Crystal identification for light-sharing Prism-PET detectors

Prism-PET is a single-ended readout PET detector module featuring a segmented array of prismatoids light guides that enhance the light sharing. We recently constructed a one-ring prototype brain-dedicated scanner using the Prism-PET detectors. However, the employing of light-guide and 4-to-1 crystal-to-SiPM coupling distorts the flood histogram, with the distortion pattern varying across the detector blocks. Although manual segmentation of the flood histogram to create the crystal look-up-table is reliable, it is time consuming as the prototype has 40 detector blocks with 10240 crystals. In this study, we developed an automatic flood histogram segmentation method to identify the crystal boundaries on the flood map. Due to the prismatoid light-guides, the corner, edge, and center crystals exhibit different light-sharing patterns with neighbor SiPM, resulting in different SiPM energy saturation effects and distinguishable energy peaks before saturation correction. For each SiPM, we detected the peaks of the energy histogram and applied energy filters to isolate the events belonging to each energy peak. The coordinates of the filtered events on the flood histograms were calculated using the truncated center of gravity (TCOG) method. A k-means segmentation algorithm was employed to segment the photoelectric spots on the flood map into crystal masks. After obtaining the mask of each crystal, we used the nearest neighbor (NN) method to determine the region and boundary of each crystal on the flood histogram. We applied this method to 40 Prism-PET detectors and the algorithm can reliably identify the boundaries of all crystals, enabling accurate crystal identifications, filtering of inter-crystal scatter events, and crystal-specific corrections. We believe this method can also be extended to segment the floodmap generated from the interleaved multiplexing readout for our next full-ring Prism-PET scanner.