Speaker
Dr
Elisabetta De Bernardi
(Bioengineering Department, Polytechnic University of Milan, Milan, Italy)
Description
INTRODUCTION: Fourier Rebinning (FORE) [1] is the most widely used algorithm for
the rebinning of 3D PET data into a stack of 2D sinograms. FORE operates into the
sinogram Fourier Transform domain and it is based on the frequency-distance
relation. The validity of this approximation holds at high frequencies; low
angular, k, and radial, ω, frequencies are instead usually rebinned with Single
Slice Rebinning (SSRB) limited to low copolar angle data. A square low frequency
region is commonly defined on empirical basis and implies an abrupt transition of
rebinning strategy. In a previous work [2] we proposed a simple index able to map
the validity of the frequency-distance relation in the sinogram frequency-space. A
gradual rather than an abrupt transition mask was consequently proposed for an
optimized FORE. Preliminary results on scanned line and uniform phantoms showed the
superior performances of the proposed method with respect to the standard abrupt
transition: ring artifacts on reconstructed images were eliminated and a better
compromise between axial resolution and noise resulted achievable. An
anthropomorphic phantom allows to test the method over more complex structures in a
condition more similar to oncological studies.
METHODS: We compared the performances of standard abrupt and gradual partitions on
the Alderson thorax-abdomen phantom, scanned with ECAT EXACT HR+. The phantom
districts were filled with different activities simulating the in vivo 18F-FDG
uptake and spherical lesions of different radii (from 2.15mm up to 10.96mm) were
positioned inside. Spherical Volumes of Interest (VOIs) were defined on the spheres
(80% of the sphere radius) and on both thorax and liver (radius 16mm). Mean imaged
activity normalised to the total activity of the reconstructed study was extracted
within each VOI. Inside the liver and thorax VOIs the std/mean ratio was also
evaluated. Two successive acquisitions were
performed and the obtained results were averaged. Both abrupt and gradual
partitions were implemented with different parameters corresponding to a different
extension of the low frequency SSRB region: for the first approach square SSRB
regions were considered starting from (ωlim=0, klim=0) up to (ωlim=6Δω, klim=6);
for the second one, in setting the gradual transition border, several values T of
the validity index were tested starting from 1% up to 13% of the maximum validity
level.
RESULTS: When FORE was applied at very low frequencies, in a low validity range,
the axial spread amount leaded to an important partial volume effect. For example
for the 7.83mm sphere the imaged activity was 79% of the true activity for (ωlim=0,
klim=0) and 84% for T=1%. The inaccurate rebinning of the low frequencies also
caused an incorrect activity definition on large organs, i.e. 130% of the true
activity inside the liver VOI for (ωlim≤Δω, klim≤1) and T≤3%. When the transition
border was moved towards higher frequencies the axial spread was reduced: for
example the imaged activity on the 7.83mm sphere was ≥92% of the true activity for
(ωlim≥3Δω, klim≥3) and T≥3%. Although, a larger parameter choice implies a
count statistics reduction, limited effects were observed with the proposed gradual
partition; e.g. in the liver VOI the std/mean was ≤0.36 for all the explored
parameter T values. The abrupt method, on the contrary, displayed std/mean ≥0.41
for (ωlim≥3Δω, klim≥3).
CONCLUSION: Results previously observed on line and uniform phantoms relevant to
minimization of axial blurring and artefacts were confirmed and were not
jeopardized by the acquisition of more complex structures. We can conclude that the
proposed method, by eliminating ring artifacts and reducing low frequencies noise
component, allows a parameter choice corresponding to a better quantification
accuracy and can be applied in clinical conditions.
REFERENCES:
[1]: M. Defrise, P. E. Kinahan, D. W. Townsend, C. Michel, M. Sibomana and D.
Newport, “Exact and approximate rebinning algorithms for 3D PET data,” IEEE Trans.
Med. Imag., vol. 16, pp. 145–158, Apr. 1997.
[2]: E. De Bernardi, M. Mazzoli, F. Zito and G. Baselli, “Evaluation of frequency-
distance relation validity for FORE optimization in 3D PET,” submitted for
publications to IEEE Trans. Nucl. Sci.
Author
Dr
Elisabetta De Bernardi
(Bioengineering Department, Polytechnic University of Milan, Milan, Italy)
Co-authors
Dr
Felicia Zito
(Nuclear Medicine Department, Ospedale Maggiore Policlinico, Mangiagalli e Regina Elena)
Prof.
Giuseppe Baselli
(Bioengineering Department, Polytechnic University of Milan, Milan, Italy)
Dr
Marco Mazzoli
(Nuclear Medicine Department, Ospedale Maggiore Policlinico, Mangiagalli e Regina Elena)
