Noise rejection in monolithic PET detectors

In PET detectors based in monolithic scintillators the interaction coordinates of each photon impact, in particular the 511 keV energy, can be determined from the light distribution sampled at the photosensors pixels. However, some of these distributions can be produced by pure dark noise from the SiPM photosensor, multi-interaction events with long travelled distance, pile-up, etc. Early rejection of these events can improve the SNR of the detector block and, as a result, in the also in reconstructed PET image. Moreover, this early rejection can reduce the statistics necessary to reconstruct those images preserving the SNR. To perform the rejection of noisy distributions, a comparison between a theoretical light distribution and the measured one is made by means of the Pearson product-moment correlation coefficient. This coefficient provides a metric, with range [-1, 1], suitable for assessing the similarity between the two distributions and, thus, allowing the rejection of uncorrelated events. The theoretical model of the light distribution has been obtained using a model based on Montecarlo simulations of a detector block formed by a thick LYSO scintillator crystals (50×50×20 mm
³
). We carried out an experiment based on two detector blocks with the described monolithic LYSO scintillators coupled to a custom photosensor SiPM array. After analyzing the FWHM of small size
²²
Na sources, an improvement of the SNR from 2.2 to 4.1 was observed when the filter was applied. In another test with reconstructed images using data belonging to the one ring prototype of the MindView brain PET insert also showed an improvement in the SNR. Moreover, the filter reduced the data used for reconstruction in about 20% but obtaining, as mentioned above, an improved image quality.