Data-Driven Modeling of Neutron Radiation Effects on Relative Biological Effectiveness

Nuclear detonations (NUDET) unleash massive amounts of radiation, impacting exposed personnel based on their distance, detonation type, and environment. To enhance the DTRA/DoD's exposure analysis capabilities, a deep understanding of Relative Biological Effectiveness (RBE) is crucial for accurately evaluating the combined effects of neutron and gamma radiation. To incorporate RBE into calculations, robust correlations must be developed using extensive training data that consider variables such as dosage, dosage rate, tissue type, lethality endpoints, and animal test specimens. Leveraging the vast archives of historical radiation studies dating back to the 1950s, we can harness these data to develop precise RBE correlations with modern machine learning (ML) and statistical analysis methods, significantly reducing costs and eliminating the need for numerous new animal experiments. However, these invaluable datasets are currently scattered across various archives and not in a format suitable for analysis. To elevate the technology readiness level (TRL) and fully utilize these datasets, CFD Research is partnering with Northwestern University to create a centralized, cloud-based database and advanced software suite. This will organize, store, and analyze meticulously selected neutron radiation studies.In Phase I, the database will employ automated methods to extract large amounts of data from spreadsheets and PDFs. The user-friendly interface will sort and display datasets, provide metadata on experiments and animal samples, and perform statistical analyses, such as histogram distributions and sample averages, to assess data quality. The software will also generate spreadsheets for use in external ML tools and process the most promising historic neutron radiation studies to calculate RBEs and populate the database. In Phase II, we will enhance data extraction and visualization capabilities, integrating automated ML tools to develop predictive RBE correlations. The software will perform feature ranking, sensitivity analysis, and uncertainty quantification, automatically retrieving data from historical PDFs to expand the dataset efficiently. The resulting ML correlations will be formatted for easy application in NUDET simulations, advancing DTRA's scenario modeling and improving predictions of radiation exposure for personnel. This cutting-edge initiative promises to revolutionize radiation exposure analysis, providing the DTRA/DoD with unprecedented precision and efficiency in safeguarding our service members.