Quantitative simultaneous positron emission tomography and magnetic resonance imaging Jinsong Ouyang Yoann Petibon Chuan Huang Timothy G. Reese Aleksandra L. Kolnick Georges El Fakhri
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Journal of Medical Imaging 1(3), 033502 (Oct–Dec 2014)
Quantitative simultaneous positron emission tomography and magnetic resonance imaging Jinsong Ouyang,a,b,*,† Yoann Petibon,a,c,† Chuan Huang,a,b Timothy G. Reese,b,d Aleksandra L. Kolnick,a and Georges El Fakhria,b
a Massachusetts General Hospital, Center for Advanced Medical Imaging Sciences, Division of Nuclear Medicine and Molecular Imaging, Department of Radiology, Boston, Massachusetts 02114, United States b Harvard Medical School, Department of Radiology, Boston, Massachusetts 02114, United States c Sorbonne Universités, UPMC Univ Paris 06, Paris 75634, France d Massachusetts General Hospital, Martinos Center for Biomedical Imaging, Department of Radiology, Charlestown, Massachusetts 02129, United States
Abstract. Simultaneous positron emission tomography and magnetic resonance imaging (PET-MR) is an innovative and promising imaging modality that is generating substantial interest in the medical imaging community, while offering many challenges and opportunities. In this study, we investigated whether MR surface coils need to be accounted for in PET attenuation correction. Furthermore, we integrated motion correction, attenuation correction, and point spread function modeling into a single PET reconstruction framework. We applied our reconstruction framework to in vivo animal and patient PET-MR studies. We have demonstrated that our approach greatly improved PET image quality. © 2014 Society of Photo-Optical Instrumentation Engineers (SPIE) [DOI: 10.1117/1 .JMI.1.3.033502]
Keywords: positron emission tomography and magnetic resonance imaging; positron emission tomography attenuation correction; surface coils; point spread function modeling; motion correction. Paper 14054PR received Apr. 22, 2014; revised manuscript received Aug. 19, 2014; accepted for publication Oct. 2, 2014; published online Nov. 3, 2014.
1
Introduction
Accurate attenuation correction (AC) is critical for quantitative whole-body positron emission tomography and magnetic resonance (PET-MR) imaging. The AC approaches for PET-MR can be grouped into three different categories: segmentation based,1–15 atlas registration based,3,16,17 and simultaneous maximum-likelihood reconstruction of attenuation and activity.18–23 Segmentation-based approach, which is to assign a global attenuation coefficient to an identified tissue class, appears to be the only one that is feasible in clinical practice for whole-body PET-MR. However, one of the major problems associated with such an approach is the large bias on PET quantification caused by the inter- and intra-patient electron density variations, especially in lungs and bones.15 This is why MRbased AC still remains a challenge for PET-MR. It may decrease electron density variation to reduce the bias on PET quantitation if patient studies are grouped according to their bed positions, patients’ ages, genders, and body mass indices and groupdependent attenuation coefficients are assigned to segmented tissue classes. Siemens whole-body PET-MR (mMR) surface coils, which can be placed above the patient during a PET-MR scan, are currently not accounted for in PET AC because the location and shape of the coils are unknown during the scan. Furst et al.24 acquired data of six patients, who had a total of eight lesions, with and without surface coils attached. The standardized uptake value difference for the eight lesions varied from −18% to 60%.
*Address all correspondence to: Jinsong Ouyang, E-mail: ouyang.jinsong@ mgh.harvard.edu †
These authors contributed equally.
Journal of Medical Imaging
The partial volume effects (PVEs), resulting from the scanner’s limited point spread function (PSF), are also a major source of image spatial resolution degradation. PVEs cause the signal to spread from a target tissue into neighboring tissues (“spill-out”) and from those neighboring tissues into the target tissue (“spill-in”). PVEs also reduce signal-to-noise ratio (SNR) and introduce a bias in measured activity concentration which significantly hampers the accuracy of the quantitation of the PET images. For lesions smaller than the reconstructed spatial resolution, PVEs may lead to more than 50% underestimation of the true radiotracer concentration.25,26 The myocardial wall, whose thickness is close to the PET resolution, particularly suffers from PVEs. PVEs can be greatly alleviated by modeling the PSF in the PET system matrix during reconstruction. Patient motion is another main contributing factor to image degradation in PET. The intrinsic spatial resolution of modern whole-body PET scanners is in the range of 4 to 5 mm. However, due to inevitable involuntary physiological patient motion, this resolution cannot currently be achieved in either abdominal or cardiac PET studies. Both respiratory and cardiac motions are the main causes of image degradation in thoracic and abdominal PET studies. Furthermore, it is important to highlight that it is not just the emission data that are corrupted by motion; there are often large reconstruction errors due to the deformation of the attenuating medium. In order to make proper use of emission data from all motion phases, one must know the corresponding attenuation at all times. There have been many attempts to develop practical and effective methodologies to remove the effects of cardiac/respiratory motion.27–37 Cardiac and/or respiratory gating strategies that “freeze” motion are 0091-3286/2014/$25.00 © 2014 SPIE
033502-1
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
Ouyang et al.: Quantitative simultaneous positron emission tomography and magnetic resonance imaging
popular in abdominal and cardiac PET studies. However, gating is not an optimal solution to the motion problem as it dramatically hampers the images’ SNR, and requires longer studies to achieve good image quality. Previously, we have studied bias atlases for segmentationbased PET attenuation correction using PET-CT and MR.15 We have also developed robust and accurate MRI methods allowing the tracking and measurement of both respiratory and cardiac motions during abdominal or cardiac studies. The MR-measured motion fields as well as PSF modeling are integrated into a PET reconstruction framework.38–40 In this PET reconstruction framework, a “time-dependent” attenuation map is deformed by the motion fields and is consistent with the PET emission distribution. In this study, we present new results on attenuation correction on MR surface coils. We also summarize our PET reconstruction framework, which incorporates AC, PSF modeling, and motion correction, and show its performance for in vivo animal and patient PET-MR studies.
2 2.1
Materials and Methods Attenuation Correction of MR Surface Coils
A bias image was then calculated using the image reconstructed with the phantom-coil-μ-map as the reference.
2.2
We measured and modeled the radially varying PSF of the mMR scanner in the sinogram domain. For a given axial position, we simultaneously placed 8 ∼ 0.5-mm 18 F point sources at radial positions with 2.5 cm spacing across the PET-MR scanner field of view. We acquired ∼20-million coincidences for each of the four different axial positions. For each point source, axial and radial sinogram profiles through the point source peak value were extracted. The obtained profiles were fitted using radially varying asymmetric exponential kernels to account for detector blurring effects along the radial and axial directions. We have previously validated this model, i.e., radially varying asymmetric exponential kernels, in phantom studies.40 In this study, we applied this model to in vivo imaging to evaluate its performance.
2.3
Because mMR surface coils include some metal components, we decided to acquire the μ-map of the surface coils using a transmission scan to avoid CT metal artifacts. A transmission scan of an anthropomorphic phantom (Data Spectrum Corporation, Hillsborough, North Carolina) with a Siemens mMR surface coil array attached [See Fig. 1(a)] was first acquired on a Siemens ECAT HR+ scanner. The same phantom was filled with F-18 in both the liver and soft-tissue compartments (liver/background concentration ratio: 2.4). Additionally, three 1-cm “tumors” were placed around the liver (tumor/liver concentration ratio: 4). The phantom was placed in a Siemens Biograph PET-CT scanner. First, a CT with 120 kVp was acquired. Second, an mMR surface coil array was placed on the top of the phantom in the same way we did for the transmission scan. Third, a PET acquisition which included total 279-million coincidence events was performed. The CT image was first transformed into a μ-map, denoted phantomμ-map, using Siemens e7 tools. The coil image obtained from the transmission scan was digitally added to the phantom-μ-map to create another μ-map, denoted phantomcoil-μ-map. The PET data were reconstructed twice: once with phantom-μ-map and the other with phantom-coil-μ-map.
Point Spread Function Modeling for PET
Motion Measurement Using MR
We have developed a set of MRI pulse sequences allowing the measurements of respiratory and cardiac motions for lower abdomen and heart PET-MR imaging studies. For imaging the lower abdomen, we have developed a dedicated MRI pulse sequence (denoted “NAV-TrueFISP”) allowing the accurate measurement of respiratory motion in the lower abdomen. The NAV-TrueFISP sequence is composed of twodimensional (2-D) multislice steady-state free precession MRI acquisitions (“TrueFISP”) interleaved by pencil-beam navigator echoes.40 The acquired TrueFISP slices are chosen in sagittal orientation to keep the main components of the motion within the acquisition plane. Pencil-beam navigators are acquired prior to each TrueFISP acquisition and are used to track the lung-liver interface during the acquisition with high temporal sampling frequency (2 to 4 Hz). The obtained internal motion surrogates allow accurate monitoring of the respiratory cycle (binning of both PET and MRI data into respiratory phases) and handling of respiratory cycle irregularities. This sequence takes advantage of the specific TrueFISP contrast to provide anatomical landmarks in homogeneous organs, such as the liver [see blood vessels with bright signal as illustrated in Fig. 6(a)]. Typical acquisition times for the NAV-TrueFISP protocol range between
Fig. 1 Phantom study using Siemens mMR surface coils. (a) An mMR surface coil array was placed on the top of a data spectrum anthropomorphic torso phantom, (b) positron emission tomography (PET) bias image, (c) PET bias profiles.
Journal of Medical Imaging
033502-2
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
Ouyang et al.: Quantitative simultaneous positron emission tomography and magnetic resonance imaging
2.5 and 3.5 min, depending on the number of slices needed to cover the area of interest. We have also developed MRI sequences for cardiac imaging. Motion of the heart has two different components: motion caused by the heart beating (cardiac motion) and motion caused by respiration. Because of their different mechanical and temporal properties, cardiac and respiratory motions of the heart are measured separately using specific MRI pulse sequences. The obtained cardiac and respiratory motion fields are then combined to compute the motion field between any cardiac/respiratory phase and a reference one in which the PET data will be reconstructed. Cardiac motion is measured using a tagged-MRI pulse sequence (denoted as “GRE-NAV-Tag”) developed specifically for this motion. Special RF/gradients series are applied, introducing periodic magnetization modulation (known as “tags”) on the myocardium. Tagging occurs at every heartbeat, triggered by the R-wave. Next, a series of dynamic GRE (Gradient Recalled Echo) MR volumes are obtained in which the tags are visible. Tag lines’ deformations visible in the images allow characterizing complex motion components, such as shearing, torsion, or radial thickening. The sequence is also navigated by a pencil-beam navigator, allowing the acquisition of tagged MR data only at end-exhalation respiratory phase. Electrocardiogram signal and navigators are simultaneously recorded to assign a cardiac and respiratory phase to each acquired PET event. Three orthogonal directions of tagging are acquired to fully characterize cardiac motion. The nonaccelerated tagging acquisition takes about 10 min using our method (∼3.5 min for each tagging direction). The acquisition can be accelerated using partial k-space sampling, parallel imaging, compressed sensing, or a combination of these techniques. Moreover, a dedicated golden-angle radial-FLASH sequence developed in our group is used to measure the respiratory motion of the heart. This sequence measures the respiratory motion of the heart in less than 2 min.41 As previously mentioned, the motion field between any cardiac/respiratory phase and the reference phase is computed by nonrigid summing of the measured cardiac and respiratory motions. For both lower abdomen and cardiac motion measurements, each subject is instructed to breath as regularly as possible during the motion tracking MR acquisitions (i.e., NAV-TrueFISP for abdomen and GRE-NAV-Tag + radial-FLASH for heart). Based on the navigator amplitude, extreme, “abnormal” events (e.g., deep breath), not representative of a normal respiratory cycle, are discarded.
2.4
PET Reconstruction
We integrated MR-based AC, PSF modeling, and MR-based motion correction into a unified iterative reconstruction framework.38–40 In PET, the estimated sinograms, y, are
y ¼ Pf þ s þ r; where f is the PET image volume to be estimated, s and r are the estimated scattered and random contributions, and P is the system matrix. The system matrix includes an MR-based warping operator registering a given motion frame to a selected reference frame, a forward-projection operator, time-dependent attenuation factors, PSF blurring effects in the sinogram space, and detector normalization factors. For dynamic reconstruction, list-mode data are divided into multiple frames according to Journal of Medical Imaging
the requirements for kinetic analysis. Motion correction is applied frame by frame using a global reference motion frame. The system matrix can be defined as
P ¼ NBAGM; where M is the MR-based warping operator that registers a given motion phase to the reference motion phase, G is the forward-projection operator, A contains the AC factors for each frame, B models the 2-D PSF blurring effects in the sinogram space, and N represents the detector normalization factors. In this study, we applied this unified reconstruction framework to in vivo animal and patient studies.
2.5
Animal and Patient Studies
PSF modeling was evaluated using a patient study performed on the mMR scanner at Massachusetts General Hospital (MGH). A 72-year-old male subject (subject 1) was enrolled to participate in the study. The subject was administrated with 370 MBq18 F-FDG. 18 F-FDG-PET list-mode data were collected during 5 min, resulting in a total of 320-million PET counts. The acquired sinogram data were reconstructed using ordinary poisson OSEM (OP-OSEM) and OP-OSEM with PSF modeling (PSF-OP-OSEM). Simultaneous PET/MR scans were performed in another subject (subject 2) with known liver lesions. A PET listmode was collected for 210 s following a 15 mCi FDG injection. Twenty-five sagittal slices covering the whole liver were simultaneously acquired using a 2-D multislice steady-state free precession TrueFISP sequence (TE ¼ 1.67 ms, TR ¼ 320 ms, 2 × 2 × 8 mm voxel size) with each slice acquisition repeated 30 times. MR slices and PET events were rebinned into seven respiratory phases using the diaphragm position. Motion fields obtained from B-spline nonrigid registration of four-dimensional MR data were incorporated in the system matrix of the fully developed three-dimensional OSEM algorithm along with motion-dependent attenuation maps. The measured shiftvarying sinogram-based detector PSF model, obtained from multiple point source measurements, was also included in the reconstruction. Image reconstruction was performed using: (1) respiratory gating with only one of the seven motion phases used (OP-OSEM “gated”), (2) no motion correction (OPOSEM), (3) no motion correction with PSF modeling (PSFOP-OSEM), (4) MR-based motion correction (MC-OP-OSEM), and (5) motion and PSF corrections (‘MC-PSF-OP-OSEM”). In the oncologic patient studies, image quality was quantitatively assessed by computing the tumor-to-background contrast (Clesion ) for each lesion, mathematically defined as
I ; Clesion ¼ max B¯ where I max is the maximal reconstructed tumor uptake and B¯ is the average uptake in a homogeneous region directly surrounding the lesion (e.g., liver). We performed a cardiac pig study using PET-MR. For this study, a domestic pig was infarcted 4 weeks prior to the scan. The infarction was a result of proximal left anterior descending occlusion after the first branch through a balloon catheter. The pig was injected with 6.5 mCi 18 F-BMS747158-02 (a flow radiopharmaceutical tracer) and was scanned for 2 h on a whole-body PET-MR scanner. Gadobenate dimeglumine
033502-3
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
Ouyang et al.: Quantitative simultaneous positron emission tomography and magnetic resonance imaging
Fig. 2 A coronal PET FDG slice reconstructed using (a) OP-OSEM and (b) PSF-OP-OSEM. PSF modeling yielded higher radiotracer thyroid uptake to background ratios (TBR).
(529 mg∕mL) was injected 15 min prior to euthanasia by pentobarbitol. In the cardiac porcine study, two regions of interest (ROI) were drawn in the myocardium (M) and the blood pool (B) in the reconstructed PET images [see Fig. 6(a)] and the myocardium-to-blood contrast (Cmyo ) was computed as follows:
coils. The mean bias is −7.7% and −6.7% for the liver and softtissue compartments, respectively. The voxel-wise standard deviation bias is 2.1% and 5.4% for the liver and soft-tissue compartments, respectively. If surface coils are not accounted for in the PET attenuation map, a large bias especially near the
¯ M Cmyo ¼ ¯ ; B where operator ð·Þ¯ denotes the mean intensity in a given ROI. In each in vivo study described above, a four classes (air, lung, soft-tissue, and nonfat soft tissues) MR-based attenuation map was obtained using the vendor’s supplied MRAC (MR Attenuation Correction) acquisition protocol. A final attenuation map combining the attenuating effects of the MR hardware (e.g., patient bed, spine coils) was then used for attenuation correction of PET emission data during iterative reconstruction. However, the attenuating effects of the flexible MR surface coils were not taken into account in this work. Random and scatter coincidences distributions were estimated using, respectively, a smoothed delayed coincidence window and the single scatter simulation method,42 and were incorporated in the forward model of PET reconstruction. Five iterations and 12-angular subsets were used for the PSF-OP-OSEM algorithm, which corresponds to 60 “EM-equivalent” image iterations, close to the manufacturer’s recommended number of iterations (3 iterations, 21 subsets). The iteration number for OP-OSEM was chosen so that almost the same background noise level was obtained as for the PSFOP-OSEM. All patient studies were approved by the Institutional Review Board at MGH and all subjects signed a written informed consent form. All the animal studies were authorized by the institutional Care and Use Committee at MGH.
3 3.1
Results and Discussions PET Attenuation Correction of MR Surface Coils
Figure 1(a) shows the anthropomorphic phantom with the MR surface coils attached. Figures 1(b) and 1(c) show a PET bias image and a plot of three different bias profiles, respectively. The absolute bias near the coils can be as high as 24%. Based on this phantom study, the absolute bias in a region can be less than 10% if the region is at least 14 cm away from the Journal of Medical Imaging
Fig. 3 Magnetic resonance (MR) motion tracking in subject 2. (a) Tagged MR images with estimated motion fields (The displacement vector fields are depicted by the small yellow arrows). (b) Breathing navigator echoes (The bright dots indicate the navigator is in the reference respiratory phase).
033502-4
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
Ouyang et al.: Quantitative simultaneous positron emission tomography and magnetic resonance imaging
3.2
Point Spread Function Modeling for PET-MR Scanner
For subject 1, Figs. 2(a) and 2(b) show the PET images in coronal orientation reconstructed with OP-OSEM and PSF-OPOSEM, respectively. Different iteration numbers were used for OP-OSEM and PSF-OP-OSEM so that the level of the background noise is almost the same for both methods. For left and right sides, PSF modeling significantly enhanced radiotracer thyroid uptake to background ratio by 22% and 20%, respectively, as compared to nonPSF modeling. High contrast improvement was achieved by PSF modeling for a small uptake region.
3.3 Fig. 4 Time-dependent PET attenuation map acquired on a subject. The reference attenuation map was directly obtained from the mMR scanner.
surface coils can be found. This is unfortunately the case for today’s PET-MR scanners because surface coils may move during the scan and incorporating time-dependent attenuation of the surface coils is challenging. One solution is to attach multiple MR markers, such as tracking coils,43,44 or vitamin E capsules, to the surface coils. The locations of these markers can be identified so that an accurate time-dependent coil attenuation map can be accounted for in PET AC. Ideally, the MR sequence, which is used in tracking the motion of the surface coils, should be interleaved with the MR imaging sequence used in the scan.
MR-Based PET Motion Correction for Liver and Cardiac Studies
Figure 3(a) shows a transverse view of the heart of subject 2. The displacement vectors, which are depicted by the small yellow arrows, show the computed myocardium displacement fields. The pencil-beam navigator, which was placed perpendicular to the dome of the right hemi-diaphragm, shows the displacement of the diaphragm with breathing. Navigator positions were calculated and returned in real time allowing to accurate tracking of the breathing motion. Figure 4(a) shows a reference attenuation map at the end of expiration acquired using the vendor’s supplied MR-based attenuation correction sequence (MRAC). Figure 4(b) shows the attenuation map obtained by transforming the reference attenuation map to the end of inspiration using the measured motion fields. Figure 5(a) shows one TrueFISP MR sagittal slice at the end of exhalation for subject 2. Figures 5(b) to 5(f) show
Fig. 5 Sagittal MRI (a) and corresponding PET slices reconstructed using (b) respiratory gating (OPOSEM “gated”) (c) no motion correction (OP-OSEM) (d) no motion correction with PSF modeling (PSF-OP-OSEM), (e) MR-based motion correction (MC-OP-OSEM) and (f) motion and PSF corrections (‘MC-PSF-OP-OSEM’). The gated volume was reconstructed using PET counts detected at end-exhalation (Approximately one-seventh of the total number of events). As can be seen, the proposed reconstruction methods (e and f) substantially improved hepatic lesion contrast and structure delineation as compared to the conventional uncorrected reconstruction method (c).
Journal of Medical Imaging
033502-5
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
Ouyang et al.: Quantitative simultaneous positron emission tomography and magnetic resonance imaging
Fig. 6 Heart short axis MRI (a) and corresponding PET slices reconstructed using (b) cardiac gating (OP-OSEM “gated”) (c) no motion correction (OP-OSEM) (d) no motion correction with PSF modeling (PSF-OP-OSEM), (e) MR-based motion correction (MC-OP-OSEM) and (f) motion and PSF corrections (“MC-PSF-OP-OSEM”). ROIs located on the myocardium for contrast calculations are shown in (a). The proposed reconstruction methods (e and f) substantially improved reconstructed myocardial contrast and wall delineation (see yellow arrows) as compared to the conventional uncorrected reconstruction method (c).
reconstructed PET coronal slices with gating, no motion correction, and motion correction with PSF modeling, motion correction, and motion correction with PSF modeling, respectively. The motion correction method removed motion blurring in the tumors without increasing the noise in contrast to the gating method. MR-based motion correction alone increased reconstructed lesion contrast (Clesion ) by 36.5% to 52%, while the combination of motion correction and PSF modeling increased Clesion by 53.8% to 98% with reference to conventional nonmotion corrected OP-OSEM reconstruction. Figure 6(a) shows a short-axis MR slice for the pig study. Figures 6(b) to 6(f) show reconstructed short-axis slices with gating, no motion correction, motion correction with PSF modeling, motion correction, and motion correction with PSF modeling, respectively. As compared to motion uncorrected, the motion-corrected method improves the myocardium wall uptake by 25% to 35%. The motion-corrected method yields similar myocardium to background contrast to the gated method but with a much lower noise level. MR-based motion correction increased reconstructed myocardium-to-blood contrast (Cmyo ) by 7% to 40.9%, depending on the ROI location, while the combination of motion correction and PSF modeling increased Cmyo by 60.6% to 122.7% with reference to conventional nonmotion corrected OP-OSEM reconstruction. The results obtained illustrate the ability of the proposed methods to substantially improve the quality of both abdominal and cardiac PET-MR studies. The potential clinical impact of these methods may be seen as twofold. First, the increase in both lesion and reconstructed myocardial wall contrast may translate into substantially higher detectability of small abdominal tumor foci and perfusion/viability myocardial defects, respectively. Second, the removal of nonrigid motion in both emission and attenuation data as well as the reduction of detectors PSF underestimation effects may result in substantially higher quantitative Journal of Medical Imaging
accuracies of both abdominal and cardiac PET images. This increased quantitative ability of PET is expected to benefit any application where quantitation is necessary, for instance, when assessing responses to chemotherapy in oncology, or reperfusion procedures in cardiac applications.
4
Conclusion
We have demonstrated that the MR surface coil must be included in the attenuation map for accurate PET quantitation. We also showed that modeling PSF and MR-based tissue deformations in the iterative PET reconstruction process significantly improves PET image quality. The results obtained suggest that PET-MR acquisition with the proposed methods can significantly enhance the performance of tumor and myocardial defect diagnosis as compared to conventional methods.
Acknowledgments This work was supported in part by NIH (Grants Nos: R21EB012326, R01-CA165221, R01-HL110241, and R01HL118261).
References 1. H. Zaidi, M. L. Montandon, and D. O. Slosman, “Magnetic resonance imaging-guided attenuation and scatter corrections in three-dimensional brain positron emission tomography,” Med. Phys. 30(5), 937–948 (2003). 2. H. Zaidi, M. L. Montandon, and S. R. Meikle, “Strategies for attenuation compensation in neurological PET studies,” NeuroImage 34(2), 518–541 (2007). 3. T. Beyer et al., “MR-based attenuation correction for torso-PET/MR imaging: pitfalls in mapping MR to CT data,” Eur. J. Nucl. Mol. Imaging 35(6), 1142–1146 (2008).
033502-6
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
Ouyang et al.: Quantitative simultaneous positron emission tomography and magnetic resonance imaging 4. A. Martinez-Moller et al., “Tissue classification as a potential approach for attenuation correction in whole-body PET/MRI: evaluation with PET/CT data,” J. Nucl. Med. 50(4), 520–526 (2009). 5. V. Keereman et al., “MRI-based attenuation correction for PET/MRI using ultrashort echo time sequences,” J. Nucl. Med. 51(5), 812–818 (2010). 6. C. Catana et al., “Toward implementation an MR-based PET attenuation-correction method for neurologic studies on the MR-PET brain prototype,” J. Nucl. Med. 51(9), 1431–1438 (2010). 7. J. Steinberg et al., “Three-region MR_based whole-body attenuation correction for automated PET reconstruction,” Nucl. Med. Biol. 37 (2), 227–235 (2010). 8. V. Schulz et al., “Automatic, three-segment, MR-based attenuation correction for whole-body PET/MR data,” Eur. J. Nucl. Mol. Imaging 38(1), 138–152 (2011). 9. M. Eiber et al., “Value of a dixon-based MR/PET attenuation correction sequence for the localization and evaluation of PET-positive lesions,” Eur. J. Nucl. Mol. Imaging 38(9), 1691–1701 (2011). 10. V. Keereman et al., “The effect of errors in segmented attenuation maps on PET quantification,” Med. Phys. 38(11), 6010–6019 (2011). 11. Y. Berker et al., “MRI-based attenuation correction for hybrid PET/MRI systems: a 4-class tissue segmentation technique using a combined ultrashort-echo-time/Dixon MRI sequence,” J. Nucl. Med. 53(5), 796–804 (2012). 12. A. Samarin et al., “PET/MR imaging of bone lesions: implications for PET quantification from imperfect attenuation correction,” Eur. J. Nucl. Med. Mol. Imaging 39(7), 1154–1160 (2012). 13. A. S. Drzezga et al., “First clinical experience with integrated wholebody PET/MR: comparison to PET/CT in patients with oncologic diagnoses,” J. Nucl. Med. 53(6), 845–855 (2012). 14. J. Kim et al., “Comparison of segmentation-based attenuation correction methods for PET/MRI: evaluation of bone and liver standardized uptake value with oncologic PET/CT data,” J. Nucl. Med. 53(12), 1878–1882 (2012). 15. J. Ouyang et al., “Bias atlases for segmentation-based PET attenuation correction using PET-CT and MR,” IEEE Trans. Nucl. Sci. 60(5), 3373– 3382 (2013). 16. M. Hofmann et al., “MRI-based attenuation correction for PET/MRI: a novel approach combining pattern recognition and atlas registration,” J. Nucl. Med. 49(11), 1875–1883 (2008). 17. A. Johansson et al., “Voxel-wise uncertainty in CT substitute derived from MRI,” Med. Phys. 39(6), 3283–3290 (2012). 18. J. Nuyts et al., “Simultaneous maximum a posteriori reconstruction of attenuation and activity distributions from emission sinograms,” IEEE Trans. Med. Imaging 18(5), 393–403 (1999). 19. A. Salomon et al., “Simultaneous reconstruction of activity and attenuation for PET/MR,” IEEE Trans. Med. Imaging 30(3), 804–813 (2011). 20. M. Defrise, A. Rezaei, and J. Nuyts, “Time-of-flight PET data determine the attenuation sinogram up to a constant,” Phys. Med. Biol. 57(4), 885–899 (2012). 21. A. Rezaei et al., “Simultaneous reconstruction of activity and attenuation in time-of flight PET,” IEEE Trans. Med. Imaging 31(12), 2224– 2233 (2012). 22. J. Nuyts et al., “Completion of a truncated attenuation image from the attenuated PET emission data,” IEEE Trans. Med. Imaging 32(2), 237– 246 (2013). 23. M. Defrise, A. Rezaei, and J. Nuyts, “Transmission-less attenuation correction in time-of-flight PET: analysis of a discrete iterative algorithm,” Phys. Med. Biol. 594, 1073–1095 (2014).
Journal of Medical Imaging
24. S. Furst et al., “Impact of MRI surface coils on quantification in integrated PET/MR,” J. Nucl. Med. 53(Suppl 1), 436 (2012). 25. L. Geworski et al., “Recovery correction for quantitation in emission tomography: a feasibility study,” Eur. J. Nucl. Med. Mol. Imaging 27(2), 161–169 (2000). 26. M. Soret, S. L. Bacharach, and I. Buvat, “Partial-volume effect in PET tumor imaging,” J. Nucl. Med. 48(6), 932–945 (2007). 27. L. Boucher et al., “Respiratory gating for 3-dimensional PET of the thorax: feasibility and initial results,” J. Nucl. Med. 45(2), 214–219 (2004). 28. S. A. Nehmeh et al., “Effect of respiratory gating on quantifying PET images of lung cancer,” J. Nucl. Med. 43(7), 876–881 (2002). 29. A. Martinez-Moller et al., “An approach for MR-based attenuation correction for combined MR/PET: effects of ignoring bones,” J. Nucl. Med. 48Suppl 2, 156P (2007). 30. F. Buther et al., “List mode-driven cardiac and respiratory gating in PET,” J. Nucl. Med. 50(5), 674–681 (2009). 31. A. Montgomery et al., “Correction of head movement on PET studies: comparison of methods,” J. Nucl. Med. 47(12), 1936–1944 (2006). 32. G. Klein, R. Reutter, and R. H. Huesman, “Four-dimensional affine registration models for respiratory-gated PET,” IEEE Trans. Nucl. Sci. 48(3), 756–760 (2002). 33. L. Livieratos et al., “Rigid-body transformation of list-mode projection data for respiratory motion correction in cardiac PET,” Phys. Med. Biol. 50(14), 3313–3322 (2005). 34. F. Lamare et al., “Respiratory motion correction for PET oncology applications using affine transformation of list mode data,” Phys. Med. Biol. 52(1), 121–140 (2007). 35. T. Li et al., “Model-based image reconstruction for four dimensional PET,” Med. Phys. 33, 1288–98 (2006). 36. F. Qiao et al., “A motion-incorporated reconstruction method for gated PET,” Phys. Med. Biol. 51(15), 3769–3783 (2006). 37. M. Reyes et al., “Model-based respiratory motion compensation for emission tomography image reconstruction,” Phys. Med. Biol. 52 (12), 3579–3600 (2007). 38. Y. Petibon et al., “Cardiac motion compensation and resolution modeling in simultaneous PET-MR: a cardiac lesion detection study,” Phys. Med. Biol. 58(7), 2085–2102 (2013). 39. Y. Petibon et al., “Towards coronary plaque imaging using simultaneous PET-MR: a simulation study,” Phys. Med. Biol. 59(5), 1203–1222 (2014). 40. Y. Petibon et al., “Relative role of motion and PSF compensation in whole-body oncologic PET-MR imaging,” Med. Phys. 41, 042503 (2014). 41. C. Huang et al., “A novel golden-angle radial FLASH motion-estimation sequence for simultaneous thoracic PET/MR,” in Proc. Int. Society for Magnetic Resonance in Medicine (ISMRM), Berkeley, California (2013). 42. C. C. Watson, “New, faster, image-based scatter correction for 3D PET,” IEEE Trans. Nucl. Sci. 47(4), 1587–1594 (2000). 43. C. Huang et al., “MR-based motion correction for PET imaging using wired active MR microcoils in simultaneous PET-MR: phantom study,” Med. Phys. 41(4), 041910 (2014). 44. C. Huang et al., “Motion compensation for brain PET imaging using wireless MR active markers in simultaneous PET-MR: phantom and non-human primate studies,” NeuroImage 91, 129–137 (2014). Biographies of the authors are not available.
033502-7
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Journal of Medical Imaging 1(3), 033502 (Oct–Dec 2014)
Quantitative simultaneous positron emission tomography and magnetic resonance imaging Jinsong Ouyang,a,b,*,† Yoann Petibon,a,c,† Chuan Huang,a,b Timothy G. Reese,b,d Aleksandra L. Kolnick,a and Georges El Fakhria,b
a Massachusetts General Hospital, Center for Advanced Medical Imaging Sciences, Division of Nuclear Medicine and Molecular Imaging, Department of Radiology, Boston, Massachusetts 02114, United States b Harvard Medical School, Department of Radiology, Boston, Massachusetts 02114, United States c Sorbonne Universités, UPMC Univ Paris 06, Paris 75634, France d Massachusetts General Hospital, Martinos Center for Biomedical Imaging, Department of Radiology, Charlestown, Massachusetts 02129, United States
Abstract. Simultaneous positron emission tomography and magnetic resonance imaging (PET-MR) is an innovative and promising imaging modality that is generating substantial interest in the medical imaging community, while offering many challenges and opportunities. In this study, we investigated whether MR surface coils need to be accounted for in PET attenuation correction. Furthermore, we integrated motion correction, attenuation correction, and point spread function modeling into a single PET reconstruction framework. We applied our reconstruction framework to in vivo animal and patient PET-MR studies. We have demonstrated that our approach greatly improved PET image quality. © 2014 Society of Photo-Optical Instrumentation Engineers (SPIE) [DOI: 10.1117/1 .JMI.1.3.033502]
Keywords: positron emission tomography and magnetic resonance imaging; positron emission tomography attenuation correction; surface coils; point spread function modeling; motion correction. Paper 14054PR received Apr. 22, 2014; revised manuscript received Aug. 19, 2014; accepted for publication Oct. 2, 2014; published online Nov. 3, 2014.
1
Introduction
Accurate attenuation correction (AC) is critical for quantitative whole-body positron emission tomography and magnetic resonance (PET-MR) imaging. The AC approaches for PET-MR can be grouped into three different categories: segmentation based,1–15 atlas registration based,3,16,17 and simultaneous maximum-likelihood reconstruction of attenuation and activity.18–23 Segmentation-based approach, which is to assign a global attenuation coefficient to an identified tissue class, appears to be the only one that is feasible in clinical practice for whole-body PET-MR. However, one of the major problems associated with such an approach is the large bias on PET quantification caused by the inter- and intra-patient electron density variations, especially in lungs and bones.15 This is why MRbased AC still remains a challenge for PET-MR. It may decrease electron density variation to reduce the bias on PET quantitation if patient studies are grouped according to their bed positions, patients’ ages, genders, and body mass indices and groupdependent attenuation coefficients are assigned to segmented tissue classes. Siemens whole-body PET-MR (mMR) surface coils, which can be placed above the patient during a PET-MR scan, are currently not accounted for in PET AC because the location and shape of the coils are unknown during the scan. Furst et al.24 acquired data of six patients, who had a total of eight lesions, with and without surface coils attached. The standardized uptake value difference for the eight lesions varied from −18% to 60%.
*Address all correspondence to: Jinsong Ouyang, E-mail: ouyang.jinsong@ mgh.harvard.edu †
These authors contributed equally.
Journal of Medical Imaging
The partial volume effects (PVEs), resulting from the scanner’s limited point spread function (PSF), are also a major source of image spatial resolution degradation. PVEs cause the signal to spread from a target tissue into neighboring tissues (“spill-out”) and from those neighboring tissues into the target tissue (“spill-in”). PVEs also reduce signal-to-noise ratio (SNR) and introduce a bias in measured activity concentration which significantly hampers the accuracy of the quantitation of the PET images. For lesions smaller than the reconstructed spatial resolution, PVEs may lead to more than 50% underestimation of the true radiotracer concentration.25,26 The myocardial wall, whose thickness is close to the PET resolution, particularly suffers from PVEs. PVEs can be greatly alleviated by modeling the PSF in the PET system matrix during reconstruction. Patient motion is another main contributing factor to image degradation in PET. The intrinsic spatial resolution of modern whole-body PET scanners is in the range of 4 to 5 mm. However, due to inevitable involuntary physiological patient motion, this resolution cannot currently be achieved in either abdominal or cardiac PET studies. Both respiratory and cardiac motions are the main causes of image degradation in thoracic and abdominal PET studies. Furthermore, it is important to highlight that it is not just the emission data that are corrupted by motion; there are often large reconstruction errors due to the deformation of the attenuating medium. In order to make proper use of emission data from all motion phases, one must know the corresponding attenuation at all times. There have been many attempts to develop practical and effective methodologies to remove the effects of cardiac/respiratory motion.27–37 Cardiac and/or respiratory gating strategies that “freeze” motion are 0091-3286/2014/$25.00 © 2014 SPIE
033502-1
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
Ouyang et al.: Quantitative simultaneous positron emission tomography and magnetic resonance imaging
popular in abdominal and cardiac PET studies. However, gating is not an optimal solution to the motion problem as it dramatically hampers the images’ SNR, and requires longer studies to achieve good image quality. Previously, we have studied bias atlases for segmentationbased PET attenuation correction using PET-CT and MR.15 We have also developed robust and accurate MRI methods allowing the tracking and measurement of both respiratory and cardiac motions during abdominal or cardiac studies. The MR-measured motion fields as well as PSF modeling are integrated into a PET reconstruction framework.38–40 In this PET reconstruction framework, a “time-dependent” attenuation map is deformed by the motion fields and is consistent with the PET emission distribution. In this study, we present new results on attenuation correction on MR surface coils. We also summarize our PET reconstruction framework, which incorporates AC, PSF modeling, and motion correction, and show its performance for in vivo animal and patient PET-MR studies.
2 2.1
Materials and Methods Attenuation Correction of MR Surface Coils
A bias image was then calculated using the image reconstructed with the phantom-coil-μ-map as the reference.
2.2
We measured and modeled the radially varying PSF of the mMR scanner in the sinogram domain. For a given axial position, we simultaneously placed 8 ∼ 0.5-mm 18 F point sources at radial positions with 2.5 cm spacing across the PET-MR scanner field of view. We acquired ∼20-million coincidences for each of the four different axial positions. For each point source, axial and radial sinogram profiles through the point source peak value were extracted. The obtained profiles were fitted using radially varying asymmetric exponential kernels to account for detector blurring effects along the radial and axial directions. We have previously validated this model, i.e., radially varying asymmetric exponential kernels, in phantom studies.40 In this study, we applied this model to in vivo imaging to evaluate its performance.
2.3
Because mMR surface coils include some metal components, we decided to acquire the μ-map of the surface coils using a transmission scan to avoid CT metal artifacts. A transmission scan of an anthropomorphic phantom (Data Spectrum Corporation, Hillsborough, North Carolina) with a Siemens mMR surface coil array attached [See Fig. 1(a)] was first acquired on a Siemens ECAT HR+ scanner. The same phantom was filled with F-18 in both the liver and soft-tissue compartments (liver/background concentration ratio: 2.4). Additionally, three 1-cm “tumors” were placed around the liver (tumor/liver concentration ratio: 4). The phantom was placed in a Siemens Biograph PET-CT scanner. First, a CT with 120 kVp was acquired. Second, an mMR surface coil array was placed on the top of the phantom in the same way we did for the transmission scan. Third, a PET acquisition which included total 279-million coincidence events was performed. The CT image was first transformed into a μ-map, denoted phantomμ-map, using Siemens e7 tools. The coil image obtained from the transmission scan was digitally added to the phantom-μ-map to create another μ-map, denoted phantomcoil-μ-map. The PET data were reconstructed twice: once with phantom-μ-map and the other with phantom-coil-μ-map.
Point Spread Function Modeling for PET
Motion Measurement Using MR
We have developed a set of MRI pulse sequences allowing the measurements of respiratory and cardiac motions for lower abdomen and heart PET-MR imaging studies. For imaging the lower abdomen, we have developed a dedicated MRI pulse sequence (denoted “NAV-TrueFISP”) allowing the accurate measurement of respiratory motion in the lower abdomen. The NAV-TrueFISP sequence is composed of twodimensional (2-D) multislice steady-state free precession MRI acquisitions (“TrueFISP”) interleaved by pencil-beam navigator echoes.40 The acquired TrueFISP slices are chosen in sagittal orientation to keep the main components of the motion within the acquisition plane. Pencil-beam navigators are acquired prior to each TrueFISP acquisition and are used to track the lung-liver interface during the acquisition with high temporal sampling frequency (2 to 4 Hz). The obtained internal motion surrogates allow accurate monitoring of the respiratory cycle (binning of both PET and MRI data into respiratory phases) and handling of respiratory cycle irregularities. This sequence takes advantage of the specific TrueFISP contrast to provide anatomical landmarks in homogeneous organs, such as the liver [see blood vessels with bright signal as illustrated in Fig. 6(a)]. Typical acquisition times for the NAV-TrueFISP protocol range between
Fig. 1 Phantom study using Siemens mMR surface coils. (a) An mMR surface coil array was placed on the top of a data spectrum anthropomorphic torso phantom, (b) positron emission tomography (PET) bias image, (c) PET bias profiles.
Journal of Medical Imaging
033502-2
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
Ouyang et al.: Quantitative simultaneous positron emission tomography and magnetic resonance imaging
2.5 and 3.5 min, depending on the number of slices needed to cover the area of interest. We have also developed MRI sequences for cardiac imaging. Motion of the heart has two different components: motion caused by the heart beating (cardiac motion) and motion caused by respiration. Because of their different mechanical and temporal properties, cardiac and respiratory motions of the heart are measured separately using specific MRI pulse sequences. The obtained cardiac and respiratory motion fields are then combined to compute the motion field between any cardiac/respiratory phase and a reference one in which the PET data will be reconstructed. Cardiac motion is measured using a tagged-MRI pulse sequence (denoted as “GRE-NAV-Tag”) developed specifically for this motion. Special RF/gradients series are applied, introducing periodic magnetization modulation (known as “tags”) on the myocardium. Tagging occurs at every heartbeat, triggered by the R-wave. Next, a series of dynamic GRE (Gradient Recalled Echo) MR volumes are obtained in which the tags are visible. Tag lines’ deformations visible in the images allow characterizing complex motion components, such as shearing, torsion, or radial thickening. The sequence is also navigated by a pencil-beam navigator, allowing the acquisition of tagged MR data only at end-exhalation respiratory phase. Electrocardiogram signal and navigators are simultaneously recorded to assign a cardiac and respiratory phase to each acquired PET event. Three orthogonal directions of tagging are acquired to fully characterize cardiac motion. The nonaccelerated tagging acquisition takes about 10 min using our method (∼3.5 min for each tagging direction). The acquisition can be accelerated using partial k-space sampling, parallel imaging, compressed sensing, or a combination of these techniques. Moreover, a dedicated golden-angle radial-FLASH sequence developed in our group is used to measure the respiratory motion of the heart. This sequence measures the respiratory motion of the heart in less than 2 min.41 As previously mentioned, the motion field between any cardiac/respiratory phase and the reference phase is computed by nonrigid summing of the measured cardiac and respiratory motions. For both lower abdomen and cardiac motion measurements, each subject is instructed to breath as regularly as possible during the motion tracking MR acquisitions (i.e., NAV-TrueFISP for abdomen and GRE-NAV-Tag + radial-FLASH for heart). Based on the navigator amplitude, extreme, “abnormal” events (e.g., deep breath), not representative of a normal respiratory cycle, are discarded.
2.4
PET Reconstruction
We integrated MR-based AC, PSF modeling, and MR-based motion correction into a unified iterative reconstruction framework.38–40 In PET, the estimated sinograms, y, are
y ¼ Pf þ s þ r; where f is the PET image volume to be estimated, s and r are the estimated scattered and random contributions, and P is the system matrix. The system matrix includes an MR-based warping operator registering a given motion frame to a selected reference frame, a forward-projection operator, time-dependent attenuation factors, PSF blurring effects in the sinogram space, and detector normalization factors. For dynamic reconstruction, list-mode data are divided into multiple frames according to Journal of Medical Imaging
the requirements for kinetic analysis. Motion correction is applied frame by frame using a global reference motion frame. The system matrix can be defined as
P ¼ NBAGM; where M is the MR-based warping operator that registers a given motion phase to the reference motion phase, G is the forward-projection operator, A contains the AC factors for each frame, B models the 2-D PSF blurring effects in the sinogram space, and N represents the detector normalization factors. In this study, we applied this unified reconstruction framework to in vivo animal and patient studies.
2.5
Animal and Patient Studies
PSF modeling was evaluated using a patient study performed on the mMR scanner at Massachusetts General Hospital (MGH). A 72-year-old male subject (subject 1) was enrolled to participate in the study. The subject was administrated with 370 MBq18 F-FDG. 18 F-FDG-PET list-mode data were collected during 5 min, resulting in a total of 320-million PET counts. The acquired sinogram data were reconstructed using ordinary poisson OSEM (OP-OSEM) and OP-OSEM with PSF modeling (PSF-OP-OSEM). Simultaneous PET/MR scans were performed in another subject (subject 2) with known liver lesions. A PET listmode was collected for 210 s following a 15 mCi FDG injection. Twenty-five sagittal slices covering the whole liver were simultaneously acquired using a 2-D multislice steady-state free precession TrueFISP sequence (TE ¼ 1.67 ms, TR ¼ 320 ms, 2 × 2 × 8 mm voxel size) with each slice acquisition repeated 30 times. MR slices and PET events were rebinned into seven respiratory phases using the diaphragm position. Motion fields obtained from B-spline nonrigid registration of four-dimensional MR data were incorporated in the system matrix of the fully developed three-dimensional OSEM algorithm along with motion-dependent attenuation maps. The measured shiftvarying sinogram-based detector PSF model, obtained from multiple point source measurements, was also included in the reconstruction. Image reconstruction was performed using: (1) respiratory gating with only one of the seven motion phases used (OP-OSEM “gated”), (2) no motion correction (OPOSEM), (3) no motion correction with PSF modeling (PSFOP-OSEM), (4) MR-based motion correction (MC-OP-OSEM), and (5) motion and PSF corrections (‘MC-PSF-OP-OSEM”). In the oncologic patient studies, image quality was quantitatively assessed by computing the tumor-to-background contrast (Clesion ) for each lesion, mathematically defined as
I ; Clesion ¼ max B¯ where I max is the maximal reconstructed tumor uptake and B¯ is the average uptake in a homogeneous region directly surrounding the lesion (e.g., liver). We performed a cardiac pig study using PET-MR. For this study, a domestic pig was infarcted 4 weeks prior to the scan. The infarction was a result of proximal left anterior descending occlusion after the first branch through a balloon catheter. The pig was injected with 6.5 mCi 18 F-BMS747158-02 (a flow radiopharmaceutical tracer) and was scanned for 2 h on a whole-body PET-MR scanner. Gadobenate dimeglumine
033502-3
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
Ouyang et al.: Quantitative simultaneous positron emission tomography and magnetic resonance imaging
Fig. 2 A coronal PET FDG slice reconstructed using (a) OP-OSEM and (b) PSF-OP-OSEM. PSF modeling yielded higher radiotracer thyroid uptake to background ratios (TBR).
(529 mg∕mL) was injected 15 min prior to euthanasia by pentobarbitol. In the cardiac porcine study, two regions of interest (ROI) were drawn in the myocardium (M) and the blood pool (B) in the reconstructed PET images [see Fig. 6(a)] and the myocardium-to-blood contrast (Cmyo ) was computed as follows:
coils. The mean bias is −7.7% and −6.7% for the liver and softtissue compartments, respectively. The voxel-wise standard deviation bias is 2.1% and 5.4% for the liver and soft-tissue compartments, respectively. If surface coils are not accounted for in the PET attenuation map, a large bias especially near the
¯ M Cmyo ¼ ¯ ; B where operator ð·Þ¯ denotes the mean intensity in a given ROI. In each in vivo study described above, a four classes (air, lung, soft-tissue, and nonfat soft tissues) MR-based attenuation map was obtained using the vendor’s supplied MRAC (MR Attenuation Correction) acquisition protocol. A final attenuation map combining the attenuating effects of the MR hardware (e.g., patient bed, spine coils) was then used for attenuation correction of PET emission data during iterative reconstruction. However, the attenuating effects of the flexible MR surface coils were not taken into account in this work. Random and scatter coincidences distributions were estimated using, respectively, a smoothed delayed coincidence window and the single scatter simulation method,42 and were incorporated in the forward model of PET reconstruction. Five iterations and 12-angular subsets were used for the PSF-OP-OSEM algorithm, which corresponds to 60 “EM-equivalent” image iterations, close to the manufacturer’s recommended number of iterations (3 iterations, 21 subsets). The iteration number for OP-OSEM was chosen so that almost the same background noise level was obtained as for the PSFOP-OSEM. All patient studies were approved by the Institutional Review Board at MGH and all subjects signed a written informed consent form. All the animal studies were authorized by the institutional Care and Use Committee at MGH.
3 3.1
Results and Discussions PET Attenuation Correction of MR Surface Coils
Figure 1(a) shows the anthropomorphic phantom with the MR surface coils attached. Figures 1(b) and 1(c) show a PET bias image and a plot of three different bias profiles, respectively. The absolute bias near the coils can be as high as 24%. Based on this phantom study, the absolute bias in a region can be less than 10% if the region is at least 14 cm away from the Journal of Medical Imaging
Fig. 3 Magnetic resonance (MR) motion tracking in subject 2. (a) Tagged MR images with estimated motion fields (The displacement vector fields are depicted by the small yellow arrows). (b) Breathing navigator echoes (The bright dots indicate the navigator is in the reference respiratory phase).
033502-4
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
Ouyang et al.: Quantitative simultaneous positron emission tomography and magnetic resonance imaging
3.2
Point Spread Function Modeling for PET-MR Scanner
For subject 1, Figs. 2(a) and 2(b) show the PET images in coronal orientation reconstructed with OP-OSEM and PSF-OPOSEM, respectively. Different iteration numbers were used for OP-OSEM and PSF-OP-OSEM so that the level of the background noise is almost the same for both methods. For left and right sides, PSF modeling significantly enhanced radiotracer thyroid uptake to background ratio by 22% and 20%, respectively, as compared to nonPSF modeling. High contrast improvement was achieved by PSF modeling for a small uptake region.
3.3 Fig. 4 Time-dependent PET attenuation map acquired on a subject. The reference attenuation map was directly obtained from the mMR scanner.
surface coils can be found. This is unfortunately the case for today’s PET-MR scanners because surface coils may move during the scan and incorporating time-dependent attenuation of the surface coils is challenging. One solution is to attach multiple MR markers, such as tracking coils,43,44 or vitamin E capsules, to the surface coils. The locations of these markers can be identified so that an accurate time-dependent coil attenuation map can be accounted for in PET AC. Ideally, the MR sequence, which is used in tracking the motion of the surface coils, should be interleaved with the MR imaging sequence used in the scan.
MR-Based PET Motion Correction for Liver and Cardiac Studies
Figure 3(a) shows a transverse view of the heart of subject 2. The displacement vectors, which are depicted by the small yellow arrows, show the computed myocardium displacement fields. The pencil-beam navigator, which was placed perpendicular to the dome of the right hemi-diaphragm, shows the displacement of the diaphragm with breathing. Navigator positions were calculated and returned in real time allowing to accurate tracking of the breathing motion. Figure 4(a) shows a reference attenuation map at the end of expiration acquired using the vendor’s supplied MR-based attenuation correction sequence (MRAC). Figure 4(b) shows the attenuation map obtained by transforming the reference attenuation map to the end of inspiration using the measured motion fields. Figure 5(a) shows one TrueFISP MR sagittal slice at the end of exhalation for subject 2. Figures 5(b) to 5(f) show
Fig. 5 Sagittal MRI (a) and corresponding PET slices reconstructed using (b) respiratory gating (OPOSEM “gated”) (c) no motion correction (OP-OSEM) (d) no motion correction with PSF modeling (PSF-OP-OSEM), (e) MR-based motion correction (MC-OP-OSEM) and (f) motion and PSF corrections (‘MC-PSF-OP-OSEM’). The gated volume was reconstructed using PET counts detected at end-exhalation (Approximately one-seventh of the total number of events). As can be seen, the proposed reconstruction methods (e and f) substantially improved hepatic lesion contrast and structure delineation as compared to the conventional uncorrected reconstruction method (c).
Journal of Medical Imaging
033502-5
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
Ouyang et al.: Quantitative simultaneous positron emission tomography and magnetic resonance imaging
Fig. 6 Heart short axis MRI (a) and corresponding PET slices reconstructed using (b) cardiac gating (OP-OSEM “gated”) (c) no motion correction (OP-OSEM) (d) no motion correction with PSF modeling (PSF-OP-OSEM), (e) MR-based motion correction (MC-OP-OSEM) and (f) motion and PSF corrections (“MC-PSF-OP-OSEM”). ROIs located on the myocardium for contrast calculations are shown in (a). The proposed reconstruction methods (e and f) substantially improved reconstructed myocardial contrast and wall delineation (see yellow arrows) as compared to the conventional uncorrected reconstruction method (c).
reconstructed PET coronal slices with gating, no motion correction, and motion correction with PSF modeling, motion correction, and motion correction with PSF modeling, respectively. The motion correction method removed motion blurring in the tumors without increasing the noise in contrast to the gating method. MR-based motion correction alone increased reconstructed lesion contrast (Clesion ) by 36.5% to 52%, while the combination of motion correction and PSF modeling increased Clesion by 53.8% to 98% with reference to conventional nonmotion corrected OP-OSEM reconstruction. Figure 6(a) shows a short-axis MR slice for the pig study. Figures 6(b) to 6(f) show reconstructed short-axis slices with gating, no motion correction, motion correction with PSF modeling, motion correction, and motion correction with PSF modeling, respectively. As compared to motion uncorrected, the motion-corrected method improves the myocardium wall uptake by 25% to 35%. The motion-corrected method yields similar myocardium to background contrast to the gated method but with a much lower noise level. MR-based motion correction increased reconstructed myocardium-to-blood contrast (Cmyo ) by 7% to 40.9%, depending on the ROI location, while the combination of motion correction and PSF modeling increased Cmyo by 60.6% to 122.7% with reference to conventional nonmotion corrected OP-OSEM reconstruction. The results obtained illustrate the ability of the proposed methods to substantially improve the quality of both abdominal and cardiac PET-MR studies. The potential clinical impact of these methods may be seen as twofold. First, the increase in both lesion and reconstructed myocardial wall contrast may translate into substantially higher detectability of small abdominal tumor foci and perfusion/viability myocardial defects, respectively. Second, the removal of nonrigid motion in both emission and attenuation data as well as the reduction of detectors PSF underestimation effects may result in substantially higher quantitative Journal of Medical Imaging
accuracies of both abdominal and cardiac PET images. This increased quantitative ability of PET is expected to benefit any application where quantitation is necessary, for instance, when assessing responses to chemotherapy in oncology, or reperfusion procedures in cardiac applications.
4
Conclusion
We have demonstrated that the MR surface coil must be included in the attenuation map for accurate PET quantitation. We also showed that modeling PSF and MR-based tissue deformations in the iterative PET reconstruction process significantly improves PET image quality. The results obtained suggest that PET-MR acquisition with the proposed methods can significantly enhance the performance of tumor and myocardial defect diagnosis as compared to conventional methods.
Acknowledgments This work was supported in part by NIH (Grants Nos: R21EB012326, R01-CA165221, R01-HL110241, and R01HL118261).
References 1. H. Zaidi, M. L. Montandon, and D. O. Slosman, “Magnetic resonance imaging-guided attenuation and scatter corrections in three-dimensional brain positron emission tomography,” Med. Phys. 30(5), 937–948 (2003). 2. H. Zaidi, M. L. Montandon, and S. R. Meikle, “Strategies for attenuation compensation in neurological PET studies,” NeuroImage 34(2), 518–541 (2007). 3. T. Beyer et al., “MR-based attenuation correction for torso-PET/MR imaging: pitfalls in mapping MR to CT data,” Eur. J. Nucl. Mol. Imaging 35(6), 1142–1146 (2008).
033502-6
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
Ouyang et al.: Quantitative simultaneous positron emission tomography and magnetic resonance imaging 4. A. Martinez-Moller et al., “Tissue classification as a potential approach for attenuation correction in whole-body PET/MRI: evaluation with PET/CT data,” J. Nucl. Med. 50(4), 520–526 (2009). 5. V. Keereman et al., “MRI-based attenuation correction for PET/MRI using ultrashort echo time sequences,” J. Nucl. Med. 51(5), 812–818 (2010). 6. C. Catana et al., “Toward implementation an MR-based PET attenuation-correction method for neurologic studies on the MR-PET brain prototype,” J. Nucl. Med. 51(9), 1431–1438 (2010). 7. J. Steinberg et al., “Three-region MR_based whole-body attenuation correction for automated PET reconstruction,” Nucl. Med. Biol. 37 (2), 227–235 (2010). 8. V. Schulz et al., “Automatic, three-segment, MR-based attenuation correction for whole-body PET/MR data,” Eur. J. Nucl. Mol. Imaging 38(1), 138–152 (2011). 9. M. Eiber et al., “Value of a dixon-based MR/PET attenuation correction sequence for the localization and evaluation of PET-positive lesions,” Eur. J. Nucl. Mol. Imaging 38(9), 1691–1701 (2011). 10. V. Keereman et al., “The effect of errors in segmented attenuation maps on PET quantification,” Med. Phys. 38(11), 6010–6019 (2011). 11. Y. Berker et al., “MRI-based attenuation correction for hybrid PET/MRI systems: a 4-class tissue segmentation technique using a combined ultrashort-echo-time/Dixon MRI sequence,” J. Nucl. Med. 53(5), 796–804 (2012). 12. A. Samarin et al., “PET/MR imaging of bone lesions: implications for PET quantification from imperfect attenuation correction,” Eur. J. Nucl. Med. Mol. Imaging 39(7), 1154–1160 (2012). 13. A. S. Drzezga et al., “First clinical experience with integrated wholebody PET/MR: comparison to PET/CT in patients with oncologic diagnoses,” J. Nucl. Med. 53(6), 845–855 (2012). 14. J. Kim et al., “Comparison of segmentation-based attenuation correction methods for PET/MRI: evaluation of bone and liver standardized uptake value with oncologic PET/CT data,” J. Nucl. Med. 53(12), 1878–1882 (2012). 15. J. Ouyang et al., “Bias atlases for segmentation-based PET attenuation correction using PET-CT and MR,” IEEE Trans. Nucl. Sci. 60(5), 3373– 3382 (2013). 16. M. Hofmann et al., “MRI-based attenuation correction for PET/MRI: a novel approach combining pattern recognition and atlas registration,” J. Nucl. Med. 49(11), 1875–1883 (2008). 17. A. Johansson et al., “Voxel-wise uncertainty in CT substitute derived from MRI,” Med. Phys. 39(6), 3283–3290 (2012). 18. J. Nuyts et al., “Simultaneous maximum a posteriori reconstruction of attenuation and activity distributions from emission sinograms,” IEEE Trans. Med. Imaging 18(5), 393–403 (1999). 19. A. Salomon et al., “Simultaneous reconstruction of activity and attenuation for PET/MR,” IEEE Trans. Med. Imaging 30(3), 804–813 (2011). 20. M. Defrise, A. Rezaei, and J. Nuyts, “Time-of-flight PET data determine the attenuation sinogram up to a constant,” Phys. Med. Biol. 57(4), 885–899 (2012). 21. A. Rezaei et al., “Simultaneous reconstruction of activity and attenuation in time-of flight PET,” IEEE Trans. Med. Imaging 31(12), 2224– 2233 (2012). 22. J. Nuyts et al., “Completion of a truncated attenuation image from the attenuated PET emission data,” IEEE Trans. Med. Imaging 32(2), 237– 246 (2013). 23. M. Defrise, A. Rezaei, and J. Nuyts, “Transmission-less attenuation correction in time-of-flight PET: analysis of a discrete iterative algorithm,” Phys. Med. Biol. 594, 1073–1095 (2014).
Journal of Medical Imaging
24. S. Furst et al., “Impact of MRI surface coils on quantification in integrated PET/MR,” J. Nucl. Med. 53(Suppl 1), 436 (2012). 25. L. Geworski et al., “Recovery correction for quantitation in emission tomography: a feasibility study,” Eur. J. Nucl. Med. Mol. Imaging 27(2), 161–169 (2000). 26. M. Soret, S. L. Bacharach, and I. Buvat, “Partial-volume effect in PET tumor imaging,” J. Nucl. Med. 48(6), 932–945 (2007). 27. L. Boucher et al., “Respiratory gating for 3-dimensional PET of the thorax: feasibility and initial results,” J. Nucl. Med. 45(2), 214–219 (2004). 28. S. A. Nehmeh et al., “Effect of respiratory gating on quantifying PET images of lung cancer,” J. Nucl. Med. 43(7), 876–881 (2002). 29. A. Martinez-Moller et al., “An approach for MR-based attenuation correction for combined MR/PET: effects of ignoring bones,” J. Nucl. Med. 48Suppl 2, 156P (2007). 30. F. Buther et al., “List mode-driven cardiac and respiratory gating in PET,” J. Nucl. Med. 50(5), 674–681 (2009). 31. A. Montgomery et al., “Correction of head movement on PET studies: comparison of methods,” J. Nucl. Med. 47(12), 1936–1944 (2006). 32. G. Klein, R. Reutter, and R. H. Huesman, “Four-dimensional affine registration models for respiratory-gated PET,” IEEE Trans. Nucl. Sci. 48(3), 756–760 (2002). 33. L. Livieratos et al., “Rigid-body transformation of list-mode projection data for respiratory motion correction in cardiac PET,” Phys. Med. Biol. 50(14), 3313–3322 (2005). 34. F. Lamare et al., “Respiratory motion correction for PET oncology applications using affine transformation of list mode data,” Phys. Med. Biol. 52(1), 121–140 (2007). 35. T. Li et al., “Model-based image reconstruction for four dimensional PET,” Med. Phys. 33, 1288–98 (2006). 36. F. Qiao et al., “A motion-incorporated reconstruction method for gated PET,” Phys. Med. Biol. 51(15), 3769–3783 (2006). 37. M. Reyes et al., “Model-based respiratory motion compensation for emission tomography image reconstruction,” Phys. Med. Biol. 52 (12), 3579–3600 (2007). 38. Y. Petibon et al., “Cardiac motion compensation and resolution modeling in simultaneous PET-MR: a cardiac lesion detection study,” Phys. Med. Biol. 58(7), 2085–2102 (2013). 39. Y. Petibon et al., “Towards coronary plaque imaging using simultaneous PET-MR: a simulation study,” Phys. Med. Biol. 59(5), 1203–1222 (2014). 40. Y. Petibon et al., “Relative role of motion and PSF compensation in whole-body oncologic PET-MR imaging,” Med. Phys. 41, 042503 (2014). 41. C. Huang et al., “A novel golden-angle radial FLASH motion-estimation sequence for simultaneous thoracic PET/MR,” in Proc. Int. Society for Magnetic Resonance in Medicine (ISMRM), Berkeley, California (2013). 42. C. C. Watson, “New, faster, image-based scatter correction for 3D PET,” IEEE Trans. Nucl. Sci. 47(4), 1587–1594 (2000). 43. C. Huang et al., “MR-based motion correction for PET imaging using wired active MR microcoils in simultaneous PET-MR: phantom study,” Med. Phys. 41(4), 041910 (2014). 44. C. Huang et al., “Motion compensation for brain PET imaging using wireless MR active markers in simultaneous PET-MR: phantom and non-human primate studies,” NeuroImage 91, 129–137 (2014). Biographies of the authors are not available.
033502-7
Downloaded From: http://medicalimaging.spiedigitallibrary.org/ on 07/12/2015 Terms of Use: http://spiedl.org/terms
Oct–Dec 2014
•
Vol. 1(3)
