A comparison of data-driven methods for patient motion estimation in cardiac SPECT imaging

We investigate two data-driven strategies of patient motion estimation in cardiac SPECT imaging. Data-driven methods use the SPECT acquisition data itself to determine motion, with little or no information from external surrogates which track patient motion. In previous work [1, 2], we investigated a data-driven strategy based on 2D–3D registration using partial reconstruction of the heart and another using maximum-likelihood for simultaneous motion and activity estimation within SPECT reconstruction. In this paper, we compare these strategies using the XCAT phantom made from the MRI studies of volunteers. These XCAT phantoms are based on segmented MRI data by first contouring the body shape, followed by scaling the skeleton to match the volunteer's skeleton, and then shaping each organ to match the MRI data. This allows modeling of realistic patient motion by replicating the volunteer's external body motion and internal organ motion. Two datasets were thus obtained: (1) Axial slide and (2) Shoulder twist. Here we present preliminary results for these datasets.