K. T. Griffin1, K. Pithadia2, Y. S. Yeom3, K. Bolton4, L. Z. Braunstein5, L. Morton2, and C. Lee1; 1National Cancer Institute, National Institutes of Health, Rockville, MD, 2National Cancer Institute-DCEG, Rockville, MD, 3Department of Radiation Convergence Engineering, Yonsei University, Wonju, Gangwon-do, Korea, Republic of (South), 4Washington University School of Medicine, St. Louis, MO, 5Memorial Sloan Cancer Center, NYC, NY
Purpose/Objective(s): Clonal hematopoiesis (CH) is characterized by genetic alterations of the hematopoietic stem and progenitor cells and is a known precursor to leukemia. Recent studies have suggested that external beam radiotherapy is significantly associated with CH. The objective of this work is to reconstruct the dose and dose-volume metrics for the active bone marrow in a cohort of patients treated with photon radiotherapy. These dose estimates will be connected to tumor and blood sequencing data for use in future analyses on the dose-response relationship between radiotherapy and CH, as well as therapy-related leukemia. Materials/
Methods: Electronic treatment records (DICOM) were retrieved from multiple institutions across one hospital system for 639 patients (5 pediatric; 634 adult) having nonhematologic cancers treated between 2003 and 2011 using photon radiotherapy. Active bone marrow dose was calculated for each patient by superimposing the recorded DICOM RT-Dose matrix calculated by the hospital’s treatment planning system onto a whole-body computational human phantom of nearest body size to the patient; thirty-four bone sites come pre-segmented in these phantoms, allowing the use of site-specific bone marrow fractions to calculate overall marrow dose. Alignment of the phantom to the patient’s anatomy was performed using an in-house 3D registration algorithm on the skeletons of the two anatomies. Evaluation of the alignment accuracy was performed both automatically and manually, through calculation of the Dice Similarity Coefficient and through visual inspection, respectively. Results: Out of the 639 retrieved patients, the automatic registration method provided acceptable segmentations for 478 patients (3 pediatric; 475 adult). The mean active bone marrow dose estimates for these patients ranged from 0.017 to 18.8 Gy. Comparison of dose across treatment regions showed that fields involving the pelvic region delivered the highest mean active bone marrow dose per prescribed dose (median: 0.245 Gy/Gy, n=176). Intracranial fields delivered the lowest of the categorized field types (median: 0.0108 Gy/Gy, n=27), with the lowest specifically being those intracranial fields that utilized stereotactic radiosurgery techniques (median: 0.0024 Gy/Gy, n=6). Conclusion: An automated patient-to-phantom registration method has been used to provide active bone marrow dose metrics for a large-scale study on CH and leukemia risk. Work is currently underway to improve bone site contours using deep learning-based automatic segmentation on the patient’s computed tomography image set.
