Automated stereotactic standardization of brain SPECT receptor data using single-photon transmission images.

UNLABELLED Intra- or intersubject registration of anatomically poorly defined SPECT data, such as in neuroreceptor imaging, is important for longitudinal or group analysis. However, accurate registration is difficult with only emission CT (ECT) data. We investigated fully automated registration using transmission CT (TCT) data as an intermediary image set. METHODS The accuracy of TCT registration was compared to that of ECT registration for four types of data: gray-matter distribution (with [99mTc]ethylcysteinate dimer (ECD)), neocortical distribution (with [123I]R91150, a highly specific 5-HT2a receptor ligand), and striatal distribution of the D2-receptor ligand (with [123I]iodobenzamide (IBZM)) and the dopamine transporter ligand (with [123I]2beta-carbomethoxy-3beta-(4-fluorophenyl)tropane (CIT)). In total, 10 datasets of the various study types were used, all collected on a Toshiba GCA9300 gamma camera with super-high-resolution fanbeam collimators and 3 x 370 MBq of 153Gd transmission sources (4-min sequential TCT scanning for receptor studies and 20-min simultaneous scanning for [99mTc]ECD studies). Per dataset, 15 random misalignments of 9 rigid-body parameters (translation, rotation, and anisotropic scaling) were conducted. All coregistrations were done twice, both to the subject's original scan and to a study-specific template. This was done manually by two independent experienced observers and with three automated voxel similarity algorithms: mutual information (M.I.), count difference (C.D.), and uniformity index (U.I.). As an outcome measure, the impact of misregistration on semiquantification for the various study types was established. RESULTS TCT matching allowed registration within 3.3 mm, 2.4 degrees, and 1.2% scaling (mean squared values for all directions) with an overall accuracy decrease in the following order: C.D. > M.I. > manual > U.I. For [99mTc]ECD and [123I]IBZM, TCT registration was as accurate as ECT registration, while it was far superior for the other receptor data types, especially for abnormal studies. The automated TCT registration accuracy corresponded to average quantification errors of 2.9% ([99mTc]ECD), 4.2% ([123I]BZM), 5.7% ([123I]R91150), and 6.1% ([123I]beta-CIT). CONCLUSION Fully automated registration through intermediary TCT images is clinically feasible, fast, and accurate. In addition to nonuniform attenuation correction, TCT scanning therefore allows coregistration for group comparisons of SPECT receptor data on a standardized or pixel-by-pixel basis.