A 49-year-old patient with a history of acute myeloid leukaemia and subsequent hematopoietic stem cell transplantation (five weeks prior to the scan) underwent a whole-body [18F]FDG PET-MRI scan in order to investigate a suspected gastrointestinal graft-versus-host disease (GVHD). This was performed on a Biograph mMR system (Siemens Healthineers, Erlangen, Germany) one hour after i.v. injection of 275 MBq [18F]FDG. Scanning was performed in a supine position with the arms lying next to the patient’s torso. Dedicated RF coils, either being corrected for in PET AC by templates (mMR head/neck coil, mMR spine coil) or optimised for minimal PET attenuation (mMR body coils), were used for MR imaging (total imaging matrix technology, Siemens Healthineers, Erlangen, Germany). A standard two-point Dixon sequence (pre-Gadolinium) was performed for AM generation, resulting in AM images of four different tissue classes (air, lung, fat-, and water-like soft tissue). Due to transaxial MR field-of-view truncation, attenuation of the arms was estimated using PET emission data by applying the maximum likelihood reconstruction of activity and attenuation algorithm (MLAA). PET imaging was performed using 6 bed positions with acquisition durations of 3 min for each bed position. PET raw data were corrected for randoms, scatter, and attenuation, and reconstructed using a 3D ordinary Poisson ordered subsets expectation maximisation (OP-OSEM) algorithm with 3 iterations and 21 subsets, followed by a 5 mm full-width-at-half-maximum (FWHM) Gaussian filter. Acquired MR data involved transaxial T1-weighted VIBE and T2-weighted HASTE sequences as well as coronal T1-weighted VIBE sequences post-Gadolinium.
Inspection of the determined AM demonstrated that most parts of the liver (approximately 82% as determined by volumetric analysis of MR and AM images) had not been identified as water-like soft tissue (with assigned linear absorption coefficients of 0.1000 cm−1), but as lung tissue (0.0224 cm−1) (Fig. 1a). This in turn led to highly suppressed PET activity values of the liver in the reconstructed images with an apparent SUVmean of the liver amounting to 0.6, far outside a typical reference range of 1.4–3.2 (Boktor et al., 2013) (Fig. 1b). The acquired Dixon sequence-based in-phase image (Fig. 2a) demonstrated markedly reduced signal intensities of the liver, similar to those of lung tissue. This translated into comparatively low values in the calculated water image (Fig. 2b). As the liver is located right next to lung tissue, it was misinterpreted as lung since the gradient of image values between lung and liver was not large enough for the connected component segmentation algorithm to handle this properly.
The loss in liver signal intensity was also clearly visible in the T1- and T2-weighted images, and was retrospectively attributed to secondary hemochromatosis caused by a history of multiple blood transfusions due to acute myeloid leukaemia. This led to elevated levels of iron in the liver (iron overload). Due to their superparamagnetic properties, these iron deposits caused increased relaxations of proton spins, hence suppressing MR signals (Queiroz-Andrade et al., 2009).
PET images revealed elevated tracer uptake levels in the rectum, the sigmoid colon, as well as the ascending and descending colon (Fig. 3), indicative of GVHD in accordance with a study by Stelljes et al. (Stelljes et al., 2008). However, as tracer uptake in the liver serves as a major reference in [18F]FDG PET-based diagnosis of gastrointestinal GVHD (Stelljes et al., 2008), a more detailed qualitative assessment of GVHD beyond simple SUV determination proved to be difficult in this case.
Manual segmentation of the lung-liver border in the AM along the visible border in the Dixon water image (Fig. 2b), assigning a soft-tissue value of 0.1000 cm−1 (Fig. 1a), and reconstructing the PET data with this modified AM resulted in PET images with visually typical [18F]FDG uptake in the liver (Fig. 1b, Fig. 3). Liver SUVmean then amounted to 3.1, well within a normal range of values.