Phantom experiment
A human left femur was obtained from an anatomical lab. The distal third was sawn away to fit the remainder into a 42 by 24 by 15 cm polypropylene container. A neck osteotomy was performed and the marrow cavity was reamed to fit the femoral stem of a prosthesis composed of a chrome-cobalt alloy (Fig. 1). The femur was covered in cellophane tape to ensure structural integrity. The titanium acetabular prosthesis component was taped over the ceramic joint in its correct anatomical position using Mefix surgical tape. 1.5cc Eppendorf tubes (Eppendorf, Aarschot, Belgium) were filled with a solution of 2.975 MBq in 0.2 l water (or 14.88 MBq/l), so as to obtain an approximate target to background ratio of 10:1. They were fixed to the assembly by surgical tape in five positions: at the medial and lateral border of the cup, at the superolateral and inferomedial borders of the prosthesis neck and at the greater trochanter (Fig. 2). These locations were inspired by the Reinartz patterns of prosthetic infection [16, 17].The whole assembly was immersed in an aqueous solution of 7.5 MBq of 18F-FDG in slightly more than 5 liter in the polypropylene container (the exact volume was unknown because some water had to be added to ensure complete immersion of the femur). A cylindrical acrylic uniformity phantom of 20 cm diameter and 18 cm height was positioned in front of the plastic container. It was filled with a solution of 8.31 MBq in 5.701l, so as to simulate abdominal/pelvic background. Fig. 3 shows the experimental setup.
Image acquisition was performed in the same PET/CT scanner as used for clinical imaging and the scanning parameters both for PET and CT were identical (see below).
Next, the prosthesis was extracted from the femur, the original femur head was taped to the diaphysis and the Eppendorf tubes were each taped to the femur in similar positions as for the first experiment, although we had to take into account the larger dimension of the anatomical neck as opposed to the prosthesis neck and the smaller surface area of the femoral head. To keep the femur immersed it was taped to the bottom of the container. An acquisition with identical parameters as the first one was subsequently performed.
Image reconstruction was the same as for patient data (see below).
On the PET images corrected for attenuation by MAR CT, volumes of interests (VOIs) were positioned on the 5 radioactive sources surrounding the femur using autocontour software with a relative threshold of 42% (Volume Viewer, GE Healthcare, Milwaukee, USA). These were then cloned to the PET attenuation corrected by the CT without MAR. In each VOI the mean and maximal activity (kBq/ml) were measured. This was repeated twice: for the PET data with and without the total hip prosthesis inserted into the phantom. Activities on the second acquisition were corrected for radioactive decay between the acquisitions.
Patients
Forty-six 2‐[18F]fluoro‐2‐deoxy‐D‐glucose (18F-FDG) positron emission tomography/computed tomography (PET/CT) scans were retrospectively selected from all clinically indicated whole-body 18F-FDG PET/CT scans performed at the department of nuclear medicine in AZ Sint-Jan Bruges (Belgium) between October 30, 2018 and July 05, 2019 on basis of the following criteria: 1. presence of at least one total hip prosthesis, to the exclusion of resurfacing prostheses and short stem prostheses. Hip screws were also excluded from the study; 2. absence of registration artefacts due to patient movement between PET and CT and 3. availability of correctly saved raw PET and CT data. When a patient underwent multiple 18F-FDG PET/CT scans during the inclusion period, only the first one was taken into account.
Most of the patients (=41) selected underwent PET for staging or follow-up of malignancy. In 2 patients a periprosthetic hip infection was suspected. In one patient PET was performed to elucidate a lung consolidation; two patients had nonspecific constitutional symptoms. Fifteen patients had bilateral hip prostheses, for a total of 61 hip prostheses analyzed. Patients aged between 60 and 91 years (mean 73, standard deviation 7.7 year); their BMI ranged from 17.7 up to 39.3 (mean 26.4, standard deviation 5).
PET/CT imaging
All PET/CT examinations were performed on a Discovery MI 15 cm axial field-of-view PET/CT camera (GE Healthcare, Milwaukee, USA) [21]. Patients fasted for at least 6 h before 18F-FDG injection and had blood glucose confirmed to be below 200 mg/dl before injection. The amount of tracer administered was based on the body mass index of the patient (BMI < 20: 1.5 MBq/kg; BMI 20-26.5: 2 MBq/kg; BMI>26,5: 2.5 MBq/kg). 18F-FDG was injected intravenously under standard conditions. Imaging was started after rest for 60 minutes in a comfortable position. Patients were positioned in the scanner with their arms raised.
PET consisted of 7 to 9 bed positions of 2 minutes duration each, from the vertex to the mid-thigh. Reconstruction used a three-dimensional ordered subset expectation maximization (OSEM) algorithm (4 iterations, 8 subsets, gaussian post-filtering 6.0 mm full width at half maximum, heavy Z-axis filter, matrix size 256 × 256, slice thickness 2.5 mm) available on our system (VPHD, GE Healthcare, Milwaukee, USA) with point spread function correction (Sharp IR, GE Healthcare, Milwaukee, USA) and time-of-flight (VUE Point FX, GE Healthcare, Milwaukee, USA).
CT used a tube voltage of 120 keV and Smart mA automatic exposure control (GE Healthcare, Milwaukee). Intravenous contrast (Xenetix 350, Guerbet, France) was used depending on the clinical indication. For metal artefact reduction the GE Smart MAR algorithm was used [15]. CT reconstructions were made with and without MAR. PET data were attenuation corrected using these MAR and nonMAR reconstructed CT data.
Image analysis
Analysis of PET images was performed using the Volume Viewer software on the Advantage Workstation (GE Healthcare, Milwaukee, USA). PET and CT datasets were spatially registered and reoriented so that the coronal plane corresponded as good as possible to the midplane of the prosthesis. Two-dimensional regions of interests (ROIs) were drawn manually in six locations around the prosthesis on the MAR corrected CT images: on the acetabulum, medially and laterally in the neck region, on the prosthesis shaft, and medially and laterally on the femoral diaphysis. The shaft of the prosthesis served as a control (Fig. 4). Afterwards, these regions were copied to the other datasets. In each ROI, mean and maximal standardized uptake values (SUVmean and SUVmax) (g/ml) as well as the standard deviation of the SUV were measured. The SUV was calculated as the activity concentration in the PET image divided by the injected activity per g of body weight. A three-dimensional volume of interest was drawn over the urinary bladder using the autocontour software with a relative threshold of 42%.
Statistical analysis
Two-way repeated measures ANOVA was performed to compare SUVmean in the various ROIs surrounding the prosthesis in images reconstructed using MAR or without using MAR. Greenhouse-Geisser correction was applied in case of violation of the sphericity assumption. Significant ANOVA-testing was followed by pairwise comparisons between groups using t-tests with Bonferroni adjustment.
In each region, the relative change of SUVmean on using MAR-corrected CT for PET reconstruction, was calculated as (SUVmean in MAR-PET – SUVmean in nonMAR-PET)/(SUVmean in nonMAR-PET). The sign test was used to test in each ROI whether the median relative change was significant. The lower limit of the one-sided confidence interval based on the sign test was reported. Differences of the relative change between regions were tested by robust ANOVA with 20% trimmed means and 2000 bootstrap samples. Robust ANOVA was followed by robust posthoc tests using 20% trimmed means and 5000 bootstrap samples, and corrected for the number of tests. All testing was performed two-sided.
Correlation plots between SUVmean in MAR and nonMAR PETs were constructed using the data from all regions in all patients. Bland-Altman plots were constructed using logtransformed data, because the differences between MAR and nonMAR SUVmean increased with increasing SUVmean.
Similar analyses were performed on SUVmax and on the coefficient of variation of the SUV.
In 45 patients, bladder volume and SUV data were compared between PET reconstructions with MAR-corrected CTs and those with uncorrected CTs. One patient was excluded from this analysis, because no substantial bladder activity was present owing to an indwelling catheter. The sign test was used for the comparison.
The relative changes of the bladder parameters on using MAR-CT versus conventional CT for PET attenuation correction were calculated. The sign test was used to test whether these were significant. Wilcoxon’s rank sum test was used to compare the relative changes between patients with unilateral and bilateral prostheses.
Significance was called when P was less than 0.05.
All statistical testing and graphics was performed in R version 4.0.1 [22] and figures were produced using the package ggplot2 [23]. Robust statistical tests were performed by the WRS (Wilcox’ Robust Statistics) package [24].