Masking for DDG in SPECT Reduces Influence of Motion Artifacts

Respiratory motion in Single Photon Emission Computed Tomography (SPECT) is typically corrected using a 1D respiratory estimate to correct for the most dominant head-to-foot motion in the data. Estimation of this respiratory surrogate signal is typically done based on all of the data measured on the detector, but there might not only be one correct 1D respiratory motion contained for a given projection dataset. Using a large set of simulated water cylinder phantoms with moving spheres, we probe the influence of different kinds of motion artifacts on Principal Component Analysis (PCA), Center of Light (COL), Isometric Feature Mapping (ISOMAP), and Laplacian Eigenmaps (LE) for Data Driven Gating (DDG) when used in combination with projection domain masking. We demonstrate that using masking for DDG improves accuracy for all methods when a motion artifact is present in the data. Overall the LE methods performed the most robust across all simulations.