Quantification in SPECT: myth or reality? A multi-centric study

Reliable quantification from Single Photon Emission Computed Tomography (SPECT) images could facilitate patient classification and follow-up. The purpose of this study was to assess the current performance of French nuclear medicine departments (NMD) and research centers (RC) in estimating parameters from SPECT images. A Tc99m SPECT brain study of the dopaminergic system was simulated using the Monte Carlo code SimSET. Projections corresponding to different energy windows, attenuation maps, anatomical images, characteristics of the detector response function, and parameters of the simulated acquisitions were made available on a Web site. The quantification task was to estimate four binding potential (BP) values measured on the left and right putamen and caudate nuclei. Four NMD and two RC participated in the study. Although they were equipped with recent workstations, NMD had limited access to sophisticated quantification tools, unlike RC, which could correct for scatter, attenuation, detector response function and even partial volume effect. Reported BP varied by factors greater than 3, due in part to differences in the corrections applied. Even for similar corrections, BP could differ by more than 50% between centres. Sophisticated processing performed by the RC yielded less biased and more reproducible BP estimates than processing available to NMD. In conclusion, this study suggests that large efforts are still required to make sophisticated processing tools available on commercial workstations equipping NMD, without which meta-analysis of values collected at different centers is impossible.