PET/MR Imaging in the Detection and Characterization of Pulmonary Lesions: Technical and Diagnostic Evaluation in Comparison to PET/CT

Patient Population

In this study, 40 randomly selected patients (23 men, 17 women; mean age ± SD, 53.2 ± 13.1 y; age range, 18–79 y), routinely referred to our institute for clinical ¹⁸F-FDG PET imaging for staging or follow-up of malignant or suspected malignant disorders, were included. Patients with the following medical characteristics were recruited between July 2011 and August 2012: bronchial carcinoma, n = 2; breast cancer, n = 4; thyroid cancer, n = 5; lymphoma, n = 3; head and neck cancer, n = 20; carcinoma of unknown primary, n = 1; testicular cancer, n = 1; adrenal gland cancer, n = 1; primary liver tumor, n = 1; rectal cancer, n = 1; and liver hemangioendothelioma, n = 1.

The local institutional review board approved this study. Written informed consent was obtained from all patients. Inclusion criteria were informed consent, ability to undergo another scan after the PET/CT examination, and clinical indication for an ¹⁸F-FDG PET/CT scan. Exclusion criteria were pregnancy, age below 18 y, and contraindications for MR imaging examinations (e.g., magnetic metal implants, pacemakers).

Imaging Protocol

All subjects underwent a single-injection dual-imaging protocol. After completion of the PET/CT scan, patients were subsequently positioned on the PET/MR imaging scanner with the smallest possible temporal delay.

PET/CT images were acquired according to standard clinical protocols, as previously reported (10) on a Biograph Sensation 64 PET/CT scanner (Siemens Medical Solutions) (18). Patients fasted for at least 6 h before scanning, and blood glucose levels were measured just before injection to ensure a value below 150 mg/dL. Patients were injected with 383 ± 35 MBq of ¹⁸F-FDG intravenously depending on their body weight, and the acquisition was started at 87 ± 35 min after injection, moving from the pelvis to the head. Acquisition time was 2 or 3 min per bed position, with 5–6 bed positions (each 21 cm) covering the trunk of the patients. For attenuation-correction purposes and anatomic correlation, a low-dose CT scan (120 keV, 20 mAs, no intravenous contrast) was acquired in shallow expiration. When clinically indicated, a diagnostic CT (120 kV, 240 mAs, 0.5 s per rotation, 5-mm slice thickness, portal venous phase 80 s after the injection of 80–120 mL intravenous contrast agent [Imeron 300; Bracco Imaging Deutschland GmbH]) was performed. In these cases, the diagnostic CT was used for attenuation correction. For evaluation of the lungs, a separate low-dose CT of the chest in deep inspiration was performed in all patients (25 mAs, 120 kV, 0.5 s per rotation, 5-mm slice thickness). CT images of the chest were reconstructed with a standardized window setting for lung parenchyma visualization (window level, −450 Hounsfield units [HU]; window width, 1,800 HU) using a sharp convolution kernel. Nonoverlapping images were reconstructed with a slice thickness of 5 mm.

All PET/MR imaging examinations were performed using a fully integrated whole-body hybrid PET/MR imaging system (Biograph mMR; Siemens Healthcare). The technical specifications have been described previously (19–21).

On average, PET/MR imaging was started 125 ± 25 min after injection. The combined imaging protocol was as follows: after obtaining a localizer MR image to define the bed positions, combined PET/MR imaging was initiated with 3–5 bed position at 4-min acquisition times per bed position. First, a coronal 2-point Dixon 3-dimensional VIBE T1-weighted MR image sequence was acquired in each bed position in shallow expiration, which was used for the generation of attenuation maps (9). The technical parameters for this sequence are summarized in Table 1. To minimize artifacts from incomplete breath-hold, a centric k-space acquisition was chosen (22). Together with the start of this Dixon MR image sequence, the PET acquisition started simultaneously in the same bed position, thus ensuring optimal temporal and regional correspondence between MR imaging and PET data. In the thorax and abdomen region, the MR images were acquired during breath-hold in end-expiratory phase. An additional fat-suppressed contrast-enhanced 3-dimensional VIBE pulse sequence was performed over the lung in deep inspiration with the sequence parameters summarized in Table 1.

Data Processing and Image Reconstruction

Identical reconstruction and correction algorithms for PET data obtained on the PET/CT and on the PET/MR imaging scanner were used. For both modalities, emission data were corrected for randoms, dead time, scatter, and attenuation. A 3-dimensional ordered-subset expectation maximization iterative reconstruction algorithm was applied, with 3 iterations and 21 subsets, 4 mm in full width at half maximum gaussian smoothing, and a zoom of 1. Attenuation maps were obtained from the CT data by bilinear transformation, as implemented in the postprocessing software of the PET/CT scanner and used for attenuation correction of the PET/CT data, as previously described (11).

For a subset of patients with ¹⁸F-FDG–positive lesions, the PET data from PET/MR imaging were reconstructed using the HUs from normal lung from the corresponding CT data of the patient instead of the fixed μ value of 224 normally used in PET/MR imaging for the lungs.

MR Imaging–Based Attenuation Correction

For attenuation correction of the PET data from the PET/MR imaging, attenuation maps were generated as recently published (9,10). The lungs were identified by connected-component analysis of the air in the inner part of the body. By application of a morphologic closing filter, virtual air artifacts induced by the absence of an MR imaging signal in cortical bone, heart, and aorta (because of blood flow) were corrected (12).

Image Analysis

Images were analyzed by 2 readers (1 nuclear medicine physician and radiologist and 1 radiologist) masked to the patient history. The different datasets of each patient (CT, T1-weighted VIBE Dixon sequence, T1-weighted VIBE fat-saturated sequence, PET of PET/MR imaging, and PET of PET/CT) were evaluated with at least a 2-wk interval. The entire rating procedure was performed on a dedicated workstation using software (Syngo MMWP [workstation] and Syngo TrueD [software]; Siemens Medical Solutions).

The ¹⁸F-FDG PET data from both modalities and CT, Dixon, and VIBE sequences were rated for every patient with regard to overall image quality: 0, non-diagnostic image quality; 1, low image quality (distinct artifacts, strong image noise); 2, satisfying image quality (little artifacts, moderate image noise); and 3, excellent image quality (no artifacts, low image noise).

Subsequently, in a first step, the detection rate of pulmonary lesions (benign or malignant) in the 2 different MR image sequences (Dixon and VIBE sequences) and both PET datasets (including both the non–attenuation-corrected and the attenuation-corrected PET) was assessed (0, absence of a lung lesion; 1, presence of one or several lung lesions), with CT being the standard of reference. If a lung lesion was found, its dignity was rated on a 3-point-scale (1, most likely unspecific/benign; 2, indeterminate; and 3 suggestive of malignancy) in each morphologic and both PET datasets. In addition, in every morphologic dataset its size in millimeters was measured. For rating of the PET/CT data, PET images were screened for any suspected focal uptake. The standard of reference was based on the CT scan including both past and follow-up examinations as well as medical history and all other clinical information.

Quantitative Assessment

For quantitative comparison of the PET and CT data, an SUV analysis of tracer uptake in the suspected lesions and in the right and left normal lung was performed. Volumes of interest were placed over matching corresponding ¹⁸F-FDG images of suspected lung lesions. To calculate SUVs, an isocontour volume of interest including all voxels above 50% of the maximum was created covering the whole lesion volume. Within all volumes of interest, mean SUV (SUV_mean) and maximum SUV (SUV_max) were measured. For areas of normal lungs, regions of interest (3-cm diameter) were placed in the right and left lung, respectively, in areas without pathologic abnormalities. For CT scans, additional measurements of HUs of pulmonary lesions and HUs of normal right and left lung parenchyma were performed.

Statistical Analysis

Statistical analysis was performed using the MedCalc software package (version 7.2.0.2., for Windows [Microsoft]). Descriptive statistical evaluation was performed; differences in image quality and size of lung lesions were compared using a 2-sample Student t test. Differences were deemed to be significant at a P value of less than 0.05. In addition, the mean SUV_mean and SUV_max of PET/CT and PET/MR imaging were correlated (correlation coefficient R, P value, and 95% confidence interval [CI]), and a Bland–Altman plot was performed. To analyze the diagnostic performance of both modalities, the detection rate was calculated on the basis of the total lung lesions determined by the standard of reference.