Performance Evaluation of the Inveon Dedicated PET Preclinical Tomograph Based on the NEMA-NU4 Standards

Abstract

The Inveon dedicated PET (DPET) tomograph is the latest generation of preclinical PET systems dedicated to high resolution and high sensitivity murine model imaging. Here, we report on its performance based on the NEMA NU-4 standards.

Methods

The Inveon DPET consists of 64 lutetium oxyorthosilicate (LSO) block detectors arranged in 4 contiguous rings, with a 16.1 cm ring diameter and a 12.7 cm axial length. Each detector block consists of a 20×20 LSO crystal array of 1.51×1.51×10.0 mm³ elements. The scintillation light is transmitted to position-sensitive photomultiplier tubes via optical light guides. Energy resolution, spatial resolution, sensitivity, scatter fraction and count rate performance were evaluated. The NEMA NU-4 image quality phantom and a normal mouse injected with ¹⁸FDG and ¹⁸F⁻ were scanned to evaluate its imaging capability.

Results

The energy resolution at 511 keV was 14.6% on average for the entire system. In-plane radial and tangential resolutions reconstructed with Fourier rebinning and filtered backprojection algorithms were below 1.8 mm full width at half maximum (FWHM) at the center of field of view (FOV). The radial and tangential resolution remained under 2.0 mm and the axial resolution remained under 3.0 mm FWHM within the central 4 cm diameter FOV. The absolute sensitivity of the system was measured to be 9.3% for an energy window of 250–625 keV and a timing window of 3.432 ns. The peak NECR at a 350–625 keV energy window and a 3.432 ns timing window was 1670 kcps at 130 MBq for the mouse-sized phantom and 590 kcps at 110 MBq for the rat-sized phantom. The scatter fractions at the same acquisition settings were 7.8% and 17.2% for the mouse- and rat-sized phantoms, respectively. The mouse image quality phantom results demonstrate that for typical mouse acquisitions, the image quality correlates well to the measured performance parameters in terms of image uniformity, recovery coefficients, attenuation and scatter corrections.

Conclusion

The Inveon system shows significantly improved energy resolution, sensitivity, axial coverage and count rate capabilities compared to previous generations of preclinical PET systems from the same manufacturer. Its performance is suitable for successful murine model imaging experiments.

Tomographic systems dedicated to non-invasive, in-vivo imaging of preclinical animal models have been widely used in research institutes in recent years (1, 2). With the ability of longitudinal imaging the same subject, each individual animal can serve as its own control. Therefore, inter-subject variability can be minimized. Due to the dramatic difference in size between humans and rodents, small animal PET imaging imposes higher performance requirements than clinical PET scanners, particularly on image resolution and sensitivity. The resolution and sensitivity improvements are mainly achieved by using smaller crystal sizes, smaller detector ring diameters and longer axial coverage. With the goal of improving the image quality of laboratory small animal studies, researchers continuously develop new techniques, including new scintillation materials, detectors, electronics, and geometries to improve the performance of preclinical tomographs.

The Inveon dedicated PET (DPET) is the latest generation of commercial tomographs from Siemens Preclinical Solutions, Inc. (Knoxville, TN). It incorporates changes in system geometry, detectors and electronics. This work reports the performance evaluation of the Inveon system in all aspects, including energy and spatial resolutions, sensitivity, scatter fraction, count rate performance and imaging capabilities. It is based on the National Electrical Manufacturers Association (NEMA) NU-4 standards for performance evaluation of small animal PET tomographs (3). The NEMA NU-4 standards offer a standardized methodology for animal PET performance evaluation and establish a baseline of system performance in typical imaging conditions.

MATERIALS AND METHODS

System Description

The Inveon DPET is a lutetium oxyorthosilicate (LSO) based, high sensitivity, high resolution preclinical PET scanner primarily used for murine imaging. The system consists of 64 detector blocks arranged in 4 contiguous rings with a crystal ring diameter of 16.1 cm and an axial extent of 12.7 cm. Each detector block is composed of a 20×20 array of LSO crystals coupled to a position sensitive photomultiplier tube via a light guide. Each crystal has a length of 10.0 mm and a cross sectional area of 1.51×1.51 mm². The crystal pitch is 1.59 mm in both axial and transverse directions.

List mode data are acquired during measurements. From the list mode data, coincidence events can then be sorted into 3D sinograms with different combinations of span and ring difference or into 2D sinograms by either single-slice rebinning (SSRB) (4), or Fourier rebinning (FORE) (5). Images can be reconstructed using analytical 2D filtered backprojection (FBP), 3D reprojection (3DRP) (6) or iterative methods, like ordered-subsets expectation maximization (OSEM) (7, 8) and maximum a posteriori (MAP) (9).

A comparison of the specifications between the Inveon DPET and three previous preclinical systems from the same manufacturer is summarized in Table 1.

TABLE 1.

Specification Comparison of microPET Systems

Category	Parameter	Concorde P4	Focus120	Focus220	Inveon
Detector	Crystal material	LSO	LSO	LSO	LSO
	Crystal size (mm)	2.2×2.2×10	1.51×1.51×10	1.51×1.51×10	1.51×1.51×10
	Crystal pitch (mm)	2.45	1.63	1.63	1.59
	Crystal array	64 (8×8)	144 (12×12)	144 (12×12)	400 (20×20)
System	No. of detector blocks	168	96	168	64
	No. of crystals	10,752	13,824	24,192	25,600
	No. of rings	32	48	48	80
	No. of crystals/ring	336	288	504	320
	Ring diameter (cm)	26.1	15	25.8	16.1
	Gantry aperture (cm)	22.0	12	22.0	12
	Axial FOV (cm)	7.8	7.6	7.6	12.7
	Transaxial FOV (cm)	19.0	10.0	19.0	10.0
	Solid angle/4π	0.29	0.45	0.28	0.62
Dataset	No. of sinograms
	3D	1024	2304	2304	6400
	2D	63	95	95	159
	Sinogram Size	192 × 168	128 × 144	288 × 252	128 × 60
	Sampling distance (mm)	1.225	0.815	0.815	0.795

Energy Resolution

An ¹⁸F point source was placed at the center of the field of view (FOV) to acquire 2D position histograms of each detector in singles mode with energy window wide open. 100,000 counts were acquired for each detector pixel. Lookup tables were generated for individual crystal identification (10).

Energy resolution was determined for each crystal in the system and calculated as full width at half maximum (FWHM) of the 511 keV energy peak divided by the center of the photopeak value. The mean value of all crystal energy resolutions was calculated and the maximum and minimum energy resolutions were obtained.

Spatial Resolution

Spatial resolution measurements were performed with a ²²Na point source conforming to the NEMA NU-4 standards. It has a nominal size of 0.3 mm, is embedded in an acrylic cube of 10.0 mm extent on all sides and has a nominal activity of 198,000 Bq. The energy window setting was 350–625 keV and the timing window was 3.432 ns (the default of the four available timing window settings on Inveon system). ²²Na has emission energy (E_avgβ⁺ = 250 keV) and positron range (~ 0.23 mm) similar to ¹⁸F, which is the most widely used positron-emitting isotope. As per the NEMA protocol, the measured spatial resolutions were not corrected for source size, positron range, or photon acolinearity.

The source was fixed in the tomograph, and located at two axial positions: (i) the center of the axial FOV and (ii) one-fourth of the axial FOV (31.75 mm away from the center along axial direction). For each of the two axial positions, the source was stepped towards the edge of the transverse FOV. For the central 5 mm of the transverse FOV, the source was stepped at 1 mm increments and then at 5 mm steps up to the edge of the FOV.

The list mode data acquired at each location were histogrammed into 3D sinograms with delayed events subtracted to correct for random coincidences. Component based normalization was applied to compensate for the differences in detection efficiency (11). The 3D sinograms were first Fourier rebinned into 2D sinograms and then reconstructed by 2D FBP with a ramp filter cutoff at the Nyquist frequency with a zoom selected to achieve 0.4 mm pixel size in plane. The axial plane separation was 0.796 mm. The response function was formed by summing one-dimensional profiles that were parallel to the radial, tangential and axial directions. A parabolic fit of the peak point and its two nearest neighboring points was used to determine the maximum value of the response function. Linear interpolation between adjacent pixels was used to determine the position of half and one-tenth of the maximum. The FWHM and full width at tenth maximum (FWTM) were determined for each extracted profile. Volumetric resolution was calculated based on the FWHMs of the radial, tangential and axial directions.

Sensitivity

The sensitivity of the system was measured with the same ²²Na point source, used in the spatial resolution measurement and an ¹⁸F point source. In addition, it was also measured with an ¹⁸F line source inserted in a set of concentric Al sleeves, which was a more traditional methodology employed in the past (12). In order to reduce the attenuation from the imaging bed, the ¹⁸F and ²²Na point sources were taped on a thin piece of cardboard and placed into the scanner FOV. The concentric Al tubes were suspended on both ends without other attenuation material in the FOV.

Due to the difficulty of accurately measuring the small nominal activity of ²²Na source in a standard dose calibrator, the ²²Na point source was only used to determine the relative sensitivity between different energy and timing window settings. The ²²Na source was positioned at the center of the FOV and scanned for 5 minutes at a fixed timing window of 3.432 ns. Two sets of energy windows were used, one with a fixed lower level discriminator (LLD) = 350 keV and changing upper level discriminator (ULD) from 550 to 700 keV and the other with a fixed ULD = 625 keV and changing LLD from 250 to 450 keV, both at 50 keV steps. The reason of using a solid ²²Na source instead of liquid ¹⁸F to measure the relative sensitivity was twofold (i) to simplify the experimental protocol and (ii) to avoid the complications of decay correction between different measurements.

The absolute sensitivity was determined by an ¹⁸F point source of ~ 40 μl and 42,550 Bq placed at the tip of a small centrifuge tube and positioned at the center of the scanner FOV. It was measured for 5 minutes at an energy window of 350-625 keV and a timing window of 3.432 ns. The activity of the ¹⁸F source was measured in a Wallac 1480 Wizard 3″ Gamma Counter (PerkinElmer life sciences, Turku, Finland).

The LSO scintillator crystals have intrinsic radioactivity (13–15). ¹⁷⁶Lu emits β⁻ particles with an average energy of 420 keV together with 3 gammas of 307, 202 and 88 keV, respectively (16). The β⁻ particles and gamma photon can make a true coincidence if both fall in the preset energy window (12, 17).

For each energy window, a background measurement was acquired for 5 minutes. The histogrammed background true counts were subtracted from the total histogrammed true counts, when the ²²Na or ¹⁸F point sources were placed inside of the scanner. The number of net true coincidences was normalized to the scan duration, divided by the source activity, and corrected for the branching ratio (0.906 for ²²Na and 0.967 for ¹⁸F). Attenuation of the 1 cm³ cube of ²²Na and the centrifuge tube for the ¹⁸F source was not compensated, but a 9% sensitivity loss was estimated based on the 0.095 cm⁻¹ attenuation of 511 keV photons in water equivalent material. The absolute sensitivity at the 350–625 keV energy window was determined by the ¹⁸F measurement. Based on the relative sensitivity determined by the ²²Na source, the absolute sensitivity for the 350 keV LLD and 625 keV ULD energy window datasets was also calculated.

We used these measurements, to investigate the system sensitivity as a function of energy window. The energy window used in typical studies should be determined based on the tradeoff between absolute sensitivity, scatter fraction and system background. For our institute, an energy window of 350–625 keV was selected as a compromise for the typical studies we perform.

The sensitivity dependence on timing window was also measured for 3 minutes with the same ²²Na point source at the 4 available timing window settings (2.808, 3.432, 4.056 and 4.680 ns) and a fixed energy window of 350–625 keV.

The axial sensitivity profile was measured with a set of concentric Al tubes and a plastic tube filled with ¹⁸F solution with both ends sealed (12). The source tubing was placed inside the smallest metal tube, suspended in the center of the transaxial FOV and aligned with the axis of the tomograph. The other Al tubes were added, one at a time and the count rate was measured for each set of metal tubes for 120 seconds at 350-625 keV energy window and 3.432 ns timing window. The set of measurements with different thicknesses of Al tubes was used to obtain the count rate without attenuation by exponential fitting. The list mode data was histogrammed with SSRB without randoms and/or scatter correction. For each slice of each acquisition and for points located greater than 1 cm on each side of the peak, the values were set to zero. The counts in all lines-of-response (LOR) in the sinograms were summed slice by slice, corrected for decay and scaled by the acquisition time and the 96.7% positron yield for ¹⁸F. The ¹⁸F activity was measured in a Wallac Gamma Counter and the counts in the central 7 cm and the whole axial FOV were added to calculate the sensitivity of a mouse size and rat size object, respectively.

Scatter Fraction and Counting Rate Performance

Scatter fraction and count rate performance were measured using two different cylindrical polyethylene phantoms, which simulate the geometries of a mouse and a rat. The design of the phantoms conformed with the NEMA NU-4 standards.

Both phantoms were made of high density polyethylene (0.96 g/cm³). The mouse-like phantom was a 70 mm long solid cylinder with 25 mm diameter. A cylindrical hole with a diameter of 3.2 mm was drilled parallel to the central axis at a radial distance of 10 mm. The rat-like phantom had similar geometry but larger dimensions. It was 150 mm long and had a 50 mm diameter. A 3.2 mm diameter hole was drilled at a radial offset of 17.5 mm.

A ¹¹C solution with concentration higher than 1,500 MBq/ml was enclosed in a flexible tubing with an outer diameter fitting the 3.2 mm hole. The initial activity in the FOV was measured in an Atomlab™ 300 Dose Calibrator (Biodex Medical Systems, Shirley, New York, U.S.A) and was higher than 500 MBq and 600 MBq at the start of the acquisition for mouse and rat size phantoms, respectively. The acquisition was performed at the 350–625 keV energy and 3.432 ns timing windows. The phantom was centered in the FOV and data was acquired until the total activity decayed below 10,000 Bq. The random coincidences were measured by the delayed window technique. The list mode data was histogrammed into 2D sinogram sets with 5 minute frame duration, by SSRB with separate prompts and delays.

For each prompt sinogram (transaxial bin size 0.795 mm, slice thickness 0.796 mm), all pixels located farther than 8 mm from the edge of the phantom were set to zero. The profile of each projection angle was shifted so that the peak pixels were aligned with the center pixel of the sinogram. A sum projection was then produced by adding up 160 angular projections in each slice and each frame. All pixel counts outside of a 14 mm centered band were assumed to be the sum of random, scatter and intrinsic counts. A linear interpolation between the left and right border of the 14 mm band was used to estimate these non-true counts under the profile peak. Counts above this line were regarded as true coincidences. Random coincidences were estimated from the delayed sinogram. A background acquisition was obtained at the same energy window and timing window for 16 hours with the cold mouse-sized and rat-sized phantoms in the FOV. The intrinsic counts were estimated from the background sinogram. The scattered count rate was then calculated by equation (1):

$$R_{\mathit{scatter}}=R_{\mathit{total}}-R_{\mathit{true}}-R_{\mathit{random}}-R_{int\mathit{rinsic}},$$ (1)

where, R_scatter, R_total, R_true, R_random and R_intrinsic are the scatter, total, true, random and intrinsic count rates, respectively. The scatter fraction was calculated by equation (2):

$$SF=\frac{R_{\mathit{scatter}}}{R_{\mathit{scatter}}+R_{\mathit{true}}}.$$ (2)

The Noise Equivalent Count Rate (NECR) of each of the 5 minute frame acquisitions was determined using the following equation (18, 19):

$$\mathit{NECR}=\frac{R_{\mathit{true}}^2}{R_{\mathit{total}}+R_{\mathit{random}}}.$$ (3)

Imaging Studies

NEMA Phantom Study

The NEMA NU-4 mouse image quality phantom (3) is composed of 3 regions.

1. A main fillable uniform region chamber with 30 mm diameter, 30 mm length.

2. A lid that attaches to the main fillable region, containing two smaller cold region chambers. One of the chambers was filled with non-radioactive water and the other chamber was filled with air. Both chambers were composed of hollow cylinders, 15 mm in length, 10 mm outer diameter (OD) and 1 mm wall thickness.

3. A solid Lucite region of 30 mm diameter and 20 mm length with 5 fillable rods drilled through (at 7 mm from the center) with diameters of 1, 2, 3, 4, and 5 mm, respectively.

The image quality phantom was filled with 5.1 MBq of ¹⁸FDG solution and acquired at 350–625 keV energy and 3.432 ns timing windows for 20 minutes. The phantom was placed on a mouse imaging chamber to simulate the situation of actual mouse imaging. The activity in the phantom was measured with a Wallac Gamma Counter.

A CT transmission scan of the image quality phantom together with the imaging chamber was performed with a MicroCAT™ II tomograph (Siemens Preclinical Solutions, Knoxville, TN). The angular sampling was 1° per projection for a full 360° scan. The x-ray source was operated at 500 μA and 70 kVp. Images were reconstructed using a Feldkamp cone-beam algorithm with a ramp filter cutoff at the Nyquist frequency. The reconstructed CT image was registered with the PET emission image to create an attenuation sinogram. The whole 3D PET sinogram data were first Fourier rebinned and then reconstructed by 2DFBP with a ramp filter cutoff at the Nyquist frequency. Normalization, deadtime, random, attenuation (20) and scatter corrections (21, 22) were applied.

Uniformity

A 22.5 mm diameter and 10 mm high cylindrical volume of interest (VOI) was drawn over the center of the uniform region of the image quality phantom. The average concentration, maximum and minimum values in this VOI and the percent standard deviation (%STD) were measured to estimate the noise performance.

Recovery Coefficient

The image slices covering the central 10 mm length of the rods were averaged to obtain a single image slice of lower noise. Circular ROIs were drawn in this image around each rod, with diameters twice the physical diameters of the rods (the exact size of the ROIs is not absolutely critical for this measurement). The maximum values in each of these ROIs were measured. The maximum values were divided by the mean value obtained in the uniformity test to obtain the recovery coefficient (RC) for each rod size.

The transverse image pixel coordinates of the locations with the maximum ROI values were recorded and used to create 10 mm long line profiles along the rods in the axial direction. The standard deviation of the pixel values measured along each line profile was calculated. The standard deviation of the RC was calculated as follows:

$${\mathit{STD}}_{RC}=RC\sqrt{{\left(\frac{{\mathit{STD}}_{\mathit{lineprofile}}}{{\mathit{Mean}}_{\mathit{lineprofile}}}\right)}^2+{\left(\frac{{\mathit{STD}}_{\mathit{background}}}{{\mathit{Mean}}_{\mathit{background}}}\right)}^2}.$$ (4)

Accuracy of Scatter Correction

VOIs were defined in the water- and air-filled cylindrical inserts. The diameter of the VOI was 4 mm and encompassed the central 7.5 mm in length in the axial direction. The ratio of the mean in each cold region to the mean of the hot uniform area was reported as spill-over ratio (SOR).

Mouse Study

Two normal mice were injected with 8.9 MBq ¹⁸FDG and 9.4 MBq ¹⁸F⁻, respectively. One and a half hours later the mice were centered inside the gantry and imaged for 1 hour. CT transmission scans of the mice were performed on a MicroCAT™ II tomograph with the same protocol as the CT scan for the image quality phantom. The PET data were reconstructed with FORE+2DFBP (with a ramp filter cutoff at the Nyquist frequency) and MAP reconstructions (β = 0.01) with all available corrections applied, including attenuation and scatter.

RESULTS

Energy Resolution

Energy resolution of the 511 keV photopeak was 14.6% on average of 25,600 LSO crystals, with 26.9% and 8.2% as the worst and best energy resolutions. Compared with the previously reported 18.5% (23), 18.3% (24) and 26% (25) energy resolutions for Focus220, Focus120 and microPET P4, the energy resolution of the Inveon DPET was significantly improved.

Spatial Resolution

Figure 1 shows the radial, tangential, and axial components of the FORE and FBP reconstructed point source images. At the center of the FOV, the image resolutions in the transverse planes were below 1.8 mm FWHM and remained under 2.0 mm FWHM within the central 4 cm diameter FOV.

FIGURE 1.

FORE+2DFBP reconstructed image resolution of the Inveon DPET system as a function of radial offset from the center of the FOV. (A) FWHM and (B) FWTM of radial (diamond), tangential (square), and axial (triangle) image resolutions and (C) volumetric resolutions for point sources located at the axial center (solid) and 31.75 mm from the axial center (hollow).

Sensitivity

Table 2 summarizes the two measurements of the absolute sensitivity, one at fixed LLD = 350 keV and the other at fixed ULD = 625 keV.

TABLE 2.

The Absolute Sensitivity of Different Energy Window Settings

LLD = 350 keV
ULD	550	600	625	650	700
Sensitivity (%)	6.32	6.64	6.72	6.74	6.85
ULD = 625 keV
LLD	250	300	350	400	450
Sensitivity (%)	9.32	7.86	6.72	5.95	4.19

At the center of the axial and transaxial FOV, the absolute peak sensitivity measured using an energy window of 350–625 keV and a timing window of 3.432 ns was 6.72%. When the LLD was lowered to 250 keV, the absolute sensitivity was measured to be 9.32%. This measurement did not include a correction for the self attenuation of the source. With consideration of self attenuation, the absolute sensitivity is expected to be higher than 10% at the 250–625 keV energy window.

The relative sensitivity measured at different timing windows was compared with the sensitivity measured with 4.680 ns timing window (the widest timing window available on Inveon). The measurement shows that the sensitivity does not depend significantly on the timing window (within 1.5%).

The average sensitivity for a mouse size object (7 cm) and a rat size object (12.7 cm) measured at 350–625 keV energy window and 3.432 ns timing window were 4.0% and 2.8%, respectively. The axial sensitivity profile is shown in Figure 2.

FIGURE 2.

The axial sensitivity profile over the axial positions.

Scatter Fraction and Counting Rate Performance

With an energy window of 350–625 keV and a timing window of 3.432 ns, the peak NECR is 1670 kcps achieved at 130 MBq for the mouse size phantom and 590 kcps achieved at 110 MBq for the rat size phantom. The scatter fraction at this acquisition setting is 7.8% and 17.2% for the mouse and rat size phantoms, respectively. The NECR as a function of activity is plotted in Figure 3 for the two phantoms.

FIGURE 3.

NECR as a function of total activity for mouse-sized (diamond) and rat-sized (triangle) phantoms.

The count loss due to deadtime was also investigated based on the mouse size phantom data (Figure 4). The count loss is 25% at an activity of 50 MBq and 50% at an activity of 110 MBq. The deadtime correction works well up to an activity of 50 MBq (within 1% accuracy compared to the expected counts).

FIGURE 4.

Raw counts (dashed line), decay corrected counts (grey solid line) and deadtime corrected counts (black solid line) as a function of total activity in the FOV based on the mouse size phantom data.

Imaging Studies

NEMA Phantom Study

The images of a transverse plane showing 5 rods (Figure 5A), a coronal plane (Figure 5B), a transverse plane of the uniform region (Figure 5C), and a profile across the uniform area (Figure 5D) of the NEMA image quality phantom are shown.

FIGURE 5.

Images of the NEMA NU-4 image quality phantom scanned for 20 minutes with 5.1 MBq ¹⁸FDG. (A) A transverse plane of the 5 rods region, (B) coronal view, (C) transverse plane of the uniform region and (D) profile across the uniform area.

Uniformity

The uniformity measurement results were summarized in Table 3. With FORE and FBP reconstruction, the percentage standard deviation was measured to be 5.29% with all corrections applied.

TABLE 3.

The Uniformity Measurement of the NEMA Phantom

	Mean	Maximum	Minimum	%STD
FBP	0.00126	0.00149	0.00101	5.29

Recovery coefficients

The recovery coefficients reconstructed with FBP for 5 different rod sizes from 1 mm to 5 mm diameter are shown in Figure 6. The RC for the smallest 1 mm rod is 0.15 and is 0.94 for the largest 5 mm rod.

FIGURE 6.

The recovery coefficients for 5 rods of different sizes reconstructed with FBP.

Accuracy of corrections

The spill-over ratios measured in the air-filled and water-filled chambers of the NEMA phantom after all the corrections applied were −0.57% and 1.65%, respectively. Without attenuation and scatter correction, the SORs for air-filled and water-filled chambers are 13.95% and −0.21%, respectively. The residual activity is a measurement of correction accuracy. With SORs below 2% after corrections for both air-filled and water-filled chambers, the corrections work reasonably well.

Mouse Study

Figure 7 shows the coronal and sagittal images of mice with ¹⁸FDG and ¹⁸F⁻ uptake, respectively. The images were reconstructed with FBP and MAP (β = 0.01) reconstructions.

FIGURE 7.

Coronal plane of a normal mouse scan with injection of 8.9 MBq ¹⁸FDG reconstructed with (A) FORE+2DFBP and (B) MAP. Sagittal plane of a mouse bone scan with injection of 9.4 MBq ¹⁸F- reconstructed with (C) FORE+2DFBP and (D) MAP.

DISCUSSION

The NEMA standards are not meant to produce absolute performance measurements, but rather define measurements that can be easily performed, analyzed and interpreted by the user community. They serve as tools for comparison of imaging instruments under specific operating conditions. These specifications represent a subset of measurements that define the performance of preclinical PET systems for specific imaging tasks.

The energy and spatial resolutions were primarily determined by the scintillator type and individual crystal size of the system. The LSO based Inveon DPET system has improved energy resolution and similar spatial resolution as the Focus220, which is also based on LSO scintillator and has the same crystal element size as the Inveon (23). The increase in energy resolution is mainly due to the use of improved light guides in the Inveon, which reduces light loss. The sensitivity is significantly higher than the previous systems because of the large solid angle coverage. The peak NECR for both mouse and rat-sized phantoms are also significantly improved with respect to previous preclinical systems due to the increased sensitivity. The scatter fractions can not be directly compared to previous measurements for other systems (23, 24) because a different size of phantom and a different location of the line source insert were used here as per the new NEMA NU-4 standard. Here we also have the addition of the NEMA NU-4 image quality phantom that puts together several aspects of the system performance in a single measurement that is relevant to a rather common acquisition protocol.

The measured recovery coefficients for different rod sizes are possibly underestimated due to scatter. The RC was calculated by comparing the rod value to the uniform filled background intensity. The rods are surrounded by cold Lucite and therefore, there are less scattered coincidences into the rod region than the uniform background area. Although, we performed scatter correction during image reconstruction, the scatter correction is possibly less effective for large uniform regions. In order to investigate this, an increased LLD from 350 keV to 450 keV was used to image the same image quality phantom. At that energy window, we saw an increase in the recovery coefficient values for all rods, which demonstrates that the recovery coefficients are underestimated at least in part due to scatter. The NEMA mouse image quality phantom is rather new and there is limited experience in its use with other systems, therefore further investigation with other tomographs needs to be performed to fully evaluate the significance of these findings.

The LSO crystals have an intrinsic background emission. With a wide open energy window, the intrinsic radioactivity will produce a uniform background and affect the ability of detecting low activities in the FOV. The total ¹⁷⁶Lu equivalent activity contained in the scintillation crystals is calculated to be 166,500 Bq based on the percentage of radioactive ¹⁷⁶Lu. Due to the relative compact geometry and large solid angle of the Inveon DPET compared with other LSO based scanners, like the Focus220, the background of Inveon might be more significant. This LSO background can be reduced by increasing the LLD at the expense of sensitivity. The absolute system sensitivity penalty is about 30% when the LLD increases from 250 keV to 350 keV.

In the design of the Inveon DPET system, small detector ring diameter and large axial extent were used to increase the solid angle coverage and therefore to achieve higher system sensitivity. Because the Inveon DPET has a long axial FOV, it can cover the whole mouse including the tail in one scan. This provides the possibility to estimate the total injected activity from the reconstructed image itself. However, due to the relatively small detector ring diameter and long axial extension comparing with other animal PET systems, the Inveon is more prone to radial and axial resolution degradation due to the photon penetration at large radial offsets and large ring differences if reconstructed with analytical FBP reconstruction. Statistical reconstruction methods, like MAP reconstruction (9, 26), should be the preferred image reconstruction algorithms to obtain high resolution images. Photon penetration and mis-positioning of the coincidence events can be modeled with Monte Carlo simulation and incorporated into a system response matrix to recover the radial and axial resolutions losses (27).

CONCLUSION

This study has evaluated the performance of the microPET Inveon system in several basic aspects based on the new NEMA NU-4 protocol. Although direct comparison with previous results acquired with non standard sources might not be straightforward, the spatial resolution performance was similar to previous generations of microPET scanners in the transverse directions. Absolute sensitivity of the system was measured as 9.3% for an energy window of 250–625 keV and a timing window of 3.432 ns, which is significantly improved with respect to previous tomographs. The peak NECR of the Inveon system was also greatly increased. The image quality phantom test demonstrated that the image uniformity and recovery coefficients were good and the scatter correction worked reasonably well. Further improvements in the overall system performance are expected to be realized with fully 3D iterative reconstruction algorithms that incorporate the estimated system response.