Analysis of the parameterization of FORE-WLS reconstruction in 3D whole-body PET imaging

The FORE-WLS (Fourier rebinning followed by a weighted least-squares iterative reconstruction) approach to 3D whole-body PET reconstruction has been investigated using a design-of-experiments (DOE) approach. Measures of image quality were analyzed at different algorithm parameterizations including lung mean value and variance, liver mean value and variance and a pseudomeasurement of resolution. One clinical 3-axial field-of-view (AFOV) study was analyzed with the DOE method and a set of parameter responses obtained. A subset of the parameters was then used in subsequent visual image quality investigations. Results: The DOE results show expected increase in resolution with number of iterations and degradation of resolution with increased post-filtering (the two main effects) but also changing the number of subsets from 20 to 40 showed a visual improvement. Lung variance degraded with number of iterations and improved with heavier post-filtering, but was somewhat less dependent on number of subsets. Liver variance showed a trend identical to lung variance as a function of parameterization, but the absolute parameter response range was over a factor of 2 for liver variance as compared to approximately a 40% range for lung variance. Mean value response for lung changed maximally by 13% while for liver it changed maximally by 31% as a function of parameterization. No significant parameter interactions were found. From subsequent visual analysis it was determined that a reasonable set of default parameters for the FORE-WLS algorithm was 3 iterations, 32 subsets, 4.30 mm loop filter and 6.0 mm post-filter. Conclusions: Parameterization sensitivity of the FORE-WLS algorithm has been established. A suggested set of FORE-WLS parameters chosen from the range investigates has been made. Visual analysis of a set of clinical patient datasets has been done covering a range of 3D whole-body PET subjects and the default parameter choice appeared has been shown to be robust.