Impact of incorrect tissue classification in Dixon-based MR-AC: fat-water tissue inversion
© Ladefoged et al.; licensee Springer. 2014
Received: 17 September 2014
Accepted: 9 November 2014
Published: 14 December 2014
The current MR-based attenuation correction (AC) used in combined PET/MR systems computes a Dixon attenuation map (MR-ACDixon) based on fat and water images derived from in- and opposed-phase MRI. We observed an occasional fat/water inversion in MR-ACDixon. The aim of our study was to estimate the prevalence of this phenomenon in a large patient cohort and assess the possible bias on PET data.
PET/MRI was performed on a Siemens Biograph mMR (Siemens AG, Erlangen, Germany). We visually inspected attenuation maps of 283 brain or head/neck (H/N) patients, classified them as non-inverted or inverted, and calculated the fat/water tissue fraction. We selected ten FDG-PET brain patients with non-inverted attenuation maps for further analysis. Tissue inversion was simulated, and PET images were reconstructed using both original and inverted attenuation maps. The FDG-PET images of the ten brain patients were analyzed using 11 concentric annulus regions of 5 mm width placed over a central transaxial image plane traversing PETDixon.
Out of the 283 patients, a fat/water inversion in 23 patients (8.1%) was observed. The average fraction of fat in the correct MR-ACDixon was 13% for brain and 17% for H/N patients. In the inverted cases, we found an average fat fraction of 56% for the brain patients and 41% for the H/N patients. The effect of the simulated tissue inversion in the brain studies was clearly seen on AC-PET images. The percent-difference image revealed a radial error where the largest difference was at the ventricles (30% ± 3%) and smallest at the cortical region (10% ± 2%).
Tissue inversion in Dixon MRI is well known and can occur when there is an error in the off-resonance correction method. Tissue inversion needs to be considered if, based on Dixon-AC, the construction of normal PET databases is performed or any quantitative physiological parameters are fitted. Visual inspection is needed if Dixon-AC is to be used in clinical routine.
Integrated PET/MR is an area of rapidly growing interest in the medical community. One challenge still to overcome is the absence of CT-like transmission sources in combined clinical PET/MR systems to perform attenuation correction (AC). Unlike the Hounsfield units of CT data that are directly related to the linear attenuation coefficients (LACs) of the 511 keV photons in PET, MRI is a function of proton density and tissue relaxation times that have no direct relation to the LACs.
The MR signal is dominated by components arising from protons in water and lipid (fat), which have different resonance frequencies. A two-point Dixon sequence consists of a 0 image with fat and water in-phase (Figure 1A) and a π image with fat and water out-phase (Figure 1B) to obtain chemical shift sensitive images. Water or fat only images can be obtained from the phase images , (Figure 1C,D). The largest signal in the water or fat images may be used to decide the classified tissue in each voxel.
A number of errors in the attenuation maps have been reported since the introduction of combined PET/MR , including tissue inversion where the fat and water classes are incorrectly identified under phase unwrapping ,,. This results in the water image looking like a normal fat image and the fat like a normal water image. As the current attenuation correction used in the combined mMR derives the attenuation map (MR-ACDixon) from fat and water images based on the Dixon MR , a global incorrect assignment of LACs will occur in these cases. Incorrect assignment of LACs can lead to large errors in estimated PET activity if LACs are misclassified or simply under- or overestimated . Here, we evaluate the prevalence of fat-water tissue inversion on the Siemens mMR in a large patient population of head and neck patients and assess the effect on the PET images after attenuation correction.
PET/MRI was performed on a fully integrated Siemens mMR Biograph . All patients were positioned headfirst, with their arms down, and data were acquired over a single bed position of 25.8 cm. Standard MR-based attenuation maps (MR-ACDixon) were derived using the Dixon VIBE sequence , as implemented by the vendor. The Dixon-water image and the MR-AC attenuation maps were reconstructed on 192 × 126 × 128 matrices (2.6 × 2.6 × 3.1 mm voxels).
was calculated, where X is fat or water. We report these results, as well as the averaged weight and age, for all brain and H/N patients.
Of the 283 patients, we selected a subgroup of 17 patients scanned with [18F]-F-fluoro-ethyl-tyrosine (FET) that had at least two scans due to therapy follow-up and inspected their attenuation maps for tissue inversion.
Simulated tissue inversion and the effect on PET images
A subset of ten patients from the original 283 referred for [18F]-fluorodeoxyglucose (FDG) PET brain scanning for suspected dementia were selected for analysis (age: (63 ± 15) years; weight: (71 ± 14) kg; injected dose: (197 ± 3) MBq; post injection time: 50.3 ± 9.9 min). The patients all had non-inverted attenuation maps (MR-ACDixon) and fat/water inversion (MR-ACInverted) was simulated by setting all voxels with μ = 0.1 cm−1 (water) to 0.0854 cm−1 (fat) and vice versa. The PET images were reconstructed with and without MR-AC using OP-OSEM (four iterations, 21 subsets, 3 mm Gaussian post filtering) on 344 × 344 matrices (2.1 × 2.1 × 2.0 mm voxels) into the resulting PETDixon and PETInverted.
This was done to capture radial differences in errors, based on the expectation that a global underestimation of tissue will result in the largest error in the center of the brain . We report the tissue-fraction distributions and Δ% difference for each of the regions and patients.
Case study: effect outside the head and neck region
FET-PET repeat scan results
Number of patients
Both scans non-inverted
Both scans inverted
One scan non-inverted, other inverted
The primary findings of the current study are that tissue inversion of water and fat has a not negligible prevalence in Dixon MRI of the brain and head/neck (23 of 283 patients). When Dixon MRI is subsequently used for attenuation correction, tissue inversion has a large quantitative effect on the resulting PET images (up to 35% error and non-uniformly distributed). Quantitative errors of this large magnitude are important for both clinical and research use of PET/MR.
For both brain and H/N patients, there was no overlap in the fraction of water voxels between inverted and non-inverted AC maps (Figure 3). This suggests that the identification of inverted MR-ACDixon can easily be automated based on the fat fraction in the image. Visual inspection is needed in the case of partial inversions, which were observed only infrequently (2 of 283 patients). We did not analyze the data with respect to anatomical anomalies occurring post surgery or from metallic implants, except for dental implants where no correlation with tissue inversion was found.
To analyze the extent of the error introduced by the fat/water inversion, we chose to use patients without inversion in order to have a gold standard PET image as a fair comparator. The images following the simulated inversion had tissue-fraction distributions similar to the brain patients with actual fat/water inversion, although with a slightly higher fraction of fat. We trust the results to be representative despite this overestimation introduced by the simulation.
The error for a given line-of-response (LOR) depends on the attenuation coefficient error and length of the path through the inverted tissue. The LORs passing the center must on average intersect more misclassified tissue than tangential LORs near the edges. This explains the radial bias observed (Figure 4G). The bias was consistent in all ten patients.
Problems using Dixon MR-based attenuation correction in PET/MR have previously been reported , including a case of false tissue assignment to lower thorax where lung tissue is assigned the attenuation coefficient of air. The phenomenon of inversion of water and fat is also well known  and occurs due to off-resonance effects caused by inhomogeneity of the B0 magnetic field. Off-resonance effects render the identification of water and fat from in- and opposed-phase images ambiguous, thus, leading to potential inversion artifacts. This ambiguity can be resolved using a number of methods; however, without knowing the exact implementation details, the reason for the fat/water inversion can only be speculated on. The most common method is the region-growing approach where the off-resonance region is determined one voxel at a time starting from one or more seed points. Choosing incorrect seed points might lead to incorrect tissue identification. This might be the explanation in the case imaging the legs (Figure 2 A,B,C,D,E). Multiple seed points are needed since the legs are not connected and choosing incorrect seed point in one leg could explain the inversion. This is further supported by the fact that the in-phase and opposed-phase images (Figure 2A,B) have identical left and right legs, indicating that the error occurs when identifying water and fat images.
The manufacturer's user manual advises that the attenuation maps should be visually inspected prior to continuation of the scan. In the case of an error, such as with tissue inversions, the manual advises a repetition of the scan. However, in our experience (supported by the few repeat examinations reported here), the error can be reproducible, so a rescan does not necessarily provide non-inverted images. For two patients, the inversion was only present in one of the repeat scans. With an underestimation of PET uptake of 10% to 35% in the case of inversion, such cases of inconsistent inversions are of course a significant problem for follow-up examinations and response evaluation which Wahl et al.  discussed in the case of solid tumors. Hence, the problem can only be corrected for by an off-line user-implemented inversion of fat and water, as was performed here. It is expected that tissue inversion is more frequent in the case of small body parts. A systematic comparison of the frequency of the effect for different body parts is out of the scope of the present work.
The findings in this paper add further concerns to those previously voiced ,,-. Tissue inversion needs to be considered if the construction of normal PET databases is performed based on Dixon-AC or if quantitative measures for simple semiquantitative analysis such as the standardized uptake value (SUV) are being used to set cut-off thresholds or follow treatment effects. We do not recommend the use of Dixon-AC for pharmacokinetic modeling of physiological parameters in the brain . In longitudinal follow-up in, e.g., dementia using FDG, regional activity may increase or decrease and using FET for tumor imaging the metabolically active area may change configuration. The study was limited to screening a large database of head and neck patients. However, the observed tissue swap effect can occur also in areas outside the head and neck as illustrated by the anecdotal example of the tissue inversion in the upper thighs. Therefore, utmost care is needed if Dixon-AC is to be used in clinical routine.
MR-based attenuation correction is still a challenge. In this paper, we have shown that incorrect identification of fat and water classes has a strong impact on the PET images following AC. We have shown that the inversion occurs frequently and is not predictable. An automatic method to detect fat/water inversions seems possible when the full volume is inverted, but in the case of inversion in only a subset of the slices, a manual inspection of the attenuation maps is required. As the non-inverted attenuation map cannot be calculated directly from the inverted, it is important to inspect the attenuation map if the PET images should be used quantitatively. In the brain, this might also be the case even if the PET images are only read visually as indicated in our simulated inversion study.
The PET/MR system at Rigshospitalet was kindly provided by the John and Birthe Meyer Foundation, Denmark.
- Schulz V, Torres-Espallardo I, Renisch S, Hu Z, Ojha N, Börnert P, Perkuhn M, Niendorf T, Schäfer WM, Bockmann H, Krohn T, Buhl A, Günther RW, Mottaghy FM, Krombach GA: Automatic, three-segment, MR-based attenuation correction for whole-body PET/MR data. Eur J Med Mol Imaging 2011, 38: 138–152. 10.1007/s00259-010-1603-1View ArticleGoogle Scholar
- Coombs BD, Szumowski J, Coshow W: Two-point Dixon technique for water-fat signal decomposition with B0 inhomogeneity correction. Magn Reson Med 1997, 38: 884–889. 10.1002/mrm.1910380606View ArticlePubMedGoogle Scholar
- Dixon WT: Simple proton spectroscopic imaging. Radiology 1984, 153: 189–194. 10.1148/radiology.153.1.6089263View ArticlePubMedGoogle Scholar
- Martinez-Möller A, Souvatzoglou M, Delso G, Bundschuh RA, Chefd'hotel C, Ziegler SI, Navab N, Schwaiger M, Nekolla SG: Tissue classification as a potential approach for attenuation correction in whole-body PET/MRI: evaluation with PET/CT data. J Nucl Med 2009, 50: 520–526. 10.2967/jnumed.108.054726View ArticlePubMedGoogle Scholar
- Keller SH, Holm S, Hansen AE, Sattler B, Andersen F, Klausen TL, Højgaard L, Kjær A, Beyer T: Image artifacts from MR-based attenuation correction in clinical, whole-body PET/MRI. MAGMA 2013, 26: 173–181. 10.1007/s10334-012-0345-4View ArticlePubMedGoogle Scholar
- Ma J: Dixon techniques for water and fat imaging. J Magn Reson Imaging 2008, 28: 543–558. 10.1002/jmri.21492View ArticlePubMedGoogle Scholar
- Keller SH, Hansen AE, Holm S, Beyer T: Image distortions in clinical PET/MR imaging. In PET/MRI. Edited by: Carrio I, Ros P. Springer, Heidelberg; 2014:21–41. 10.1007/978-3-642-40692-8_2View ArticleGoogle Scholar
- Huang SC, Hoffman EJ, Phelps ME, Kuhl DE: Quantitation in positron emission computed tomography: 2. Effects of inaccurate attenuation correction. JCAT 1979, 3: 804–813.Google Scholar
- Delso G, Fürst S, Jakoby B, Ladebeck R, Ganter C, Nekolla SG, Schwaiger M, Ziegler SI: Performance measurements of the Siemens mMR integrated whole-body PET/MR scanner. J Nucl Med 2011, 52: 1914–1922. 10.2967/jnumed.111.092726View ArticlePubMedGoogle Scholar
- Andersen FL, Ladefoged CN, Beyer T, Keller SH, Hansen AE, Højgaard L, Kjær A, Law I, Holm S: Combined PET/MR imaging in neurology: MR-based attenuation correction implies a strong spatial bias when ignoring bone. Neuroimage 2014, 84: 206–216. 10.1016/j.neuroimage.2013.08.042View ArticlePubMedGoogle Scholar
- Wahl RL, Jacene H, Kasamon Y, Lodge MA: From RECIST to PERCIST: evolving considerations for PET response criteria in solid tumors. J Nucl Med 2009, 50: 122–150. 10.2967/jnumed.108.057307View ArticleGoogle Scholar
- Hitz S, Habekost C, Fürst S, Delso G, Förster S, Ziegler S, Nekolla SG, Souvatzoglou M, Beer AJ, Grimmer T, Eiber M, Schwaiger M, Dzezga A: Systematic comparison of the performance of integrated whole-body PET/MR imaging to conventional PET/CT for 18F-FDG brain imaging in patients examined for suspected dementia. J Nucl Med 2014, 55: 923–931. 10.2967/jnumed.113.126813View ArticlePubMedGoogle Scholar
- Dickson JC, O'Meara C, Barnes A: A comparison of CT- and MR-based attenuation correction in neurological PET. Eur J Nucl Med Mol Imaging 2014, 41: 1176–1189. 10.1007/s00259-013-2652-zView ArticlePubMedGoogle Scholar
- Ladefoged CN, Andersen FL, Keller SH, Löfgren J, Hansen AE, Holm S, Højgaard L, Beyer T: PET/MR imaging of the pelvis in the presence of endoprostheses: reducing image artifacts and increasing accuracy through inpainting. Eur J Nucl Med Mol Imaging 2013, 40: 594–601. 10.1007/s00259-012-2316-4View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.