Euromedim 2006 :1st European Conference on Molecular Imaging Technology

Estimating accidental coincidences for pixelated PET detectors and singles list-mode acquisition

9 May 2006, 14:00

1h

Palais du Pharo, Marseille

poster • System simulation, design and implementation Poster Session :Simulation, Modeling, Reconstruction

Speaker

Dr Magdalena Rafecas (Instituto de Fisica Corpuscular (IFIC), Universidad de Valencia/CSIC)

Description

* Motivation: High spatial resolution and high sensitivity are mandatory in small animal PET. To achieve high resolution, pixelated crystal detectors with individual read-out are often employed. Sensitivity can be improved by reducing the scanner diameter, and also by lowering the low energy threshold (LET). The latter allows photons which undergo Compton interactions in the object or in the detector elements to be detected. A low LET also increases the number of detected accidental coincidences (randoms). To estimate the randoms rate, either a delayed coincidence window or the singles rate is used, although the data acquisition system usually constraints the choice. However, modern systems which are capable of acquiring singles list-mode data allow both techniques to be applied post-acquisition. This is the case for our small animal PET prototype MADPET-II, a high resolution scanner. * Objective: This work is aimed at studying the validity of random estimation techniques for low energy thresholds and singles list-mode data sets. We also investigate the role of inter-crystal scatter (ICS) in the estimation of accidental coincidences. * Methods: The scanner MADPET-II (under construction) currently consists of two opposing sectors, where each sector comprises three dual-layer detector modules. A layer consists of 4x8 LSO crystals coupled one-to-one to the channels of an APD matrix. The crystal surface is 2x2 mm2, with a thickness of 6 mm (front layer) or 8 mm (back layer). The scanner diameter is 71 mm. The present configuration has been simulated using GATE. Point sources were both measured and simulated, where the output in both cases was a singles list-mode data set. Additional simulations of sources of varying activity and geometry were also carried out. Coincidences were sorted post-acquisition in software using coincidence window widths from 5 ns to 60 ns, and LETs from 100 keV to 450 keV. The values of the time and energy resolution were 10 ns and 21% (at 511 keV), respectively. The number of random events was estimated (a) using a delayed coincidence window (DW method) and (b) the singles rate formula (SR method): Rij= 2 Tau Si Sj. The GATE file was also used to compute random events: A prompt coincidence was considered a GATE random if the two single photons detected within the coincidence window were related to different annihilations. * Results: For both simulated and measured data, the SR method predicted a higher number of random events than the DW method, and both estimation methods yielded more random events than the computed GATE randoms. This overestimation of randoms decreased with increasing LET values. The ratio R between estimated and GATE randoms, which accounts for the mismatch, depended on the source geometry, being smaller for extensive sources. On the other hand, R remained almost constant for coincidence windows larger than 10 ns. For LET=200 keV and for a centered planar source, R was approximately 1.6 for the SR method, and 1.15 for the DW method. For a simulated point source located outside the FOV (OFOV) and LET under 255 keV, the mismatch between the DW method and the computed GATE ranoms was R=1.12. We also observed that the number of prompt coincidences was larger than the number of GATE randoms. This was due to inter-crystal scatter coincidences, as the analysis of the simulated file showed. The contribution of ICS events to the promts was 27% for LET=150 keV and, as expected, diminished for higher LETs (10% for LET=200 keV). For LETs over 255 keV, the number of promts, GATE and DW random events was almost identical. The mismatch between DW and GATE randoms was less pronounced for this source location than for sources within the FOV. In the measurement of the OFOV source, scatter in the object also contributed to increase the number of prompts, so that the number of detected coincidences remained higher than the number of DW estimated randoms for all tested LETs. * Discussion and Conclusion: Our study shows the need to revisit the issue of random estimation when using pixelated crystals read out individually and low energy thresholds below 255 keV. The analysis of simulated data reveal that inter-crystal scatter plays a decisive role in the estimation of randoms, regardless of the method of choice. This is because pixellated detectors with individual crystal readout allow the single interactions related to a ICS event to be detected, if the LET is low enough. Moreover, since ICS single hits are related to the same annihilation, they may thus give rise to true coincidences. ICS can also cause more complex events to appear, such as those consisting of three single hits, two of them related through ICS to the same annihilation. The contribution of these complex events to the discrepancy between estimated and GATE randoms is currently under investigation. Our future work is also aimed at studying the role of energy resolution and object scatter in the estimation of random coincidences. For this purpose, further measurements and simulations will be carried out.