View sampling requirements for cardiac SPECT

The new generation of dedicated cardiac SPECT systems achieve high detection efficiency by focusing onto a small field of view (FOV) that encloses a patient´s heart. They also generally use a relatively small number of views for angular sampling. The objective of this study is to use simulations to investigate the relationship between number of views and image quality when iterative reconstruction is used for cardiac SPECT. In our simulations, sinograms, with number of views from 6 to 60 evenly spaced over an appropriate angular range, were generated from a digital heart phantom. A depth-dependent Gaussian PSF was modeled for a SPECT system with an inverse-fanbeam geometry. We varied the system resolution at the center of the FOV from 9 mm to 18 mm FWHM in 3 mm steps. We performed reconstructions using conventional FBP, and OS-EM with and without PSF modeling. We compared the reconstructed images with different number of views and different system resolutions qualitatively and quantitatively. Normalized root mean squared error (NRMSE) and recovery coefficient (RC) were calculated as indices of image quality. Noise performance was also analyzed as a function of number of iteration. Our results indicate that, NRMSE decreases and RC increases initially as the number of views increases then flatten out around 18-24 views. For a 9-15 mm system resolution, 15-24 views are optimal for a 16 cm FOV since further increase of the number of views only improves image quality marginally. Similar to FBP, OS-EM shows the same trend that the number of views required is inversely proportional to the system resolution. Under the same system resolution and number of views, OS-EM achieves better image quality than FBP does, especially when PSF is modeled in the reconstruction. The simulations demonstrate that modern cardiac SPECT using iterative reconstruction with PSF modeling allows the use of reduced number of views compared to conventional SPECT using FBP.