Reconstruction using Depth of Interaction Information of Curved and Flat Detector Designs for Quantitative Multi-Pinhole Brain SPECT

Publication Type:

Journal Article


Journal of Nuclear Medicine, Volume 61, p.103 (2020)



103Objectives: Brain SPECT has many clinical applications, especially for cerebral blood flow and dopamine transporter imaging [1,2]. In this context, a dedicated brain imaging, multi-pinhole system, AdaptiSPECT-C, is being developed by our group. Recent studies in cardiac and small animal imaging have demonstrated that the use of curved detector could improve image quality, by reducing parallax errors due to the depth of interaction (DOI) effect [3,4]. In this simulation study, we proposed to investigate using reconstruction with DOI modeling the potential advantage in imaging performance of curved over flat detectors for 123I-IMP perfusion imaging using the AdaptiSPECT-C system. Methods: The AdaptiSPECT-C design used in this work consists of 26 detector modules, 158 by 158 mm2 in size, arranged around the patient’s head in three rings. The simulated detector modules were composed of a 8 mm thick NaI(Tl) crystal coupled to a 5 cm thick back-scattering compartment, representing components behind the crystal, to model 123I down-scatter interactions. Each detector module is associated with a 1.36 mm radius direct knife-edge pinhole aperture collimator. Two different system designs were considered, one based on curved detectors, and the other on flat detectors. The curved detectors were designed so that the radius of curvature corresponds to the detector to system center distance (e.g. 30.5 cm). This distance was the same for the flat detectors. GATE simulation [5] was employed to compute the system matrix [6,7] for both detector designs by forming the system response for the activity within each three-dimensional image-voxel of a 24 cm diameter sphere, thus including DOI variations and corrections in reconstruction [6,7]. An XCAT brain phantom [8] emulating 123I-IMP perfusion source distribution was simulated using the two designs. Data were acquired following three scenarios: noise free case (S1), equal number of counts comparison (S2) (e.g. 5.5M detected counts [9]), and equal imaging time comparison for the typical scan time (e.g. 30 min [10]) (S3). For S3, the total number of counts for the curved and flat detector designs, were respectively 9.24M and 9.17M. Projections were reconstructed with 3D-MLEM into images of 1203 voxels of (2 mm)3 and reconstruction compared to the ground truth image. The normalized root mean squared error (NRMSE) as well as percentage of activity recovery (%AR) for several brain regions were used to evaluate the image quality. Results: Only a small gain in volumetric sensitivity ( 0.8%) was obtained with the curved detector design. Qualitatively, the reconstructions for the curved and flat detector designs appear similar for all the 3 noise scenarios. Differences are mostly for the peripheral regions of the head where the differences in the obliquity of the gamma-rays passing through the apertures would be the greatest. Due to lower activity, those regions are also more impacted by noise. Quantitatively, a slight NRMSE improvement using curved detectors was seen. The curved detector design leads on average to the best ARs, especially for the striatum and putamen. Regions at the edges of the brain (e.g. cortex and cerebellum), more impacted by DOI effect, are similarly recovered by the two designs. Conclusion: We demonstrated that using curved instead of flat detector for AdaptiSPECT-C with solely centered pinholes leads to small improvement in sensitivity and image quality based on visual inspection, NRMSE, and activity recovery analysis. Flat detector associated with a sophisticated DOI correction was found to lead to similar results than those obtained with the curved detector. Further investigation will be performed using additional pinholes irradiating the 4 quadrants of the detectors which will increase the obliquity of the rays striking the detectors and may thus result in larger difference. Research Support: Grant No R01 EB022521 (NIBIB).

Research area: