104 - Deep Learning-Based Personalized Blood Dosimetry (OncoBlood) for Circulating Blood Cell Exposure Predicts Radiation-Induced Severe Lymphopenia and Survival Outcomes in Craniospinal Irradiation
Presenter(s)
S. H. Seo1,2, N. Kim3, J. S. Chang4, S. G. Moon2, S. Kang3, S. Y. Park3, D. H. Lim3, J. Shin5, and J. S. Kim6,7; 1Department of Clinical Medicine, Sungkyunkwan University School of Medicine, Suwon, Korea, Republic of (South), 2Oncosoft Inc., Seoul, Korea, Republic of (South), 3Department of Radiation Oncology, Samsung Medical Center, Sungkyunkwan University School of Medicine, Seoul, Korea, Republic of (South), 4Department of Radiation Oncology, Yonsei Cancer Center, Yonsei University College of Medicine, Seoul, Republic of Korea, Seoul, Korea, Republic of (South), 5National Cancer Institute/National Institutes of Health, Rockville, MD, 6Oncosoft, Seoul, Korea, Republic of (South), 7Department of Radiation Oncology, Yonsei Cancer Center, Yonsei University College of Medicine, Seoul, Korea, Republic of (South)
Purpose/Objective(s): Craniospinal irradiation (CSI) is widely adopted in pediatric central nervous system tumors and is also utilized for leptomeningeal metastases in solid tumors. Despite its therapeutic efficacy, CSI frequently induces severe hematologic toxicity, particularly radiation-induced severe lymphopenia (RISL), which has been associated with poor oncologic outcomes across various cancer types. Accurately estimating the radiation dose to circulating blood cells (rCBC) is challenging, yet it may serve as a predictive biomarker for RISL. In this study, we introduced OncoBlood, an innovative platform that combines deep learning-based whole-body segmentation, individualized blood volume estimation, and user-friendly cCBC calculation, enabling personalized assessment of rCBC during CSI and its impact on RISL development.
Materials/Methods: Patients who received CSI from January 2018 to June 2024 were retrospectively reviewed. Laboratory data were collected at baseline, weekly during CSI, and 1-2 weeks after CSI. RISL was defined as the occurrence of grade 4 lymphopenia during or after CSI. To calculate rCBC, we implemented HEDOS (Hematological Dose) within a commercial AI contouring platform. This system automatically performed deep learning-based segmentation of the whole body, brain, and spine and integrated a flow model, temporal distribution analysis, and compartment dose calculation to estimate individualized blood dosimetry. The developed OncoBlood module was validated to ensure accuracy, and rCBC dose was ultimately calculated for all patients included in this study.
Results: Among 180 patients (median age: 12 years, range: 2-68), 58.9% were under 15 years. Embryonal tumors (e.g., medulloblastoma, atypical teratoid rhabdoid tumor) were the most common diagnosis (n=86, 47.8%), followed by germ cell tumors (n=54, 30.0%). Proton therapy was administered in 162 patients (90.0%) with a median total dose of 23.4 Gy in 13 fractions. Overall, RISL occurred in 76 patients (42.2%), and was associated with significantly worse overall survival and event-free survival outcomes compared to those without RISL (3-year, 63.1% vs. 85.4%, p=0.002; 51.7% vs. 66.9%, p=0.034). Among various rCBC metrics, D5% was the most contributing factor of RISL development (LASSO coefficients, 0.922). Also, the number of RISL events during CSI correlated positively with rCBC D5% (Pearson r=0.522). Stratification by rCBC D5% (cut-off, 3.7 Gy) effectively differentiated overall survival and event-free survival outcomes (p<0.001).
Conclusion: We successfully developed the OncoBlood platform, enabling precise rCBC dosimetry during CSI. Our findings confirmed that RISL events during CSI showed a dismal impact on oncologic outcomes regardless of cancer types. Among various rCBC metrics, rCBC D5% is the most critical determinant of RISL risk. Further studies are needed to validate these findings and to develop strategies to mitigate RISL risk based on rCBC dosimetry.