受動散乱陽子線療法を受けた小児患者の正常組織に対する個別化線量体積再構成ワークフロー

The United States National Cancer Institute has launched a large epidemiological cohort study comparing subsequent cancer risks from photon and proton radiotherapy in pediatric cancer patients. Clinical data, including electronic radiotherapy files, are being collected from 17 treatment centers across the United States and Canada for dose reconstruction and risk analyses. Although clinical dose distributions exported from treatment planning systems (TPS) are available, they typically represent only partial-body dose and are less accurate in heterogeneous regions with large density variations, such as bone or lung, and in areas far out-of-field. Monte Carlo (MC) simulations are considered the most accurate method for addressing these issues but can be computationally intensive. In this pilot study, we developed a framework for large-scale MC calculation of normal tissue dose for patients treated with passive scattering proton therapy. Our automated MC workflow was developed for a supercomputing environment to simulate patient-specific treatments, incorporating detailed modeling of the beamline nozzle and the patient's anatomy from computed tomography images. For patients with limited imaging data, our workflow extends partial-body images into whole-body anatomical models. Normal tissues on the patient images were delineated using automated segmentation tools. We tested the approach by calculating dose and dose volume for 30 patients treated at a single institution. For organs far out-of-field, the MC simulation method allowed characterization of scatter dose not captured by the TPS, spanning magnitudes from tenths to tens of cGy, depending on treatment technique, body size, and the organ dose being calculated. For most near-field normal tissues, MC and TPS dose differences were within 5%, but for organs near the distal beam edge, MC simulations occasionally revised dose estimates by multiple Gy. Neutron dose estimates ranged from 0.1 cGy far out-of-field to 15 cGy near-field. Use of the supercomputing environment allowed dose reconstruction in a practical timeframe, achieving statistical uncertainties of <5% for nearly all organ dose estimates, typically in under one h per patient. Future extension of these methods to the full study cohort will enable large-scale dose estimation and dose-response analyses to advance our understanding of the late health effects of radiotherapy.