3D position determination in LaBr3 monolithic crystals with convolutional neural networks

Convolutional neural networks have been used to determine the positions of photon interactions in a monolithic crystal. Data for the model training process was generated through Gate 8.2 simulations. Realistic simulations of a collimated source of 511 keV photons impinging a LaBr 3 crystal coupled to a silicon photomultiplier array were carried out. The light collected by the pixels of the photomultiplier array was recorded to generate images employed to train and test the model. The optimal architecture was found using the Weights & Biases platform tools. The model's predictions were evaluated by analyzing the FWHM, euclidean distance and mean absolute error. Once an optimal model was found for predicting x and y coordinates, the DOI was estimated using the same model. Results were compared with previously published works, and the model performed well both in the 2D and 3D studies. This method shows promise for determining photon interaction positions in medical imaging applications.