Chemical radioanalysis of production of positron-emitting radioisotope Gallium-68 via (p,n) and (p,2n) reactions using compact cyclotron for tomography applications.

In this study, the proton-induced reactions of Zn and Ga aimed at generating Ga were simulated and modeled using Talys code and neural network software. In the first step, both targets were simulated under different proton energies and at different bombardments times to generate a total of six thousand data. Then, the obtained data from the Talys, including the various cross-sections, contaminations, the main product i.e. Ga, and other options were completely saved in the output file. Afterwards, the inputs of the neural network were selected from the output of the Talys by analyzing and considering most of the key features. A total of four inputs, two of which are different energies related to the reaction, the other is the process sequence and the fourth input is the bombardment time, were recognized as suitable inputs and the model was trained differently depending on the type of target. The selected model was a feed-forward neural network with 5 nodes in a middle layer, which was able to estimate the output of Talys code by changing the input parameters with extremely high accuracy. Two different models including the main model for estimating the output of the main sample (product) and the sub-model for estimating process pollution or impurity were trained, and then the trained model was tested on the deduced process data. The implementation results fully demonstrated the high accuracy of the method. The neural network model is much easier to implement than the Talys code, and its execution speed is very high. In addition, it can be used appropriately as a system alternative for optimization and different structures in medical and biological engineering.