Using Longitudinal MRI to Manage Proton Range Uncertainty for Pediatric Proton Craniospinal Irradiation

Abstract

Clinical evidence has shown that proton therapy can effectively reduce side effects for pediatric patients undergoing vertebral body-sparing craniospinal irradiation (VBS CSI), compared to conventional photon treatment modalities. However, radiation-induced growth impairment remains challenging for VBS CSI due to proton range uncertainty, compromising vertebral body sparing for growing children. Previous studies have shown that fatty marrow replacement can be observed in vertebral bodies 4-48 weeks after treatment is complete. This study aims to detect and quantify the fatty marrow replacement in vertebral bodies using longitudinal magnetic resonance (MR) to manage proton range uncertainty. A prospective clinical trial of proton VBS CSI was designed, and ten pediatric patients were enrolled with prescribed doses of 15-36 Gy. The thecal sac and neural foramina were the clinical target volumes, and a Monte Carlo planning system was used to robustly optimize treatment plans with a 3.5% range margin. We analyzed patients' T1/T2 MR images acquired before, during, and after proton treatment to investigate the hematopoietic marrow transformation induced by irradiation. A metric was defined to calculate the ratio of fatty and hematopoietic marrow based on relative MR intensity histograms. We proposed a machine learning method via Gaussian fitting process (ML-GFP) to explore hidden correlations between marrow transition and radiation dose to 2 cm3 of the bone marrow (D2cc). We also leveraged this method to embed uncertainty to support potential proton range management for VBS enhancement. The results indicated that fatty marrow replacement could be observed during inter-fractional treatment. For instance, an individual patient showed that the fatty marrow generation ratios were 0.54, 0.74, and 0.45, corresponding to 11, 18, and 65 days after the treatment started. Using ML-GFP, the fatty marrow transition was found to be quadratically correlated to treatment fractions, and the maximum transformation ranged from 40 to 50 days. Then marrow regeneration was observed due to the decrease in fatty marrow ratios. The fatty marrow ratios were also positively correlated to the D2cc doses ranging from 10 Gy to 36 Gy. Limited by insufficient low-dose data, the ML-GFP model extrapolated the data to predict the marrow transformation below 10 Gy. We demonstrated the feasibility of using non-invasive longitudinal MR to quantify the fatty marrow transition from inter-fractional treatment. Based on this prospective study, the method can detect early fatty marrow generation in vertebrae caused by proton irradiation due to the conservative range margin used for robust optimization. The proposed method could be used to validate the actual proton range, allowing an accurate range margin to be defined to preserve bone marrow. Future investigation will likely focus on clinical implementation to improve life quality for pediatric CSI patients.