NEMA Performance Evaluation of CareMiBrain dedicated brain PET and Comparison with the whole-body and dedicated brain PET systems

This article presents system performance studies of the CareMiBrain dedicated brain PET according to NEMA NU 2-2012 (for whole-body PETs) and NU 4-2008 (for preclinical PETs). This scanner is based on monolithic LYSO crystals coupled to silicon photomultipliers. The results obtained for both protocols are compared with current commercial whole body PETs and dedicated brain PETs found in the literature. Spatial resolution, sensitivity, NECR and scatter-fraction are characterized with NEMA standards, as well as an image quality study. A customized image quality phantom is proposed as NEMA phantoms do not fulfil the necessities of dedicated brain PETs. The full-width half maximum of the radial/tangential/axial spatial resolution of CareMiBrain reconstructed with FBP at 10 and 100 mm from the system center were, respectively, 1.87/1.68/1.39 mm and 1.86/1.91/1.40 mm (NU 2-2012) and 1.58/1.45/1.40 mm and 1.64/1.66/1.44 mm (NU 4-2008). Peak NECR was 49 kcps@287 MBq with a scatter fraction of 48% using NU 2-2012 phantom. The sensitivity was 13.82 cps/kBq at the center of the FOV (NU 2-2012) and 10% (NU 4-2008). Contrast recovery coefficients for customizing image quality phantom were 0.73/0.78/1.14/1.01 for the 4.5/6/9/12 mm diameter rods. The performance characteristics of CareMiBrain are at the top of the current technologies for PET systems. Dedicated brain PET systems significantly improve spatial resolution and sensitivity, but present worse results in count rate measurements and scatter-fraction tests. As for the comparison of preclinical and clinical standards, the results obtained for solid and liquid sources were similar.

National Electrical Manufacturers Association (NEMA) NU 2-2007 1 and NU 2-2012 2 standards constitute a set of methods under specific conditions that allow estimating the performance evaluation and comparison of Positron Emission Tomograph (PET) scanners. NU 2-2007 is expressly intended for Whole Body PETs (WB-PET) and NU 2-2012 presents minor modifications of the previous protocol. The measurements performed by these standards are the spatial resolution, sensitivity, counting rate performance, accuracy (correction for count losses and randoms) and image quality (accuracy of attenuation and scatter corrections).
In this work, we present the performance evaluation of the dedicated brain PET CareMiBrain based on NU 2-2012 and NU 4-2008 3 (dedicated to preclinical equipment). The reason for including the small animal standard is the reduced dimensions of the CareMiBrain scanner (260 mm of gantry). Finally, we compare the performance evaluation for the most used WB-PETs and dedicated brain PETs found in the literature that fulfills the NEMA procedures. Some of these studies used the standard NU-2007 and therefore it is also included in this study. Table 1 shows the phantoms used in each standard and the main differences between them. Detailed information can be found in 1-3 . Spatial resolution. The spatial resolution of a system represents its ability to distinguish between two points after image reconstruction. The measurement of spatial resolution was performed according to both standards. For the preclinical standard a 22 Na point source (a sphere with 0.3 mm diameter embedded in an acrylic cube of 10.0 mm) with an activity of 370 kBq (July 2011) was used. The source was located at the axial center of the FOV, and one-fourth of the axial FOV from the center of the axial FOV, at radial distances from the center of 0, 5, 10, 15, 25, 50, 75 and 100 mm. Each measurement took 300 seconds.
Following WB-PET PET standard performance, a capillary glass tube with 18 F-FDG (FluoroDeoxyGlucose) was used, with an inner and outer diameter of 1 and 2 mm respectively. The starting activity was 121 kBq with a length of 1 mm inside the tube. The source was located at the axial center of the FOV, and three-eighths of the axial FOV from the center of the axial FOV, at 10 and 100 mm from the center in a radial direction. Each measurement took 300 seconds. www.nature.com/scientificreports www.nature.com/scientificreports/ The experiments were repeated thrice. The acquisitions were reconstructed using 2D-Filtered BackProjection algorithm (FBP) 7 with Single Slice ReBining method (SSRB) 6 , using an energy window of 30% and were analyzed according to their NEMA standards.
Scatter fraction and count rate measurements. The objective of these measurements is to obtain system performance curves. These curves are the result of the analysis of the acquisitions in which the data are sorted into sinograms by SSRB, and processed to obtain an estimation of the true and random-scattered coincidences. With this aim and assuring certain conditions, several measures must be taken starting with an activity that guarantees the saturation of the detector until the detection of radiation is negligible. The Scatter Fraction (SF) estimation is performed with low activity rates ensuring the absence of random coincidences.
In our case, scatter fraction and count rates tests were performed following WB-PET specifications (Fig. 2c). The phantom described in this standard was placed in the center of the tomograph employing support specifically designed for this purpose, achieving the same position as indicated for WB-PET devices.
In our case, each measurement lasted 100 seconds, with a delay time of 300 seconds in between acquisitions. The initial activity of the line source was 602.5 MBq. The SF estimation following NEMA indications was performed with acquisitions in the range of 7.4-14.8 MBq. The energy window used was 50%.
Sensitivity. Sensitivity is a measure that indicates how many true coincidence events have been detected for a given source. For this purpose, a source is placed along the axial axis in order to acquire data over the entire length of the scanner. These acquired data are processed according to the standards to obtain a sensitivity value, utilizing analysis of sinograms (in case of NU 4-2008) or data extrapolation (in case of NU 2-2012).
According to NU 4-2008, the same solid source as in spatial resolution section is used to perform this measurement. In our case, the source was placed at the center of the transaxial FOV and moved along it in steps of 2 mm. The duration of each acquisition was 180 seconds.
According to NU 2-2012, we used the sensitivity phantom described in the standard. 600 seconds scans were taken with the 5 sets of aluminium sleeves with increasing thickness. The phantom was suspended in the center of the transaxial FOV, aligned with the axis of the tomograph. This measurement was repeated at 100 mm off the central axis ( Fig. 2a,b). A 700 mm length plastic tube was used and filled with water mixed with 18 F-FDG with an initial activity of 12.56 MBq for the center test and 7.16 MBq for the off-center test.
image quality, the accuracy of attenuation and scatter corrections. The main objective of image quality testing is the measurement of Recovery Coefficients (RC). The recovery coefficients evaluate the system ability to discern hot or cold lesions contained in a radioactive background, giving an idea of the reconstructed image quality taking into account corrections for scattering, random counts, lost counts, positron range and partial volume effect. The spill-over ratio (SOR) is the ratio of the mean in each cold region and the mean of the hot uniform area. WB-PET standard uses a torso-like phantom, while the preclinical standard has its own phantom. Dedicated tomographs have a very specific application and NEMA phantoms are not the most appropriate. With this motivation, we designed a custom phantom, with dimensions closer to a human head; 135 mm diameter and 103 mm height. Inside, six independent cylindrical rods with 50 mm height and diameters of 20, 15, 12, 9, 6 and 4.5 mm are placed (Fig. 3). The phantom designed is able to evaluate the recovery coefficients from 4.5 mm to 20 mm within a radioactive background covering the range of spatial resolutions provided by WB-PET scanners reported in the literature. The values of RC, SOR and their standard deviations were calculated based on the procedure described in NU 4-2008. For the measurements, the containing cylinder was filled with 5.3 kBq/ml of 18 F-FDG and the four small rods were filled at a 4:1 ratio. The rod of 15 mm was filled with non-radioactive water while the 20 mm rod contained only air. The acquisition lasted 1200 seconds and was processed using an energy window of 30%. To simulate the activity outside of the FOV, the count rate phantom was placed on the sofa with a linear source of 116 MBq (Fig. 2d). Finally, a reconstructed image of a patient from Hospital Clinico San Carlos from Madrid (Spain) is shown in this work. The acquisition took 660 seconds and the activity injected at the moment of the acquisition was 123 MBq. www.nature.com/scientificreports www.nature.com/scientificreports/ The acquisitions were reconstructed using List Mode Ordered Subsets algorithm (LMOS) 8 with 3 iterations, 12 subsets, voxel size of 1 × 1 × 1 mm 3 and virtual crystal pixel size of 2 × 2 mm 2 . The scatter correction is based on the dual-energy window method 9 , whereas the random correction follows the single rate method 10 . The attenuation correction is performed using an attenuation map generated by segmentation of the reconstructed image 11 . In the case of the phantom, two different materials were considered for attenuation map: air and tissue. However, in the patient image, three attenuation materials were considered: air, bone and tissue. No further post-processing filters were applied.   Sensitivity. The results for NU 4-2008 are shown in Fig. 5a. The sensitivity peak was 7% (for 30% of energy window) and 10% (for 50%). According to the NU 2-2012, the total sensitivity in the center was 13.82 cps/kBq and 11.05 cps/kBq (for 50% and 30% respectively), while 17.83 cps/kBq and 13.57 cps/kBq respectively at 100 mm-off-center as shown in Fig. 5b.
To compare both protocols, the average of the contributions of the 22 Na source along the whole axial axis can be considered as an approximation of the sensitivity of a line source of 154 mm (5.69% and 3.59% for the 50% and 30% energy windows respectively). If we consider the liquid source homogeneous, the activity in the 154 mm FOV can be linearly estimated (i.e. 154/700 times the original activity). The values obtained with this estimation were similar: 1.25% vs. 1.38% for the 50% window and 0.79% vs. 1.10% for the 30% window. The difference between the measurements could be reduced with a higher sampling of the 22 Na data, which would lead to a better approximation to a line source.
image quality, the accuracy of attenuation, and scatter corrections. The results for the recovery coefficients were 0.73, 0.78, 1.14 and 1.01 (from smaller to bigger rod diameters) with standard deviations of 45-46% for all rods. The SOR were 0.002 and 0.0001 with standard deviations of 12.3% and 17% for air and water respectively. The patient image is shown in Fig. 6. comparison with other tomographs. The most relevant characteristics of the brain dedicated and WB-PET systems are listed in Table 5, and NEMA results in Tables 6 and 7. There is no consensus on the image  www.nature.com/scientificreports www.nature.com/scientificreports/ quality study for dedicated brain PETs and any work found in the literature present results for the NEMA "accuracy" section. These results for WB-PETs are given in Table 8, for completeness (procedures and phantoms are in NU 2-2007 and NU 2-2012).

Discussion
Dedicated brain PETs present better results than WB-PETs in spatial resolution tests, CareMiBrain being the system that reports the best results. When both protocols are compared, the solid source results improve when compared with liquid sources (with larger dimensions). The use of 22 Na solid source is easier, cost-effective (even ecological) compared to 18 F liquid sources. For logistical reasons, we recommend a 22 Na source. Since the results are sensitive to the reconstruction algorithm (2D-FBP, 3D-FBP, rebinning, filters, etc.) it could be interesting to perform this test free of algorithmic dependence i.e., Siddon-like-back-projection 12 Table 4. Spatial resolution (NEMA NU 2-2012).  www.nature.com/scientificreports www.nature.com/scientificreports/ Regarding count rate measurements, WB-PETs present a higher NECR peak and lower SF than dedicated brain PETs in all cases, but total count rates are comparable. The scatter detection is related to the system geometry, and smaller gantry translates into greater SF. The NECR curve depends (in the denominator) on the scatter, therefore, its peak is smaller for smaller gantry systems even if the total count rates are similar. The SF for CareMiBrain is in the range of other brain PET systems and NECR and trues peaks are overcome by jPET and G-PET. All systems have used NU 2-2012 phantom for the counting rate performance test and that is the main reason to use it in the present study. Figure 6 shows a patient image from CareMiBrain system, acquired at typical injection activity for a dedicated brain PET study, far below the NECR peak measured. Usually, dedicated PETs work at lower rates than the whole-body systems due to its proximity to the organ of study, allowing a higher frequency of patient monitoring.
Signa PET and Discovery IQ report the best results for the sensitivity test, and the brainPET 4 layers MPPC DOI, followed by jPET, NeuroPET and CareMiBrain present the best results for dedicated brain PETs. Finally, the values for the recovery coefficients showed an acceptable performance for attenuation and scatter corrections with the proposed measurements.
conclusion Dedicated brain PET systems improve spatial resolution and sensitivity, but present worse results in count rate measurements and scatter fraction tests. However, the study should be re-performed for all the tomographs with a different phantom that appropriately adjusts the dimensions and characteristics of the brain to draw further conclusions for brain devices.
NEMA standards are an extremely useful tool for the comparison of different PET scanners, but, given the emerging dedicated brain PET systems, it could be interesting to redefine a standard exclusively for these