ESTRO 2020 Abstract book

S420 ESTRO 2020

Center (HIT), the RBE-dependence on linear energy transfer (LET) and fractionation has to be investigated. Material and Methods The segments C1-C6 of the CSC of female Sprague Dawley (SD) rats were irradiated with increasing dose levels of 16 O- ions. Experiments were performed with either 1 or 2 fractions (fx) and at 4 different positions within a 6 cm Spread-out Bragg-peak (SOBP) covering the plateau-region (LET of 26 keV/µm) and different positions within the SOBP (LETs of 68, 100 and 140 keV/µm). Dose-response curves were measured for the endpoint paresis grade II (PII – palsy of the forelimbs similar to irradiation-induced myelopathy) within 300 days after irradiation. RBEs were calculated based on the TD 50 -values (dose at 50% complication probability) and using previously measured values for photons [Radiat Res. 2003 Nov;160(5):536-42]. Results For all positions RBEs rose from 1 to 2 fx. Also RBEs increased with increasing LETs up to 100 keV/µm and this increase was seen to be stronger for irradiations with 2 fx than with 1 fx. At 140 keV/µm, however, both the 1 and 2 fx studies showed RBEs lower than that at 100 keV/µm. Conclusion In accordance with our previous studies with 12 C-ions [Radiat Oncol. 2018 Jan 11;13(1):5] we found a clear fractionation- and LET-dependence of the RBE also for 16 O- ions. Excluding the data point at 140 keV/µm, comparison with the previously measured 12 C-ion RBEs at the same SOBP-positions in addition revealed a higher effectiveness of 16 O-ions. Interestingly, the RBE of 16 O-ions decreased beyond 100 keV/µm for both fractionation schedules, which might be interpreted as an “overkill-effect". This study provides the first systematic RBE-data for late effects in the CSC, which can be used to benchmark RBE- models. OC-0688 Risk factors for late brain lesions in proton treated glioma patients: ventricular proximity and RBE E. Bahn 1,2,3,4 , J. Bauer 1,2,3 , S. Harrabi 1,2,3 , K. Herfarth 1,2,3 , J. Debus 1,2,3,4 , M. Alber 1,2,3 1 Heidelberg University Hospital, Radiation Oncology, Heidelberg, Germany ; 2 Heidelberger Institut für Radioonkologie HIRO, Quantitative klinische Strahlenbiologie, Heidelberg, Germany ; 3 National Center for Tumor Diseases NCT, Integrative Radiation Oncology, Heidelberg, Germany ; 4 German Cancer Research Center DKFZ, Clinical Cooperation Unit Radiation Oncology, Heidelberg, Germany Purpose or Objective Late radiation-induced contrast enhancing brain lesions (CEBL) can be precursors for radiation necrosis. We employed voxel-based analysis of 110 proton treated low- grade glioma patients to investigate the role of linear energy transfer (LET) and of proximity to the ventricular system (VP) in CEBL localization. Material and Methods We analysed a retrospective cohort of 110 low-grade glioma patients treated with proton therapy during the years 2010-2015. Follow-up T1 and T2 MR images were rigidly registered and regions of interest were manually contoured. The contours at the earliest available MR image for each CEBL were recorded. We extracted dose and dose-averaged LET (LET d ) from the original radiation treatment plans by Monte Carlo recomputation with an in-house developed platform (based on FLUKA) on the original planning CT images. We aimed to create a predictive, not an explanatory model, which necessitates certain hypotheses about the

origin of CEBLs. CEBLs present typically as sparsely seeded, clearly delineated, round structures that grow in size over time. Thus, we hypothesized that each RCC originates from a single focus and we aimed to model this in terms of a voxel-wise Probability of Lesion Origin (POLO). We performed voxel-wise stratified 10-fold cross-validated logistic regression with the model coefficients dose (D), the dose-LET product LET d ∙D and VP (VP = 1: distance ≤ 4 mm; VP = 0: distance > 4 mm): Finally, we derived a patient-level risk model by extrapolating from each patient’s voxel-level distribution of POLO, under the assumption of a serial dose-volume relation (occurrence of CEBL depends only on the focal POLO and is statistically independent of surroundings). Results The voxel-level logistic model has an accuracy of: AUC = 0.94 (0.84-1.00). The model coefficients are: β 1 = -(27.4 ± 0.5), β 2 = (0.20 ± 0.01) Gy -1 , β 3 = (0.021 ± 0.002) (Gy·keV/µm) -1 , β 4 = (1.19 ± 0.83). This amounts to the linear LET-RBE relation:

RBE = 1 + 0.10 (keV/µm) -1 ·LET d .

The localization of CEBLs is accurately predicted: voxels from the 99 th POLO percentile are present in 57 out of 67 CEBLs. Figure 1 shows exemplary MR image slices from three patients, overlaid with dose and POLO distribution.

The extrapolated patient-level risk model has an AUC of 0.79 (0.67-0.87). For comparison, a logistic model based on the patient-level information prescribed dose has an AUC of 0.66 (0.58-0.77) and is significantly less accurate (p=0.003, based on likelihood-ratio). Conclusion Our findings present clinical evidence for two risk factors for late contrast change in the brain: an increased