Hybridrecon

All patient images were acquired on a Discovery NM/CT 670 SPECT/CT system (GE Healthcare, Cleveland, USA) with a parallel-hole medium-energy general purpose collimator. SPECT data were acquired with a 109.1–134.2 keV window for 120 views over ${360}^{\circ }$ with 30 s/view. Quantitative reconstructions were performed on HybridRecon (Version 1.3, Hermes Medical Solutions) and consisted of 3D ordered-subset expectation maximum (OSEM) reconstructions with attenuation correction based on a low-dose helical CT, scatter corrections based on a Monte Carlo convolution-based forced detection algorithm, and collimator detector response modeling [20 (link)]. All quantitative reconstructions were performed with an equivalent 75 iterations (15 subsets and 5 iterations) based on manufacturer recommendations [6 (link)]. No post-filter was applied and the reconstructed SPECT voxel sizes were isotropic at 4.42 mm.
When required, CT mass densities were calculated using a scanner-specific linear lookup table based on electron density phantom scans. The CT voxels were reconstructed with sizes 0.976 mm $\times$ 0.976 mm $\times$ 3.75 mm. All contours were drawn on the low-dose helical CT using tools from MIM Maestro v6.6 (MIM) and followed established contouring guidelines [21 (link)]. Any overlapping contours were corrected with Boolean operations. The contours were verified by an experienced nuclear medicine physician.

HybridRecon (Version 1.1C, HERMES Medical Solutions AB) was used to perform all reconstructions whilst HybridViewer (Version 2.6F, HERMES Medical Solutions AB) was used to view and analyse the reconstructed images. All images were attenuation corrected using CT data and had the resolution recovery option switched on. Where full MC collimator modelling was not enabled, standard MC scatter correction was carried out.
One acquisition of the phantom was reconstructed with OSEM 1–7 iterations and 15 subsets in order to ascertain the optimum OSEM reconstruction. Previous work on optimisation of ^99mTc reconstruction in SPECT imaging had shown that five iterations and 16 subsets yield the best compromise between resolution and noise [20 (link)] but owing to the number of projections at which data were acquired at this institution, 16 subsets were not a possible option. Instead, the nearest alternative of 15 subsets was used. The optimum number of OSEM iterations for ⁹⁰Y was found via quantitative analysis as described below.
Once the optimised OSEM reconstruction had been identified, all three of the acquisitions were reconstructed with the GE default OSEM protocol (2 iterations and 10 subsets) and the optimised OSEM protocol, both with and without full MC collimator modelling.