Quantification of reconstructed SPECT non-uniformity

A simple method is presented for the quantification of reconstructed image non-uniformity which is both reproducible and sensitive to changes in system response. Each reconstructed transaxial slice is converted into a polar image by stacking together radial profiles measured from the image centre at 15 degree intervals. Calculating the intensity variations of each polar image in the horizontal and vertical directions then generates measures of radial and tangential uniformity, respectively. Axial uniformity is assessed by stacking together the average radial profiles for each slice and monitoring the variation in intensity of each one from the average profile for the whole phantom. The reproducibility and sensitivity of the indices were assessed using synthetic projection data generated at different count densities and with varying levels of planar non-uniformity. There is little improvement in either the reproducibility or the sensitivity of the indices for count densities in excess of 1 million counts per slice provided that the data have been prefiltered using an appropriate filter. The indices are capable of detecting a range of artefacts relevant to clinical imaging and can be used as part of a quality control programme or to compare the performance of different single photon emission computed tomography (SPECT) systems.