PurposeInterdetector scatter (IDS) is a triple coincidence caused by the Compton scatter of an annihilation photon from one detector block to another which frequently occurs in small-animal positron emission tomography (PET). By finding the true lines-of-response (LORs) of annihilation photon pairs among three possible LORs in IDS events, we can utilize these recovered events to improve the sensitivity of PET systems. IDS recovery should be accurate to yield reliable images with relatively short scan times. We systemically investigated physical factors affecting IDS recovery performance, focusing on the reconstructed image quality of small-animal PET. We evaluated sensitivity increase, recovery accuracy, and image quality by applying different combinations of energy window, recovery scheme, and scanner properties. MethodsWe used GATE Monte Carlo simulation to acquire coincidence events from a NEMA NU 4-2008 image quality phantom using small-animal PET scanner with axial field of view of 55 mm and diameter of 64 mm. We first defined energy window criteria to obtain valid IDS events. Their role was to assign triple coincidences as IDS events and to restrict the number of LOR candidates to two. We tested three different energy windows around 511 keV. Second, we applied four different recovery schemes (maximum energy, Compton kinematics, neural network, and proportional) to assigned IDS events. To measure the effects of scanner properties, energy resolutions of 0-20% and one to four depth-of-interaction (DOI) layers were simulated. For every combination of the factors, we measured sensitivity increase and recovery accuracy. We also analyzed the reconstructed images for each IDS recovery method in terms of mean pixel intensity, noise, signal-to-noise ratio (SNR), contrast, and recovery coefficients. ResultsSensitivity increase depended on the energy window and energy resolution. The maximum increase in sensitivity was 33% when energy window of [250, 750] keV was applied. Higher energy resolution yielded larger sensitivity increase, especially for narrow windows. Recovery accuracy was affected by all the factors tested in this study. Accuracy increased with narrower energy window, and a neural network scheme was the most accurate. The better energy resolution and DOI capability improved accuracy by providing precise measurement of energies and interaction positions. In image quality analysis, noise and SNR were highly dependent on the sensitivity increase and energy window. When the same energy window was applied, SNR, contrast, and recovery coefficients were higher with higher accuracy of the scheme. Meanwhile, the proportional scheme yielded the best image quality among the schemes and reduced 20% of scan time to achieve the same SNR as that of double coincidence images. ConclusionsAs a fundamental research for real implementation of IDS recovery, we conducted a simulation study to evaluate the factors affecting sensitivity increase, recovery accuracy, and image quality. Sensitivity increase was dependent on the energy window and energy resolution, while the recovery accuracy was affected by energy window, recovery scheme, energy resolution, and DOI capability. In image quality analysis, sensitivity increase and recovery accuracy dominantly affected the noise and quantitative accuracy, respectively. Among the recovery schemes, the proportional scheme obtained the best image quality.
