Christoph Kolbitsch1,2, Radhouene Neji3, Matthias Fenchel4, and Tobias Schaeffter1,2
1Division of Imaging Sciences and Biomedical Engineering, King's College London, London, United Kingdom, 2Physikalisch-Technische Bundesanstalt (PTB), Braunschweig and Berlin, Germany, 3MR Research Collaborations, Siemens Healthcare, Frimley, United Kingdom, 4MR Oncology Application Development, Siemens Healthcare, Erlangen, Germany
Synopsis
Quantitative PET requires accurate attenuation
correction (AC) information. For simultaneous PET-MR acquisitions in the thorax
or abdomen these MRAC images are obtained in a single breathhold which can lead
to misregistration errors between breathhold MRAC and free-breathing PET data.
Here we present a method which obtains accurate AC information (Dice
coefficient higher than 0.85) during free-breathing and yields additional
respiratory motion fields which can be utilised in motion-compensated MR and
PET reconstructions. The proposed Dixon-GRPE method led to improvements of up
to 50% in sharpness (FWHM) and a 33% improvement in the quantification of the
specific uptake value (SUV).Introduction
Simultaneous PET-MR yields highly sensitive PET and
versatile MR data in a single examination. In order to obtain quantitative PET
information, the varying density of different tissue types has to be taken into
consideration using attenuation correction (AC) data.
Commonly, this AC map is obtained in a short two-point
Dixon MR scan before every PET acquisition during a single breathhold
1. However, the
following PET scans are obtained during free-breathing which can lead to misregistration
errors between breathhold AC map and free-breathing PET data causing artefacts
at tissue boundaries (Fig. 1a).
Here we present a technique which obtains MR data
during free-breathing using a multi-echo acquisition with a 3D Golden radial
phase encoding (GRPE) sampling scheme
2. This approach
yields AC information and respiratory motion information allowing for the
transformation of AC information to the required respiratory motion phases. In
addition, motion information can be utilised in motion compensated image
reconstructions to improve MR and PET image qualities
2 (Fig. 1b).
Methods
Dixon-GRPE: GRPE was
implemented as a prototype sequence for multi-echo acquisitions on a 3T PET-MR scanner (Biograph mMR,
Siemens Healthcare, Erlangen, Germany) and three-point Dixon data were obtained
in a phantom and four volunteers during free-breathing: T1-weighted GRE (FA 10o,
TR/TE 5.94/1.23/2.72/4.21ms FOV: 400mm3, 1.92x3.2x3.2mm3 resolution,
total acquisition time 96s). Standard MRAC data were also acquired (FA 10o,
TR/TE 3.60/1.23/2.46ms FOV: 500x328x400mm3, 2.6x4.1x3.1mm3
resolution) in a single 19s breathhold.
Respiratory motion estimation: A
self-gating signal from 1D foot-head projections is used to bin the MR data
into respiratory bins. 3D images defining different respiratory phases are
reconstructed with an iterative non-Cartesian SENSE reconstruction3 with a generalised
total-variation4 constraint in
spatial and temporal direction. Non-rigid image registration was performed to
derive motion fields describing the transformation between end-expiration and
all other respiratory phases5.
Motion compensated MR image reconstruction: The
final 3D high-resolution images are reconstructed using a motion compensated
image reconstruction which utilises the non-rigid respiratory motion fields to
transform the image data to the same respiratory phase at each iteration6.
Calculation of AC information: The motion-compensated
MR images are separated into fat and water components7. The images are then
classified into the following tissue types: air outside the body (Aout),
air inside the body (Ain), fat (F) and water (W).
Evaluation of AC images: The AC
information was validated by calculating the Dice similarity coefficient
between Dixon-GRPE and standard MRAC images8. In the phantom
images this was carried out for different undersampling factors ranging from 2x
oversampled (R=0.5) to 16x undersampled (R=16) in the angular direction.
PET simulations: Respiratory
resolved PET simulations were carried out based on the motion resolved GRPE
images using STIR9. Realistic
emission values were assigned to tissue classes and tumor lesions were added in
the liver10. PET images were
reconstructed using either the Cartesian MRAC data without motion correction
(uncorr) or with the Dixon-GRPE MRAC information and a respiratory motion
compensated image reconstruction (MCIR10). In the simulated PET images full-width-at-half-maximum (FWHM) and maximum uptake
(SUVmax) were assessed and compared to motion-free reference images.
Results
Dice coefficients calculated
from the phantom scans were larger than 0.85 for a wide range of undersampling
factors suggesting a robust fat-water separation and AC classification (Fig. 2).
Based on these phantom experiments the number of respiratory bins was set to 8,
i.e. respiratory resolved images with a maximum undersampling factor of 8. A comparison
between fat, water and AC images for volunteers shows also very good
agreement (Fig 3). In addition, motion vectors from expiration to inspiration
are displayed. The Dice coefficients for A
out, A
in and W
were 0.99±0.001, 0.85±0.031 and 0.86±0.017, respectively. The Dice coefficient for fat in healthy volunteers was 0.51±0.115, due to the small size of fat structures in healthy volunteers which
led to strong reductions of the Dice coefficient even for small differences in
the classification. Respiratory motion and misregistration errors between MRAC
and PET data can lead to strong artefacts in the PET images (Fig. 4). The error
of FWHM is between 10% and 50% and the error of SUV
max is up to 33%.
Conclusion
We have presented an
MR technique which yields accurate 3D AC images and respiratory motion
information within one short free-breathing scan. The motion information can be
utilised in a motion-compensated PET reconstruction to correct for respiratory
motion artefacts and to ensure there are no misregistration errors between AC
and PET image data. Evaluation of Dixon-RPE in clinical PET-MR scans will be
carried out to validate the PET simulation results.
Acknowledgements
No acknowledgement found.References
1. Martinez-Möller A,
Souvatzoglou M, Delso G, et al. Tissue classification as a potential approach
for attenuation correction in whole-body PET/MRI: evaluation with PET/CT data. J
Nucl Med. 2009;50(4):520-526.
2. Buerger C, Clough RE, King AP, Schaeffter T,
Prieto C. Nonrigid motion modeling of the liver from 3-D undersampled
self-gated golden-radial phase encoded MRI. IEEE Trans Med Imaging.
2012;31(3):805-815.
3. Pruessmann KP, Weiger M, Boernert P, Boesiger
P. Advances in sensitivity encoding with arbitrary k-space trajectories. Magn
Reson Imaging. 2001;46(4):638-651.
4. Knoll F, Bredies K, Pock T, Stollberger R.
Second order total generalized variation (TGV) for MRI. Magn Reson Med.
2011;65(2):480-491.
5. Rueckert D, Sonoda LI, Hayes C, Hill DL,
Leach MO, Hawkes DJ. Nonrigid registration using free-form deformations:
application to breast MR images. IEEE Trans Med Imaging.
1999;18(8):712-721.
6. Batchelor PG, Atkinson D, Irarrazaval P, Hill
DLG, Hajnal J, Larkman D. Matrix description of general motion correction
applied to multishot images. Magn Reson Med. 2005;54(5):1273-1280.
7. Berglund J, Kullberg J. Three-dimensional
water/fat separation and T 2 * estimation based on whole-image
optimization-Application in breathhold liver imaging at 1.5 T. Magn Reson
Med. 2012;67:1684-1693.
8. Dice LR. Measures of the Amount of Ecologic
Association Between Species. Ecology. 1945;26(3):297-302.
9. Thielemans K, Tsoumpas C, Mustafovic S, et
al. STIR: software for tomographic image reconstruction release 2. Phys Med
Biol. 2012;57(4):867-883.
10. Tsoumpas C, Buerger C, King a P, et al. Fast generation of 4D PET-MR data
from real dynamic MR acquisitions. Phys Med Biol. 2011;56(20):6597-6613.