Integrating Structure Propagation Uncertainties in The optimization of Online Adaptive Proton Therapy Plans
L Nenoff1,2*, G Buti1,2,3, A Sudhyadhom4, G Sharp1,2, H Paganetti1,2, (1) Harvard Medical School, Boston, Massachusetts, USA (2) Department of Radiation Oncology, Massachusetts General, Boston, MA, USA (3) Universite Catholique de Louvain, Institute of Experimental and Clinical Research (IREC), Center of Molecular Imaging, Radiotherapy and Oncology (MIRO), Belgium(4)Radiation Oncology, Brigham and Women's Hospital, Boston, Massachusetts, USA
Presentations
MO-H345-IePD-F4-4 (Monday, 7/11/2022) 3:45 PM - 4:15 PM [Eastern Time (GMT-4)]
Exhibit Hall | Forum 4
Purpose: Fast structures definition is a large source of uncertainty in online plan adaption. Clinically, online adaption requires a manual check of target and organ structures in the proximity to the target. This study investigates methods of including uncertainties of propagated structures during optimization, without the need for manual correction.
Methods: Proton therapy plans with 50Gy-RBE in 5 fractions for 6 pancreas and 5 liver cancer patients were optimized in Raystation using the clinical constraints from the original MRI-linac treatment. Daily MRIs were deformed to the planning MRI using three different deformable image registration (DIR) algorithms to create deformed structure sets. Six strategies were investigated. ‘Simple’ directly uses each algorithm for propagated structures. ‘Conservative’ uses the intersection of all algorithms as target, and the union for organs. ‘Anatomically robust’ uses the Raystation built-in function for optimizing over multiple scenarios using the same image with structures from all DIRs. ‘Probabilistic’ adjusts the optimization weight according to how often voxels were classified as target or organ amongst the DIRs. ‘Ideal’ uses the clinical, MD corrected structures. ‘Not adapted’ doses were simulated by recalculating in the treatment plan without reoptimization. Each method was calculated on deformed CTs and evaluated using the clinical structures. For comparison, plans were scored by summing the weighted difference between clinical constraints and the corresponding DVH parameter. Lower scores mean better plans.
Results: The ‘Ideal’ strategy scored best (0 for liver/pancreas patients), followed by ‘Conservative’ (5.7 liver/6.9 pancreas). The ‘Simple’ performance varied (8.2-9.7 liver/7.8-13.3 pancreas), the best and worst DIR were patient dependent, but they generally scored better than ‘not adapted’ (12.7 liver/17.7 pancreas). ‘Anatomically robust’ (12.1 liver/17.6 pancreas) scored slightly better than ‘not adapted’. ‘Probabilistic’ performed worst (12.7 liver/27.3 pancreas).
Conclusion: The ‘conservative’ method performed best if no manual corrected structures are available for the plan optimization.
Funding Support, Disclosures, and Conflict of Interest: This study was supported by the Swiss National Science Foundation (P2EZP2_199943).
Keywords
Protons, Deformation, Optimization
Taxonomy
IM/TH- Image Registration: General (Most aspects)
Contact Email