Event Localization in Continuous Crystal Scintillation Cameras using Distribution Matching Neural Networks

R.G.COLMEIRO; C.VERRASTRO; D.M. MINSKY; GROSGES, T.

.

Congreso; IEEE- CAE- EAMTA; 2021

IEEE

(Trabajo final ya aceptado y en prensa, peresentación en abril)The localization of gamma interactions on solid scin-tillation crystals is normally estimated from the light distributionof the crystal?s scintillation. The estimation of the interactionposition is not an error nor bias free process, mostly due to lightreflections within the crystal. Complex models exist to reduce thiseffect. These methods often rely on calibration measurementsperformed with collimated beams along the crystal?s surface,making the process slow and troublesome. This paper presentsa method to improve the interaction position estimation basedon neural networks and an interaction distribution matchingloss. The method requires only a flood acquisition with knowninteraction distribution. The neural network does not need adataset with matched ?light distribution-interaction position?data. The method is tested using an experimental acquisition andMonte Carlo simulation of a large scintillation camera composedof a 406.4 × 304.8 × 25.4 mm3 continuous NaI(TI) crystal. Themethod improves the interaction localization and reduce the edgepacking effects of the center of gravity algorithm, increasing thedetector?s effective area from 48.7% to 72.1%. Moreover themethod is able to estimate depth of iteration.