Motion-compensated fully 4D PET reconstruction using PET data supersets

Patient head movement not only inhibits the full potential of high spatial-resolution neurological PET imaging, but it also significantly degrades fully 4D reconstruction when using temporally-extensive basis functions. In this case a single movement in just one of the time frames propagates to impact the reconstruction of all other time frames. Here we propose a motion-compensation strategy through the use of PET data supersets and demonstrate its application to fully 4D reconstruction. The richly-sampled superset is populated by considering head motion as equivalent to shifts in scanner position. This requires the positioning of the list-mode events into the superset and the creation of time-dependent normalisation factors for the superset. An advantage of this approach is that the attenuation factors for the superset need only be computed once for the reference position. This approach adapts readily for use with existing fully 4D reconstruction methods with the only modification being the introduction of time-dependent normalisation factors. Using simulated as well as real high-resolution PET data from the HRRT, we demonstrate both the detrimental impact of head motion in fully 4D PET reconstruction as well as the efficacy of our proposed motion-compensation method. This is an important step towards realising the potential of fully 4D reconstruction methods for patient studies.