Speaker
Description
Accurate timing characterization of detected radiation events in Positron Emission Tomography (PET) provides Time-of-flight (TOF) information for detected coincidences, which improves the signal-to-noise ratio of the reconstructed images. In this work, we propose a method to train machine learning (ML) models to assign accurate timestamps to events measured in radiation detectors making use of just the signals detected by individual detectors, without the need of resorting to other detector signals. We performed a proof-of-concept study using four different neural network (NN) architectures, trained with event pairs constructed with the measured signal and a delayed version of it. The delay is used as a label in the training process. The proposed trained model was evaluated with data acquired with a $^{22}$Na point source using a pair of LaBr$_3$(Ce) crystals in the shape of truncated cones coupled to fast photomultiplier tubes (PMTs). The detected $\gamma$-ray pulses were sampled at 5 Gsamples/s. A decrease of 15$\%$ in Mean Absolute Error (MAE) and of 16$\%$ in Coincidence Time Resolution (CTR) compared to the analog constant fraction discrimination (A-CFD) -from 284 $\pm$ 3 ps to 238 $\pm$ 1 ps- method was achieved. The proposed ML methods requiere moderate computing power and are fast enough (>250kcps) for its practical implementation in PET scanners with many detectores modules.
