Genetic Algorithm Modeling of 47Sc Production via Proton Irradiation of 49Ti

Abstract

The β^−-emitting radionuclide 47Sc is an attractive theranostic agent for nuclear medicine, offering both SPECT imaging capabilities and therapeutic efficacy for small tumors. However, its clinical deployment remains limited by production challenges, including insufficient radionuclidic purity, low yields for some reaction channels, and reliance on restricted infrastructure. In cyclotron-based irradiation of enriched 49Ti, long-lived impurities such as 48Sc and 46Sc are co-produced through competing reactions, complicating purification compliance and increasing the dosimetric burden for patients. Reactor-based alternatives are not universally accessible and present logistical constraints. In this study, we analyze the 49Ti(p,x)47Sc reaction using the TALYS 1.95G nuclear reaction code [1,2], calibrated against the only available experimental cross-section dataset for this route, recently reported by Pupillo et al. [3]. A genetic algorithm (GA) is employed to optimize twelve level-density parameters across six compound nuclei and nine reaction channels, including 46Sc, 47Sc, 48Sc, 43Sc, 44Sc, 42K, 43K, and 48V. This represents the first application of GA-based joint optimization across multiple reaction pathways for a medically relevant radionuclide, demonstrating the feasibility of integrated AI-driven calibration in nuclear modeling. TheGA-driven optimization significantly improves the agreement between TALYS description and measured data. However, the refined simulations confirm that high co-production of 48Sc and 46Sc are intrinsic to the proton–49Ti reaction path. As a result, this route remains unsuitable for routine clinical applications. These findings emphasize the potential of evolutionary algorithms for nuclear model tuning and underscore the need for alternative production strategies. The methodology provides a scalable framework for integrating AI into predictive nuclear theory and applied isotope research.