Voxel-wise case-control analysis of dose, linear energy transfer and symptomatic brainstem necrosis in pediatric brain tumors treated with proton therapy

INTRODUCTION: Brainstem necrosis is a rare but serious toxicity following radiotherapy for pediatric brain tumors. In proton therapy (PT), concerns have emerged around elevated linear energy transfer (LET) and the relative biological effectiveness (RBE) in relation to normal tissue damage. While previous studies have largely focused on asymptomatic image changes in adults, we investigated voxel-wise correlations between dose, LET, and symptomatic brainstem necrosis in a pediatric case-control cohort. MATERIALS AND METHOD: From a cohort of exactly 1200 central nervous system (CNS) tumor patients treated with double scattering proton therapy between 2006 and 2023, 48 patients were included in a case-control cohort. Twelve of these patients had developed brainstem necrosis following PT and 36 were matched controls (1:3 case-control ratio). Necrotic volumes were contoured on magnetic resonance images (MRI) and related to LET and RBE-weighted dose distributions from Monte Carlo simulations. Mixed effect modelling was used to investigate potential correlations between dose, LET and necrosis at voxel-level while accounting for intergroup heterogeneities. Further, a recently published prediction model for MRI contrast enhancement in posterior fossa tumor patients was utilized to evaluate whether its parameters demonstrated predictive value within our cohort. To control for dose, voxel matching was also performed between necrotic and healthy voxels to investigate LET, and LET-weighted doses were calculated and compared. RESULTS: Mean dose in necrotic brainstem voxels was 54.1 Gy(RBE) range: 17.7-58.7 Gy(RBE) compared to 42.6 Gy(RBE) 0.0-63.1 Gy(RBE) in healthy voxels. Mean LET in necrotic brainstem voxels with dose over 54 Gy(RBE) was 2.9 keV/µm 2.2-5.0 keV/µm in necrotic voxels and 2.7 keV/µm 1.9-5.0 keV/µm in healthy voxels. Mixed effect modelling revealed a significant positive interaction between dose and LET, indicating that elevated LET was associated with an increased probability of necrosis in high-dose regions. However, notable intergroup heterogeneities were observed, with more pronounced LET-related trends in specific case-control pairs. Dose matching similarly only found higher LET in the necrotic voxels of a subset of the case-control groups. The prediction model estimated higher probability of contrast-enhancement in the necrotic voxels, primarily driven by dose and most necrotic voxels being located in the brainstem pons. LET-weighted dose was significantly elevated in necrotic voxels compared to healthy voxels. CONCLUSION: Brainstem necrosis was primarily driven by dose, with elevated LET contributing to increased risk in high-dose regions. Nevertheless, considerable intergroup heterogeneity was observed, and dose-matched analyses did not support a uniform LET effect.