has great potential to make previously unreliable IMPT plans reliable.