Performance evaluation of the PET component of a sequential APD-based micro-PET/MR imaging system

The LabPET/MR imaging system is a new preclinical prototype with sequentially-acquired PET/MR scan capabilities. The dual modality system is composed of a modified APD-based LabPET4 subsystem, coupled with a cryogen-free superconducting 3.0 T MRI subsystem. The objective of the study was to characterize the imaging performance of the PET subsystem and identify the change in PET performance due to the presence of the magnetic field. NEMA NU-4-2008 performance characteristics were conducted with the MR subsystem switched ON and OFF. Influence of the temperature was also investigated by repeating selected performance tests. In the ON configuration, the radial/tangential FBP spatial resolution ranged from 1.62 ± 0.08/1.88 ± 0.08 mm to 2.83 ± 0.11/2.03 ± 0.03 mm at the axial center field of view. The peak absolute sensitivity for an energy window of 250–650 keV was 0.87%. The maximum noise equivalent counting rates were 133 kcps at 3.8 MBq ml−1 and 25 kcps at 0.4 MBq ml−1 for the mouse- and rat- like phantoms, respectively. The corresponding scatter fractions were 13% and 23%. No differences could be identified between the ON and OFF configuration, except for a mean loss in peak absolute sensitivity of 5.1 ± 0.5% which was attributed to a mean rise in temperature of 1.2 °C when the magnet was ON. Compared with the LabPET standalone scanner, the LabPET/MR imaging prototype is working in a warmer environment and this impacts on the sensitivity of the APD-based LabPET scanner.


Introduction
In preclinical studies, combined PET and MR imaging (MRI) are now gaining increasing interest in all areas of molecular imaging: oncology, cardiology, neurology, infectious disease (Wehrl et al 2014). There is a clear opportunity to combine the unique information attainable with each modality: high sensitivity and accurate quantification of PET functional information with high spatial resolution and good tissue contrast from morphologic MR. In recent years, the main challenge has been a technical one, especially due to the fact that photomultiplier tubes, (PMTs), commonly used in photo-detection in nuclear medicine, were extremely sensitive to even weak magnetic fields and might contain conducting components that could interfere with the MR system (Pichler et al 2008, Lecomte 2009).
Two main types of solid-state photodetectors have been selected as a replacement of PMTs. Firstly, avalanche photodiodes (APDs) are very compact photodetectors insensitive to magnetic fields up to 9.4 T (Pichler et al 1998), but with internal gain and timing properties inferior to those found in PMTs (Lecomte 2009). Secondly, silicon photomultipliers (SiPMs) are being increasingly used because of their many attractive properties for implementing integrated PET/MR designs. SiPMs exhibit insensitivity to magnetic fields even higher than APDs, an internal gain similar to PMTs, a low operating voltage and a fast response (Lecomte 2009).
However, a key drawback of these semiconductor sensors is their high sensitivity to temperature changes (Lecomte 2009, Seco et al 2014. For example, APDs are known to have a gain factor strongly dependent on ambient temperature and therefore require adequate Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. temperature stabilization for APD-based PET scanners (Keereman et al 2013). If not properly compensated, these temperature variations might impact the imaging performance of the system through changes in sensitivity and a normalization calibration inadapted to the temperature state of the detectors.
The first prototype MR-compatible PET scanners were introduced in the mid-1990s (Christensen et al 1995, Shao et al 1997. In those early days, PMT performance degradation was eliminated by using very long optical fiber connections between the scintillators and the PMTs. Later on, prototypes were developed using APDs with or without an optical-fiber connection to the scintillators (Catana et al 2006, Judenhofer et al 2007, Maramraju et al 2011. Since then, considerable efforts have been made to propose integrated PET/MR hybrid systems (for a review, see Vandenberghe and Marsden 2015). Integrated PET/ MR hybrid systems have been developed commercially in a clinical setting (Delso et al 2011, Grant et al 2016, as well as prototypes in the preclinical environment (Wehner et al 2015). Integrated systems do allow simultaneous PET and MR acquisitions with shorter acquisition times. However, these systems require the development of a completely new MR compatible PET technology. In a preclinical environment, PET and MR scans acquired sequentially represent a simple, straightforward method to realize spatially registered PET/MR scans as small animals are usually anesthetized during acquisitions and are less likely to move between the two acquisitions. Moreover, interactions between both scanners are reduced to a minimum. One commercial PET/MRI preclinical system with a PMT-based PET scanner mounted in line with a 1 T permanent magnet MR has already been described (Nagy et al 2013). The aim of our work was to assess the performance characteristics of an APD-based micro-PET/MR in-line prototype. The PET component was an adaptation of the LabPET4 scanner . Changes in performance of the PET component due to the proximity of the MR device were investigated. The PET performance characteristics were also compared with the published version of the original standalone LaBPET4 scanner (Bergeron et al 2014).

System description
The PET component of the system is based on the LabPET4 standalone system (Trifoil Imaging, Chatsworth, CA, USA), whose main physical characteristics are reported in table 1 (Bergeron et al 2009. Briefly, detectors are composed of two different types of scintillators, optically coupled side-by-side in the axial direction, to form phoswich pairs (figure 1). Each phoswich pair is readout by one single APD to minimize signal multiplexing and reduce the number of electronic channels. Crystal identification in the phoswich pair is performed by advanced digital signal processing methods dedicated to the specific architecture of the LabPET system (Fontaine et al 2009). This unique architecture (phoswich assembly along the axial direction, readout by one APD) has been shown to yield to a very good ratio of spatial resolution per crystal size (Goertzen et al 2012). However, this 1.2 (2D) or 0.6 (3D) Figure 1. Schematic drawing of the LabPET4/MRI scanner. The labeled components are the technical support unit (1), which is common to both modalities, the dockable LabPET ring and new associated electronics (2), and the MRI system (3).
architecture is not to be confused with a phoswich in the transverse direction, aimed at correcting depthof-interaction (DOI) effects in the transaxial direction. Four phoswich detectors, enclosed in a hermetic package, are stacked in the transverse plane to form one module and forty-eight of these modules are used to form a 162 mm ring diameter. In the LabPET4 version, eight layers of modules (or 16 rings of detectors) are axially stacked to reach a 3.75 cm axial field of view (FOV). Some minor improvements have been implemented to accept the docking of a MR component in the prototype. Firstly, the electronics have been repackaged along the axial axis so as to reduce the size of the LabPET system and to provide better shielding against PET/MR interferences (figure 2). Secondly, a new cooling system has been designed to match the new layout of the electronics and the new system packaging. The cooling system consisted in flowing heated forced air through the detector and front-end electronics. Thirdly, the power supply has been made simpler and more compact. The LabPET has been coupled with a 17 cm bore, cryogen-free superconducting 3.0 T MRI subsystem (MR solutions, Guilford, UK, Europe). This 'dry' magnet uses a compression/expansion cycle of a small quantity of helium gas to create 'mechanical' cooling of the gas without changing its gaseous state. The specific architecture of the magnet, with the elimination of the helium cooling system, has allowed the optimum installation of an additional solenoid, which reduces the stray field to only a few centimetres. The magnetic field can be easily switched ON and OFF without any difficulty and associated costs. Typical discharge/charge cycle at 3 T was less than two hours.
The technical support unit (Minerve, Esternay, France) is common to both modalities and allows a smooth, precise, motorized positioning over 60 cm length between PET and MRI systems. The control software provides a unified interface to manage both modalities.

Evaluation of the micro-PET
Performance characteristics were evaluated according to the National Electrical Manufacturers Association (NEMA) NU4-2008 report (NEMA 2008). To assess the influence of the presence of a MRI system close to the micro-PET in the in-line system, these measurements were carried out for two different configurations. The ON configuration referred to the normal operating conditions of the prototype, with the magnet operating at 3 T. The OFF configuration referred to the magnet being completely discharged and served as a baseline for the micro-PET performance. The ON and OFF measurements were performed sequentially in a short time delay, with the PET calibrated for normal operating conditions (ON). Preliminary results showed that the room temperature change between both configurations remained within manufacturer specifications (±1°C) for temperaturedependent PET calibrations. The short time delay (within two weeks, one week for each configuration) was chosen to allow the measurements to be independent of performance drifts of the PET detectors or external environmental changes which might impact the temperature adjustment of the air-conditioning system. The room temperature was monitored during all experiments using two temperature and humidity loggers (Klimalogg Pro, TFA Dostmann, Wertheim, Germany). The first one was positioned next to the PET gantry and the second one was placed 2 m away from the gantry. All the measurements were performed with the default coincidence time window of 22 ns and the default energy window of 250-650 keV.
All the post-processing tasks were performed using ImageJ (National Institutes of Health, Maryland, USA) (Rasband 1997).

Temperature dependence and calibration of the scanner
Temperature dependence of the LabPET scanner has been thoroughly investigated in the literature (Keereman et al 2013). It has been proposed that, when the room temperature changes, both the scintillation light yield of the crystal and the gain factor of the APD are modified. The pulse amplitude resulting from the interaction of a gamma photon in the detector at a given modified temperature is then different from that obtained at the calibration temperature. As the change is proportional to the energy of the photon, the total energy spectrum is uniformly scaled along the energy axis, shifting the photopeak towards either end of the energy window and so leading to different photon counts recorded inside the energy window. For the LGSO and LYSO detectors of the LabPET standalone system, Keereman et al showed that the expected average peak drift per°C was respectively 6.85±1.78% and 5.10±1.53%. When the room temperature change is outside the manufacturer recommendations, the PET detectors need some energy re-calibration, a process that is automated for each individual crystal and last only a few minutes on the LabPET scanner (Tetrault et al 2008. To check if this calibration could impact our results even on a small temperature change, albeit within manufacturer specifications, additional measurements were conducted on some selected NEMA tests detailed below to better characterize the differences observed between the two configurations ON and OFF. 2.4. PET measurements 2.4.1. Spatial resolution Spatial resolution was measured using a 0.25 mm diameter 22 Na point source embedded in a 1 cm 3 acrylic cube (Eckert & Ziegler Isotopes Product, Valencia, CA). The source had an activity of 0.46 MBq and was carefully positioned at the four transaxial positions and the two axis locations of the NEMA NU 4-2008 report. Data were collected for 90 s in each position resulting in more than 10 5 prompt counts per measurement. Sinograms were sorted using singleslice rebinning (SSRB), with data sampled from one ring difference and reconstructed using the 2D filtered backprojection (FBP) algorithm implemented in the LabPET scanner. The voxel size was 0.25×0.25×0.6 mm 3 , which was the setting of the high-resolution (HR) mode of the scanner. No DOI correction was taken into account. Random count rates were estimated in a delayed window and then subtracted from the prompt rates before starting the reconstruction. Normalization was performed using a direct data-based method with a 10 MBq 68 Ge rod source, rotating around the detectors during almost 5 h. To ensure statistical significance between the ON and OFF configuration, measurements were repeated three times in each configuration. These measurements were obtained at different times over a period of 3 months and after calibration of the detectors at the ambient temperature. Radial and tangential spatial resolution were reported as full width at half maximum (FWHM) and full width at tenth maximum (FWTM) obtained from slice profiles fitted with a Gaussian function. Statistical significance between both configurations was evaluated by multiple t test using GraphPad Prism version 6.02 for windows (GraphPad Software, La Jolla California USA, www. graphpad.com).
Iterative algorithms are usually more adapted to take into account detector designs in micro-imaging studies. This is particularly true for the LabPET scanner since its design is characterized by irregular crystal spacing in both azimuthal and axial directions (Bergeron et al 2014). Therefore, spatial resolution was also assessed visually using an Ultra Micro Hot Spot phantom filled with a solution of 18 F. Also, this made it possible to have an activity distribution more representative of that obtained in small animal imaging studies. For each configuration (ON/OFF), the phantom was imaged three times during decay, starting with an initial activity of 115 MBq. The acquisition time was set to 30 min at each time point measurement. The reconstruction was performed using 3D MLEM reconstruction (10-200 iterations, all oblique incidences included) operating in HR mode. Random count rates were estimated in a delayed window and then corrected for, inside the loop of the iterative reconstruction. Normalization was performed with the rotating 68 Ge rod source. No attenuation and scatter corrections were applied.

Count rate performance and scatter fraction
Scatter fraction and counting rate performance were carried out according to the NEMA NU4-2008 methodology (NEMA 2008) for the ON and OFF configurations. Two phantoms were used: the mouse phantom (70 mm long, 25 mm diameter) and the rat phantom (150 mm long, 50 mm diameter). The line source was 1 cm shorter than the phantom's length and was initially filled with 400 MBq and 350 MBq of 18 F activity in the rat and the mouse phantoms, respectively. Then, the phantoms were scanned until the activity decayed to less than 0.2 MBq. Intrinsic scanner counting rate was assessed by performing a 12 h acquisition with each phantom without any radioactivity. For all acquisitions, prompt and random 2D-sinograms were generated using the SSRB method. Data processing was performed in accordance with the NEMA NU4-2008 specifications using an in-house imageJ plugin (Rasband 1997). For each configuration, system event rates, peak noise-equivalent counting rate (NECR), the activity at which the NECR peak occurs and low-counting rate scatter fraction were reported.

Sensitivity
The absolute sensitivity was assessed using the same 22 Na point source used for spatial resolution determination. For each configuration (ON/OFF), the source was carefully centered in the axial and the transaxial directions, within ±0.5 mm, and then stepped in 1.2 mm increments along the entire axial FOV. Counts were acquired over 20 s to ensure the collection of at least 50 000 true events at the center of the scanner. A background scan was also acquired over the same interval to take into account intrinsic radiation of the detectors. Data were rebinned into 2D-sinograms using the SSRB method and processed in accordance with the NEMA NU4-2008 specification using an inhouse imageJ plugin (Rasband 1997). Absolute sensitivity was calculated at each position by summing the counts over all the masked slices and by dividing the background-corrected counts by the acquisition time, the activity and the branching ratio of 22 Na. For each configuration, total absolute system sensitivity of the scanner over the entire axial FOV and peak absolute sensitivity were reported. Average absolute sensitivity was calculated as the average absolute sensitivity over the entire axial FOV.
To better characterize the relationship between sensitivity and ambient temperature in the two configurations ON and OFF, a second experiment was performed. In this experiment, sensitivity was measured in each configuration after a 24 h temperature stabilization period and after re-calibration of the detectors at the proper room temperature. This experiment was repeated three times over a period of three months. Positioning errors were minimized by taping the source to the imaging bed. Relation between sensitivity and temperature was also investigated in each configuration by linear regression analysis using GraphPad Prism version 6.02 for windows.

Image quality
To assess overall image quality and accuracy of corrections, the NEMA NU-4 image quality phantom (internal dimensions : length, 50 mm; diameter, 30 mm) was filled with 10 MBq of 18 F and then scanned for 20 min once the activity had reached the required 3.7 MBq at the start of the acquisition. Two bed positions with a combined FOV of 56.25 mm were needed to acquire the entire length of the phantom in each configuration (ON/OFF). Images were reconstructed using the 3D MLEM (maximum likelihood expectation-maximization) algorithm with all oblique coincidences being included. The number of iterations (30) was chosen to obtain optimal compromise between MLEM bias and noise (Bergeron et al 2009. The reconstructed voxel size was set to 0.5×0.5×0.6 mm 3 . The algorithm takes into account random corrections as well as dead time correction and normalization, but attenuation and scatter were not corrected. Image quality was assessed according to the NEMA NU4-2008 standard. For the uniform cylinder region, the non-uniformity was reported as a percentage standard deviation (%SD). For the cold and hot rod regions, the spill-over ratios (SORs) and the recovery coefficients (RC) were respectively calculated with their associated percentage standard deviation (%SD).
To further investigate any temperature dependence in these measurements, three repeated acquisitions of the image quality phantom were performed on alternate days after re-calibration of the detectors in each configuration. For each phantom acquisition, corresponding normalization was acquired overnight within a delay of no more than 24 h. This ensured that a normalization correction was adapted to the temperature state of the detectors in each configuration. The figures of merit were analyzed in the same manner as described above. All the individual uncertainties were propagated through the repeated measurements and the statistical significance between the two configurations was evaluated by Student's t test using GraphPad Prism version 6.02 for windows.

Room temperature
During the ON measurements, the temperature was 22.3±0.2°C next to the gantry and 21.1±0.2°C, 2 m away from the gantry. During the OFF measurements, the temperature decreased to 20.8±0.1°C and 19.6±0.2°C, respectively.

Spatial resolution
The results for the radial and tangential spatial resolutions obtained from FBP reconstruction are graphically presented in figure 3 as a function of radial offsets. In the ON configuration, the FWHM/FWTM radial resolutions ranged from 1.62±0.08/ 2.95±0.06 mm at a distance of 5 mm from the center to 2.83±0.11/5.19±0.27 mm at a distance of 25 mm from the center. Tangential resolutions (FWHM/FWTM) were found to be more stable across the FOV and ranged from 1.88±0.08/3.57±0.06 at a distance of 5 mm from the center to 2.03±0.03/ 3.82±0.04 mm at a distance of 25 mm from the center. Mulitple t tests showed no significant differences (p>0.05) between the ON and the OFF configuration. FWHM and FWTM resolutions were slightly degraded at the one fourth-axial-offset position but exhibited no significant differences (p>0.05) between both configurations.
Iterative algorithms are expected to give better spatial resolution results because they are more adapted to take into account detector designs through accurate modeling of the detection process in the system matrix and robust compensation of the missing data between detector modules (Dumouchel et al 2009). Figure 4 show images of the Ultra Micro Hot Spot phantom obtained with 3D MLEM reconstruction at 3 different activity levels, in the ON configuration. Images obtained in the OFF configuration are not presented as they were visually indistinguishable from the ON configuration. Regardless of the activity levels, the phantom images exhibited no distortions. With 200 MLEM iterations, the 1.0 mm rods were always resolved. Figure 4(D) also show the reconstructed images obtained with different number of MLEM iterations. The optimal number of iterations will depend on the image statistic in order to have a good compromise between improved spatial resolution and noise amplification. At the optimal number of iterations chosen for the image quality phantom (30 iterations), the 1.1 mm rods were resolved. Figures 5(A) and (C) show total, true, random, scattered event rates and NECR, plotted as a function of the average effective activity concentration for the mouse and rat phantoms in the ON configuration. Comparison of NECR performance between the ON and OFF configuration is plotted in figures 5(B) and (D) for the mouse and rat phantoms, respectively. The peak NECR was 133 kcps at 3.8 MBq ml −1 for the mouse phantom and 25 kcps at 0.4 MBq ml −1 for the rat phantom in the ON configuration. The respective values were 132 kcps at 3.4 MBq ml −1 and 28 kcps at 0.4 MBq ml −1 in the OFF configuration. The scatter fractions of the mouse and rat phantoms were 13% and 23% in the ON configuration versus 14% and 25% in the OFF configuration.

Sensitivity
Absolute sensitivities are listed in table 2 for the two configurations investigated at the default 250-650 keV energy window. The magnet was switched OFF three times over a three-month period and recalibrated after temperature stabilization. The ON configuration exhibited a 5.1±0.3% loss in average sensitivity compared with the OFF configuration. Figure 6 show the influence of temperature on peak absolute sensitivity during these experiments in both configurations. In the OFF configuration, linear regression analysis revealed an estimated decrease of sensitivity of  approximately 5.6% per°C between 20°C and 22°C (correlation coefficient: −0.97) while in the ON configuration, the corresponding estimated decrease was approximately 6.4% per°C (correlation coefficient : −0.95). Slopes and intercepts were not found statistically different (p>0.05) leading to a global estimated decrease of sensitivity of approximately 5.9% per°C between 20°C and 22°C. Table 3 reports the SORs, uniformity and RCs for 30 iterations of the 3D MLEM reconstruction. Individual variabilities associated with each figure of merit make it difficult to identify any difference between the two configurations. To clarify some of the trends observed between the two configurations, three repeated acquisitions were then performed in each configuration after calibration of the detectors at the corresponding ambient temperature. Temperature conditions were similar to the first experiment. For each acquisition, 3D MLEM reconstructions were carried out with a fresh normalization. No significant differences (t test, p>0.05) could be observed between the ON and OFF configuration except for non-uniformity (%SD) which was slightly worse in the ON configuration (figure 7).

Discussion
Performance tests of the PET component of a new micro-PET/MRI system were investigated to highlight possible interferences due to the presence of the magnetic field. The PET component was evaluated using the NEMA NU4-2008 standard to compare the performance characteristics of the in-line system with and without the field activated (ON/OFF) and to allow comparisons with the LabPET4 standalone system.

PET performance in the ON and OFF configuration
The effect of MRI on the PET was found to be negligible except for the average sensitivity which was almost 7.7% lower when MRI was functioning (table 2). This sensitivity loss was attributed to the temperature rise (1.5°C) observed when the magnetic field and all associated electronic equipment of the MR component were switched on. This finding is close to the theoretical model previously proposed for temperature dependence of APD-based LabPET scanners in the case of stabilized temperature changes (Keereman et al 2013) (−8.5%°C −1 with ambient temperature monitored at 1 m from the gantry). To check if the automatic PET detector calibration could impact the results observed over one experiment, the magnetic field was switched OFF three times at different occasions over a three-month period. Each time, the PET detectors were carefully calibrated at the room Figure 5. Total, true, random, scattered event rates and noise-equivalent counting rate for the mouse (A) and rat (C) phantoms in the ON configuration. Comparison of the noise-equivalent counting rate measured between the ON and OFF configuration for the mouse (B) and rat (D) phantom. Table 2. Total, peak and average absolute sensitivities over the entire axial FOV of the LabPET4 scanner when the magnet is ON or OFF. First three columns refer to a measurement performed with the PET detectors calibrated in the ON configuration. Last columns refer to three repeated measurements obtained from three different switches of the magnetic fields, after energy calibration of the detectors. temperature before and after the switch. In the ON configuration, the mean loss in average sensitivity was found to be 5.1±0.3%, which confirmed the reproducibility of the experiment. As shown in figure 6, this loss in sensitivity of the APD-based detectors is uncorrelated to the presence of the magnet field and so is believed to be solely correlated to the mean increased temperature of 1.2°C, measured next to the gantry. Besides, it showed that even a small ambient temperature change might impact the sensitivity of PET detectors and that the simple automatic re-calibration of the PET detectors at the proper temperature did not counter this loss of sensitivity. A beginning of explanation should lie in the very specific calibration mechanisms of the LabPET scanners. For example, in the signal processing architecture of the LabPET system, acceptance of an impinging event is also constrained by an additional noise threshold for each channel (1 channel per APD, 1536 APDs in the LabPET4 scanner). These noise thresholds are adjusted independently and could reduce sensitivity if set too high compared to the low-energy threshold of the corresponding channel. Indeed, LabPET scanners are complex systems which might be fully optimized at a given temperature , with individual detectors adjustments. For the end-user, LabPET scanners are usually calibrated by the manufacturer to obtain the best compromise between detector stability at a given temperature and sensitivity. In the framework of a sequential PET/MR configuration, these results emphasized the need for adequate temperature stabilization of LabPET APD-based detectors. These requirements might be achieved by optimal designs of dual modality packaging and associated cooling systems.
No other differences could be seen between the two configurations. Taking into account experimental errors and streak artefacts of FBP reconstruction, spatial resolution results were indistinguishable. Differences observed in count rate indices could be explained by experimental errors such as activity measurements or filling and centering of the line source phantom. In these two experiments, adequate calibration of the PET detectors in the OFF configuration was not believed to bring any further improvement to counter effects caused by a mean temperature difference of only 1.5°C. As for the NEMA image quality phantom results, which might be of interest in the case of small animal studies, non-uniformity was found to be slightly worse in the ON configuration. This effect was the consequence of a mean number of counts slightly lower in the uniform region of the phantom and could be related to the loss of sensitivity in the ON configuration. SOR and RC were not significantly different in both configurations and so uncorrelated to room temperature changes between the ON and OFF configurations.
As a whole, the PET imaging capabilities of the LabPET4/MR prototype scanner are adequate for preclinical imaging applications. Despite its relatively low sensitivity (Goertzen et al 2012), a key advantage compared with other available sequential PMT-based PET/MRI system (Nagy et al 2013) might be its high count rate linearity. As shown in figure 4, uncompromised image quality of the Ultra Micro Hot Spot phantom can be obtained up to 115 MBq, corresponding only to 87% of the peak NECR obtained using the mouse phantom. Any temperature change (>1°C) should be taken into account by proper re-calibration of the scanner, as in the standalone version of the system. According to the manufacturer, good practice should encourage the normalization calibration to be performed every week and the energy calibration every month.

LabPET subsystem versus LabPET standalone
Selected NU4-2008 performance characteristics of the LabPET4 standalone version were extracted from the publication of the original system  and compared with the present results in table 4. Firstly, we found a peak absolute sensitivity that is almost 35% lower than the standalone system published in Bergeron et al (2014). Secondly, the count rate capabilities that we obtained with the rat phantom  appeared to be much more degraded. Thirdly, spatial resolution results obtained with FBP reconstruction were found to be slightly worse than with the standalone system. Concerning sensitivity, the Lab-PET4 scanner, as published in Bergeron et al (2014), takes advantage of many fine-tuned calibrations of the detectors from the inventors of the system. These calibrations, such as APD polarization or timing alignment for example, are not readily available or require specifics devices to optimize the system. Therefore, the present results are believed to be more representative of the capabilities of commercial Lab-PET systems, once they are properly calibrated at the installation site. Concerning count rate capabilities, the original system was mounted with shields at each end of the axial FOV to limit the detection of photons originating from outside the FOV. The shielding height extended well beyond the diameter annulus, leaving an inner diameter bore of only 131 mm. Without shields, random coincidence rates are expected to rise much faster with increasing out-of-FOV activity, and can be particularly disadvantageous for the 150 mm long rat phantom. Accordingly, true coincidence rates and NECR performance are expected to worsen in these situations. Concerning spatial resolution, our results were slightly poorer (up to 0.27 FWHM) than the standalone system (table 4). These differences could be attributed to the electronic architecture of our present scanner and the analysis method of the line-profiles. FBP reconstruction is suboptimal with labPET systems and might be influenced by the geometrical distribution of active channels. Differences of up to 0.26 FWHM have already been reported on different models of the same LabPET system (Prasad et al 2011. Also, in our study, FWHM (and FWTM) results were the mean of three replicate measurements and the line profiles were fitted with a Gaussian function.

Conclusion
In this work, we assessed the PET performance characteristics of a new in-line prototype composed of a LabPET scanner modified to accept the docking of a cryogen-free superconducting 3.0 T MRI. In accordance with the NEMA NU4-2008 standard, the microPET subcomponent was shown to exhibit some noticeable differences compared with the standalone version: the scatter fractions were 20% better but the counting rate capabilities of the scanner were degraded for the rat-like phantom. Moreover, the sensitivity was found to be lower and correlated with the temperature room: an average 5.1% loss in peak sensitivity was demonstrated to be due to an average room temperature rise of 1.2°C when all the electronic equipment of the MRI device was switched on inside the imaging room. This sensitivity loss could not be compensated for by the automated energy calibration of the system. This finding should encourage optimizing the packaging and cooling system of APD-based scanners even in a sequential configuration in order to minimize the temperature variation.  Sensitivity (250-650 keV) Absolute peak sensitivity (%) 1.4 0.9