Comparison of 10 versus 20 min SPECT 99mTc-MDP bone scans: use of 3D-OSEM image reconstruction with distance-dependent resolution modelling

Iterative reconstruction with system response modelling has been implemented in commercial software by manufacturers for distance-dependent resolution modelling (DRM) of the collimator physical effects. Initial experience with such algorithms also shows improvements in noise characteristics with lower dependency on counting statistics. In this study the performance of one such algorithm, the Philips Astonish, was assessed for bone single-photon emission computed tomography (SPECT) acquired at count levels reduced by half on technetium-99m methylene diphosphonate scans.

For every SPECT scan, two sets of images were generated with the aid of concurrent data acquisition: (i) a conventional scan used routinely for reporting at 20 s per projection reconstructed with filtered back-projection (FBP20 s) and (ii) a scan at 10 s per projection reconstructed with Astonish (DRM10 s). Phantom and pilot patient data were used to initially establish optimal reconstruction parameters. Subsequently, patient studies (n=28) were scored independently by two experienced observers (blinded to reconstruction method or acquisition time) for image quality based on a scale of 1-5. Observers were also asked to report the number of observed lesions in each scan.

Results show that scores were better or equivalent for the vast majority of DRM10 s images compared with FBP20 s with statistically significant differences between the two methods (observer A: mean DRM10 s=4.3±0.5, mean FBP20 s=3.8±0.8, P=0.0064; observer B: mean DRM10 s=3.6±0.8, mean FBP20 s=3.1±0.9, P=0.0073). Improvements in image quality for DRM10 s were reported on 16 out of 28 scans for observer A and 15 out of 28 scans for observer B, whereas 8 out of 28 and 9 out of 28 scans received equivalent scores, respectively. The total number of lesions reported for both DRM10 s and FBP20 s was 72 for both observers showing no differences between the two methods.

These results indicate that the use of the DRM algorithm has the potential for reducing bone SPECT acquisition times by half without compromising current levels of image quality and diagnostic value, or reduce the injected dose when radioactivity supply is limited.