|
ABSTRACT
The
iterative image reconstruction (IIR) is a promising approach
to achieve a better
image quality in PET. However, limitations exist with respect
to the required computation time and the influence of reconstruction
parameters on quantitative PET data. We implemented different
reconstruction algorithms in a PC based reconstruction program
and evaluated the effect of the reconstruction algorithms
as well as reconstruction parameters on the quantitative PET
results.
The following IIR algorithms were implemented: maximum likelihood
expectation maximization (LMEM), weighted least squares (WLS),
image space reconstruction algorithm (ISRA), space alternating
generalized expectation maximization (SAGE). The ordered subsets
(OS) method and the median root prior (MRP) correction were
provided and can be used in combination with each reconstruction
algorithm. A dynamic PET study, showing small liver metastases,
was used for the evaluation of the properties of the reconstruction
parameters. Regions-of-Interest (ROI) were placed in a small
high uptake area as well as in a larger low uptake region
for quantification purpose using standardized uptake values
(SUV). The 128x128 image matrix was generally not suffient
to detect the metastases as separate lesions and a 256x256
matrix was required for the delineation of the lesions. Furthermore,
the use of the iterative attenuation correction improved the
image quality significantly. The lesion detectability deteriorated
when more than six iteration steps were used without applying
the median root prior correction. In contrast, the median
root prior correction improved the lesion detectability with
a higher number of iteration steps. The quantitative evaluation
of the hot lesion demonstrated a dependency of the uptake
values on the number of iterations for all reconstruction
methods. In contrast, the SUV of the low uptake area did not
show a major variation with the number of iteration steps.
Both convergence and noise reduction were improved when the
median root prior correction was applied. All reconstruction
algorithms showed an increase of the SUV and noise with higher
number of subsets. The increase of the median root prior correction
value (0.1 to 1.0) resulted in an decrease of the SUV in the
hot area. Regarding reconstruction speed, image quality, and
accuracy of quantitative data, best results were obtained
with OSEM and OSISRA. The image quality of OSSAGE was comparable,
but the reconstruction speed slower. OSWLS showed instable
results with higher number of iterations. Based on our results,
we prefer for routine PET studies the OSEM method, 8 iterations,
4 subsets, and median root prior correction with mrp=0.3.
key
words:
PET, iterative image reconstruction
|
