Artifact-free partial volume correction for ¹⁷⁷Lu-PSMA SPECT

Partial volume effect degrades the quantification accuracy of SPECT across the field-of-view at varying extent. However, traditional partial volume corrections (PVCs) usually apply a spatially invariant point spread function and deconvolution-based PVC may cause Gibbs ringing artifacts.

We aim to develop a spatially variant PSF (svPSF) enhanced PVC without Gibbs ringing artifact and noise amplification for ¹⁷⁷Lu-PSMA SPECT.

The svPSFs were generated through linear interpolations of Monte Carlo simulated ¹⁷⁷Lu point sources across the field-of-view. Reblurred Van Cittert (RVC) and iterative Yang (IY) were enhanced by applying svPSFs voxel by voxel. In RVC, the non-negativity constraint (NC) was further implemented by scaling the negative values in the intermediate images with an empirical coefficient. All PVC methods were validated via simulated SPECT data using a digital geometrical cylindrical phantom with five spheres and 20 XCAT phantoms, as well as real sequential ¹⁷⁷Lu-PSMA-617 SPECT data of 10 clinical patients. Coefficient of variation (CoV) was measured on a uniform background on the geometrical phantom. Recovery coefficients (RC) of spheres and mean absolute activity errors (MAE) of the kidneys, parotid glands, and tumors in XCAT phantoms were evaluated, while the activity difference was evaluated in clinical data before and after PVC.

For the cylindrical phantom, the central sphere suffered from a more severe partial volume effect (RC of 0.50) than one of the peripheral spheres on the edge of the phantom (RC of 0.63). Their RCs were improved to 1.02 and 1.22 after IY, 1.03 and 0.98 after svPSF-IY, 1.02 and 1.15 after RVC, 0.99 and 1.12 after NC-RVC, and 0.98 and 1.01 after svPSF+NC-RVC, respectively. The background CoV was 0.048 for non-PVC, 0.090 for RVC, 0.078 for NC-RVC, and 0.082 for svPSF+NC-RVC. The MAEs of kidneys, parotid glands, and tumors before PVC were 20.4% ± 1.1%, 51.2% ± 6.1%, and 51.8% ± 8.1% respectively, and were best reduced to 4.7% ± 2.4%, 6.2% ± 4.5%, and 8.8% ± 6.4% after svPSF-IY. The Gibbs ringing artifacts and noise in RVC were reduced when NC was implemented. The activity differences of clinical data showed a similar trend with those of XCAT simulations.

The proposed svPSF enhanced PVC methods improve quantification accuracy in ¹⁷⁷Lu-PSMA-617 SPECT. NC suppresses the Gibbs ringing artifacts and reduces noise in standard RVC images.