Modeling Organ Dose from Industrial Radiography Sources: Parameter Sensitivity and Predictive formulation

Abstract Industrial radiography accidents, often involving high-activity sources such as ¹⁹²Ir, ¹³⁷Cs, and ⁶⁰Co, are among the most common radiation incidents globally. The absence of comprehensive dosimetric datasets and rapid predictive models for dose estimation in emergencies represents a critical gap. This study addresses this by systematically analyzing the impact of source-to-body distance, source height, irradiation geometry, and photon energy on organ- absorbed dose. Monte Carlo simulations, performed using MCNP6 with the ICRP reference voxel phantom, modeled doses for organs across anterior-posterior, posterior-anterior, and lateral irradiation geometries, with source distances ranging from 0.5 to 300 cm and heights spanning ground to upper torso levels. Results were validated against ICRP Publication 145. Three machine learning models—Random Forest (RF), XGBoost, and Lasso regression—were developed in Python (version 3.9.6). RF and XGBoost achieved high predictive accuracy (R² = 0.63–0.77, MSE = 0.007–0.010), with source-to-body distance identified as the most influential factor and irradiation geometry the least. Lasso regression provided a simplified predictive formula (R² = 0.56) for sensitive organs in rapid dose estimation in time-critical scenarios. These models offer a robust framework for precise clinical dose assessment and efficient safety evaluations in radiation emergencies.