We report the extended application of an automated computer technique
for three-dimensional spatial registration of SPECT and PET studies. M
ethods: The technique iteratively reslices a misaligned data set until
the sum of the absolute differences (SAD) from a reference data set i
s minimized. The registration accuracy was assessed in Hoffman brain p
hantom studies collected with known misalignments and transmission stu
dies of a thorax phantom with fiducial markers. The SAD was compared w
ith three other cost functions: stochastic sign change criterion, sum
of products and standard deviation (s.d.) of ratios. In clinical neuro
logical and myocardial perfusion studies, registration accuracy was es
timated from the relative locations of landmarks in the reference and
registered data sets. Results: Registration accuracy in the Hoffman br
ain phantom studies was -0.07 +/- 0.46 mm (mean +/- s.d.) for translat
ions and -0.01 +/- 0.20 degrees for rotations, with maximum translatio
n and rotation errors of 1.2 mm and 0.8 degree, respectively. The SAD
was the most accurate and reliable cost function. Registration errors
in the thorax phantom were 3.1 +/- 1.7 mm. Mean accuracy in the neurol
ogical studies, estimated from landmark pairs, was 2.0 +/- 1.1 mm for
SPECT to SPECT and 1.8 +/- 1.1 mm for PET to SPECT registrations. Aver
age registration accuracy in (TI)-T-201 myocardial perfusion studies w
as 2.1 +/- 1.2 mm. Conclusion: Our registration method (a) provided ac
curate registrations for phantom and clinical SPECT and PET studies, (
b) is fully automated, (c) simplifies comparison of data sets obtained
at different times and with different modalities, and (d) can be appl
ied retrospectively.