ESTRO 37 Abstract book

ESTRO 37

S487

3.5% sys ±1mm rand range error with 1mm sys ±2mm rand setup error perfectly correlated within a beam. Our APM results are compared against sample mean and standard deviation of the metric/objective functions for 3 patient cases. Results Closed-form computation of mean dose, DVH points and F +/- was performed by integration (1). For EUD, we used Taylor expansion (2). Approximations for d min/max are found exploiting that with larger negative/positive k , EUD converges to the min/max of the data, respectively. Fig. 1 shows analytical and sampled moments of DVHs for each case with 30 fractions.

Tab. 1 completes with validation of E[ F +/- ]. rel. abs. diff. [%] APM vs sampling 1 frac 30 frac F +/- O F +/- T F +/- O F +/- T prostate 0.6 22 0.4 43 paraspinal 3.1 0.3 2.3 0.2 intracranial 1.1 0.2 0.5 1.0

Factoring in sampling error, most analytical moments are in reasonable agreement with sampling, except for the prostate boost. Conclusion Analytical formulations for E[ I ( d )] and Var[ I ( d )] of common dose metric/objective functions I ( d ) were derived and evaluated for 3 patients. We found good agreement between APM and sampling, especially for fractionated therapy. All derivations build upon an E[ d ] and Σ d estimate. Hence our work is generally applicable, also in combination with random or importance sampling. The APM formulations provide concrete handles on the error bars of dose quality metrics/objectives which may allow for more intuitive probabilistic treatment planning workflows in the future. PO-0910 Nonlinear robust optimization methods for 4D treatment planning in carbon ion therapy M. Wolf 1 , C. Graeff 1 , K. Anderle 1 1 GSI Helmholtz Centre for Heavy Ion Research, Biophysics, Darmstadt, Germany Purpose or Objective In conventional treatment planning safety margins on CTV are used to handle uncertainties during treatment. Recently, robust optimization has proven to be a more effective method, especially in intensity modulated particle therapy (IMPT). We introduced robust non-linear biological optimization into our treatment planning system for carbon ions. Additionally, we mitigated motion with a state of the art conformal robust 4D optimization. Material and Methods We implemented a worst case scenario method with 9 different scenarios: nominal scenario, under- and overestimation of particle ranges and anisotropic shifts of patient’s isocenter in the 6 major anatomical directions. In every iteration step of the optimization RBE-weighted doses are calculated for all scenarios from the current set of beam fluences. In the conformal 4D optimization approach a treatment plan is calculated for each motion phase resulting in a library of plans, delivering a homogeneous dose to each motion phase, which requires synchronized beam delivery (see figure 1).

Fig. 2 shows a similar comparison for the remaining metrics for 1 and 30 fractions.