The method and some of the materials described here are adapted from NEMA Standard Publication NU 2-2001 for PET image quality and accuracy of attenuation (NEMA, 2001), which has been used for assessing image quality of PET systems and instrumentation (Daube-Witherspoon et al., 2002; Ziegler et al., 2015).
The arrays under evaluation and phantom
A NEMA Body phantom IEC-2007 (National Electrical, Manufacturers Association Washington, DC, USA) was employed to carry out all PET quantification in this work (Fig. 1a). As shown in Fig. 1 b, c and d, the three arrays used in this study are as follows: the novel 32-channel PET/MRI array (Ceresensa, Canada 2016) consisting of two parts, anterior (flexible) and posterior (fixed); the 12-channel mMR array (Siemens Healthcare Limited, Erlangen, Germany) comprised of a 6-channel body matrix (flexible anterior) and 6-channel spine-matrix (fixed posterior); and the standard 32-channel MRI array (In-Vivo Corporation, Gainesville, FL, USA) with anterior (flexible) and posterior (fixed) parts.
AC maps
In order to achieve accurate PET quantification, AC maps for all hardware within the scanner must be developed and applied during attenuation correction; this includes the NEMA phantom housing (i.e. phantom without its liquid content). Therefore, we developed all necessary hardware AC maps using a Dual Energy CT (GE Healthcare, Discovery CT750 HD, Waukesha, USA) scan of the hardware components following the method of bilinear transformation of Hounsfield units (using mono-energetic reconstruction at 70 keV and 140 keV) to 511 keV linear attenuation coefficient (Carney et al., 2006). The developed DCTAC maps were for both the empty and filled NEMA body phantom and the posterior part of each array, while the spine-matrix array was excluded, since its AC map is provided by the vendor.
For non-hardware AC maps, such as that related to the de-ionized water included in the NEMA phantom, two scenarios were examined: (1) rely on MRAC to correct for water attenuation, with the rationale that this is the scenario of a non-research/ clinical study where pre-acquired patient CT data is not normally available to the PET/MRI system, and (2) develop DCTAC map for the phantom filled with water and use it for off-line attenuation correction of the PET image, with the rationale that this is the scenario of a research setting, where patient PET/CT or CT-only data could be available for offline reconstruction.
PET phantom preparation and measurements’ set-up
The phantom background is defined to be the phantom volume without the six spherical inserts nor the lung insert. The phantom background (volume ~10.2 L) was filled with deionized water and injected with 18 F-FDG to reach an activity concentration of 5.3 ± 0.1 kBq/ml. All six spheres (with inner diameters 10, 13, 17, 22, 28 and 37 mm) were filled with 18F-FDG to reach an activity concentration of 201 ± 5 kBq/ml, making the phantom activities reach a 1:38 background-to-sphere ratio, where the background activity corresponds to a typical injected dose for total body studies (NEMA, 2001).
Markers were produced on the patient table, as well as the phantom body, and when both markers were matched, the same position of the phantom on the table was consistently achieved with ± 1 mm tolerance. During all measurements, the isocentre was selected to be at the centre line between position number 4 and 5 of the spine-matrix (marked on the patient table). This set-up ensured that no additional objects employed for positioning during the PET measurement, and hence, data are free from added errors. The set-up also ensured the elimination of attenuation variation that the patient table itself may contribute.
Image acquisition and attenuation correction
For ease of reading throughout this manuscript, the anterior part of an array is denoted with letter A while the posterior part is denoted with letter P, and therefore, both parts of an array are denoted with A&P, and part(s) may be referred to as case(s).
All PET acquisitions were performed on a 3.0-T PET/MRI system (Biograph mMR Software Version VE11P, Siemens Healthineers, Erlangen, Germany) and followed the identical workflow routine. Evaluation of one complete array required one imaging session consisting of four PET measurements, where the first measurement was started once the activity reached the concentration described above. Each measurement comprised of a Dixon MRAC acquisition (2.5 min) and a PET acquisition (10 min). The four PET measurements were carried out in the following order: (1) the complete array (A&P) and NEMA phantom, (2) the posterior part (P) of the same array and NEMA phantom, (3) the anterior part (A) of the same array and NEMA phantom and (4) the NEMA phantom with no-array. Two imaging sessions were performed for each array, yielding two data sets for data analysis.
MRAC
By MRAC, here, we refer to the PET/MR scanning conditions where only the posterior part of the array and the patient table hardware AC maps, together with MR-based AC map of the liquid in the phantom, are used for PET attenuation correction during image reconstruction. A MRAC map of the liquid portion of the NEMA phantom was acquired with the manufacturer’s standard 2-point Dixon (3D dual-echo spoiled gradient sequence) using TR/TE = 4.14/2.51 ms, Matrix size = 240 × 126 × 127 at a resolution of 2.08 × 2.08 × 2.02 mm3 and is intrinsically registered to the PET images. The AC maps of each array’s posterior part along with the NEMA phantom housing were added to the scanner. For clarity, the reconstruction of the no-array AC images on the scanner used the phantom housing and patient table hardware AC maps, together with the phantom water MRAC map. The Biograph mMR’s standard PET reconstruction algorithm, an Ordinary Poisson Ordered Subsets Expectation Maximization (OP-OSEM) (Comtat et al., 2004) approach, was applied using 3 iterations, 21 subsets and a 4-mm Gaussian filter.
DCTAC
In order to obtain the ideal AC for both the hardware and liquid portion of the NEMA phantom, the raw, nonattenuation-corrected (NAC) data of each measurement was reconstructed offline using DCTAC of the NEMA phantom including the liquid portion and other hardware DCTAC maps. All PET reconstructions were performed offline using the manufacturer’s e7-tools (Siemens Molecular Imaging, Knoxville, USA). The PET images were reconstructed using the same algorithm and parameters as described under the MRAC section above. The PET images’ output matrix size was 344 × 344 × 127 to achieve ideal reconstruction modality (Ziegler et al., 2015).
Data analysis
All data were corrected for the PET decay, due to the time used for hardware changing and sequence set-up, scatter and attenuation, before carrying out the analysis. All image processing and data analysis were computed using Matlab 9.3.0 (The MathWorks, Natick, MA, USA).
NAC data
NAC data are important in this analysis since it does not include any errors that an AC process might introduce. From NAC data, the global mean and standard deviation of each masked image (slice) were estimated. This mean and standard deviation were estimated for all cases of the three arrays, as well as for the no-array. The no-array data was considered the frame of reference that each part of an array was compared to. Therefore, the RPD between no-array and each part of an array was estimated and reported using equation 4 as reported in (Farag et al., 2019) and presented here for convenience, where v1 is the no-array case and v2 is any given case (A, P, or A&P).
$$ \mathrm{RPD}=\frac{v_1-{v}_2}{\ 0.5\ \left({v}_1+{v}_2\right)}\ 100\% $$
AC data
According to the NEMA NU 2–2001 standard, PET image quality can be analysed from parameters such as CR, BV and CNR. Therefore, each sphere’s CR, BV and CNR were estimated from the relationships described in (NEMA, 2001), and the mean and standard deviation for each array and for no-array were computed and reported. Relative percentage difference maps between PET images from no-array and images from each array part were computed. PET image quality parameters were estimated for both scenarios of AC data, images reconstructed by the scanner and images reconstructed offline using e7-tools. Statistical evaluation of the resulting means used a one-factor analysis of variance (ANOVA) with ‘array’ as the between-subject factor (alpha = 0.05). In the null hypothesis, no significant difference exists between all three arrays (equal means), and in the alternative hypothesis, means are significantly different with no a priori direction. The test was applied on the data from each array part (A, P and A&P), and the degree of freedom, F-factor, and p values are reported.