Motion correction for gated PET using sinogram registration

The aim of this study was to investigate the feasibility of a motion correction method for dynamic positron emission tomography (PET) using sinogram registration. The NCAT phantom was simulated with respiratory motion and either 10 nun or 20 nun lesions located at the upper, middle, and base of the right lung. Additionally, a Jaszczak phantom filled with F-18 was placed on a motion platform. The NCAT data were binned into 16 gates, and the Jaszczak data were binned into 12 gates. In both studies lesions had a source to background ratio of either 5:1 or 10:1, and data were reconstructed with OSEM, 4 iterations and 8 subsets. Three methods were used to obtain the transformations between each frame: non-rigid image registration, rigid image registration, and sinogram registration. The sinogram method was derived from forward projecting only the lesion from image space to projection space. The transformations were applied to the gated sinograms to produce motion corrected sinograms, which were summed prior to reconstruction and resulted in a motion corrected image volume. The mean lesion signal, lesion contrast to noise ratio (CNR), and registration accuracy were evaluated. Compared to the other motion correction methods, sinogram registration resulted in comparable or improved lesion signal recovery, lesion CNR recovery, and registration accuracy. With respect to the NCAT data, sinogram registration resulted in a 5%-23% and 5%-10% higher CNR recovery compared to the non-rigid and rigid registration, respectively. With respect to the Jaszczak data, sinogram registration resulted in a 10%-14% higher mean signal recovery of the 1 mL lesion compared to rigid registration, and a 5-25% higher CNR recovery compared to the rigid motion correction method. As lesion size and contrast decreased, the mean signal recovery and CNR recovery achieved with the motion correction methods increased.