Calibration stability in a 1 mm 3 resolution, clinical PET system and its impact on real-time data processing and coincidence sorting

Real-time feedback of system performance, such as a non-diagonstic quality image to the operator, is important for clinical imaging where time is constrained. This task becomes more difficult in systems with large numbers of components. We are constructing a two-panel clinical PET system dedicated to imaging the breast that has 294,912 LYSO crystals read out by 4608 Position-Sensitive Avalanche Photodiodes (PSAPD). The system requires 907,776 calibration parameters to be estimated for an image to be produced. Applying stored parameters in real-time requires understanding their stability. We show that using previously stored calibration estimates causes no significant degradation of system measured energy resolution at 12.2 ± 0.1 %, with a photopeak shift of 3.6 ± 0.2 keV. Timing resolution is degraded by 0.3 ± 0.2ns to 15.0 ± 0.1 ns. Crystals are correctly identified for 93.5 % of events, with 3. 5 % of events being mispositioned within 1 mm. A singles event data structure is proposed and implemented given the measured calibration stability and used to benchmark the speed of singles calibration and coincidence sorting. We demonstrate we that we are able to process our systems output at a rate of 9.24 Mevents/s or 15.4 % slower than expected real-time. We show that this is I/O limited and that with storage upgrades we can process 16.5 % faster than expected real-time at an event rate of 12.7Mevents/s.