Fully adaptive temporal regression smoothing in gated cardiac SPECT image reconstruction

A data-driven weighted polynomial regression estimator with variable bandwidth and adaptation of the order of the polynomial can enhance the diagnosis of both ventricular function and myocardial perfusion in gated cardiac SPECT. We propose a temporal approach which estimates a mean regression function of either time frame projections or reconstructed images. We investigated several implementations of the regression estimation procedure: a) prior to the reconstruction, applied on the projection data, b) iteratively, as a semi-parametric reconstruction method, and c) applied as a post-reconstruction regression smoother. Implementation of the regression method in step a) and c) is independent from the algorithm used to reconstruct the projection data; b) requires an iterative image reconstruction algorithm. Monte Carlo simulated projection data as well as clinical data have been processed to evaluate the technique. Unique regression smoothing of projection data, a), or reconstructed images, c), improves markedly the image statistics while preserving spatial and temporal resolution. Iterative application, b), without the use of a relaxation parameter may result in over-smoothed time activity curves.