Simultaneous partial volume correction and noise regularization for cardiac SPECT/CT

Partial volume correction (PVC) methods typically improve quantification at the expense of increasing image noise. In this study, we developed a novel voxel-based PVC method that incorporates anatomical knowledge to improve quantification while suppressing noise for cardiac SPECT/CT imaging. In the proposed method, the SPECT images were first reconstructed using anatomical based maximum-a-posteriori with Bowsher´s prior (AMAP) to penalize noise while preserving boundaries. A template was then obtained by assigning initial estimations of the mean activity in the target regions on a segmented contrast CT dataset. This template was forward projected, and reconstructed using AMAP to derive a correction map that reflects the partial volume effects (PVE) introduced by both the intrinsic system resolution and the smoothing applied during reconstruction. This map was then applied to the non-PVC SPECT images on a voxel-by-voxel basis to correct PVE. To evaluate the proposed Simultaneous PVC and Regularization method (SPR), we first simulated two SPECT scans with ^99mTc-tetrofosmin and ^99mTcred blood cell (RBC) tracers on a dedicated cardiac multiple pinhole SPECT/CT at low count levels. We then applied the proposed method on a dog study injected with ^99mTc-RBC tracer. We also retrospectively rebinned the dog study into shorter acquisitions to assess the performance of SPR on high-noise low-count data. The proposed method was compared to MLEM, a conventional multi-target correction (MTC) PVC method applied on the MLEM reconstruction and the AMAP reconstruction, in terms of quantification, noise level and visual quality. The results show that MTC corrected PVE but amplified noise and yielded the worst performance among all the methods tested on the low-count data. AMAP suppressed noise effectively, however, it was unable to recover the activity in the myocardium and other organs. SPR yielded superior performance in both quantitative assessment- and image quality for visual detection by recovering the activity in each organ while suppressing noise. The results also show that SPR was robust to the initial estimation of the regional mean values in the template.