The Nuclear Medicine Software Quality Group of the Institute of Physics and Engineering in Medicine has conducted an audit to compare the ways in which different manufacturers implement the filters used in single-photon emission computed tomography. The aim of the audit was to identify differences between manufacturers’ implementations of the same filter and to find means for converting parameters between systems.
Computer-generated data representing projection images of an ideal test object were processed using seven different commercial nuclear medicine systems. Images were reconstructed using filtered back projection and a Butterworth filter with three different cutoff frequencies and three different orders.
The audit found large variations between the frequency–response curves of what were ostensibly the same filters on different systems. The differences were greater than could be explained simply by different Butterworth formulae. Measured cutoff frequencies varied between 40 and 180% of that expected. There was also occasional confusion with respect to frequency units.
The audit concluded that the practical implementation of filtering, such as the size of the kernel, has a profound effect on the results, producing large differences between systems. Nevertheless, this work shows how users can quantify the frequency response of their own systems so that it will be possible to compare two systems in order to find filter parameters on each that produce equivalent results. These findings will also make it easier for users to replicate filters similar to other published results, even if they are using a different computer system.