Artificial Neural Network-Based prediction of production cross sections of the medical radioisotopes 67Ga and 89Zr

In this study, the excitation functions of the 67Zn(p,n)67Ga, 68Zn(p,2n)67Ga, and 89Y(p,n)89Zr reactions were investigated. Nuclear reaction simulations were performed using the TALYS 1.95 and EMPIRE 3.2 codes, by using Two Component Exciton Model, Geometry Dependent Hybrid Model, Exciton Model, and Hybrid Monte Carlo Simulation. In parallel, a feedforward artificial neural network (ANN) trained with the Levenberg-Marquardt (LM) algorithm was employed to predict the same reaction cross-sections using experimental data. Comparative analysis revealed a strong consistency between TALYS and EMPIRE calculations, ANN predictions, and experimental results, with the ANN providing enhanced predictive performance in energy regions where experimental data are sparse or uncertain. Furthermore, activation and yield calculations confirmed the feasibility of producing clinically relevant quantities of 67Ga and 89Zr under realistic irradiation conditions. In conclusion, the use of physics-based nuclear reaction codes with data-driven ANN models has been demonstrated to be useful in medical radionuclide production studies by providing a complementary framework for cross-section estimation.