/02/$ IEEE 1109

Transcription

1 Performance Measurements for the GSO-based Brain PET Camera (G-PET) S. Surtil Student Member, IEEE) J.S. Karpl Muchllchncr Senior Member, IEEE) L.-E. Adam1 * Senior Member. IEEE AbstractPerformance measurements on the high sensitivity, high resolution G-PET scanner have been completed. This scanner a diameter of 42.cm and axial field-ofview of 25.6cm was designed for brain receptor imaging, as well as regular clinical and blood perfusion studies in the brain. The transverse and axial resolution near the center are 4. and 5. mm (fwtm of 8. and 1. mm), respectively. At a radial offset of 1Ocm these numbers deteriorate by less than 12%. The absolute sensitivity of this scanner measured a 7 cm long line souxe is 4.57 cps/kbq. Scatter fraction measured a line source in a 2cm diameter by 19cm long cylinder is 34.5%. For the same cylinder, the peak NEC rate is measured to be 75 kcps at an activity concentration of kbq/ml (.35 pci/cc), while the peak true coincidence rate is 2 kcps. Image contrast measured six s placed in the cylinder an activity concentration ratio of 8:l is 21-24% for the smallest (diameter of 1mm). However, due to the loss of counts near the edges, the image contrast for this is as high as 36% when using a region half the size of the diameter and a 3 minutes acquisition time. We also show results from the 3D Hoffman brain phantom as well as 18F-FDG patient scans. These images illustrate the high visual quality of images acquired on the G-PET scanner. I. INTRODTJCTION A high sensitivity. high resolution PET scarum using a gatlolir~ium orthosilicatc (GSO) crystal basctl Anger-logic tlctcctor [I] has bccri tlcvclopctl for brain imaging (G-PET) [a]. The irltcrdctl applications inclutlc rcscarch stutlics using llc-taggctl radio-ligaritls for ricuro-rcccptor irnagirlg ~~rotocols. as well as 18F-taggctl ligards and O-water brain stutlics (Figure 1). This scarmcr was tlcsigrlctl to hvc irnprovctl spatial rcsolutiori. high scrisitivity. arid high count-rate capability cornparctl to the curmnt NaI(Tl)l-~~tl HEAD PENN-PET (H-PET) scarmcr [3]. The GPET scmrm uses 4 x 4 x 1Omm GSO crystals. which have an cncrgy rcsolutior~ of 9-1X at 511 kcv. rclativcly high tlcrisity (or liricar attcmmtiori cocfficicrit j1). arid short tlccay tirnc. Thcsc physical ~mrarnctcrs of GSO cornparc favorably those of NaI(T1) and LSO. The scarmcr has a tliarnctcr of 42 cm. an axial ficltl-of-view (FOV) of 25.6 cm. and it opcratcs only in the 3D or volume imaging rnotlc (no scpta). The G-PET scmrm will bc critical for study of structures which fall blow tlic tlirmlioltl of most corivcritiorml PET tcchology. while its high count-rate capability will bc irriportarit for study of physiological cffccts fast temporal tlyriamics. The authors arc the ldcpartrr~cr~t Pennsylvania, Philadelphia, PA 1914 phia, PA surti~rad.upcnn.cdu and of Radiology, Lnivcrsity of ADAC ITGiLl, Philadcl- Fig. 1. The Hospital G-PET of the Member, IEEE) G. brain scanner installed in the Lnivcrsity of Pennsylvania. II. SCANNER PET Ccntcr at the DIXSIGN A schematic tliagrarn of the scarmcr s $mton tlctcctiorl systcrn is shown in Figure 2. A total of 1856 crystals (32 colurrms by 58 rows) and rnrn tliarnctcr PMTs (36 colurriris by 8 rows in a licxagorial lattice) couplctl to a siriglc curvctl lightguitlc arc usctl. The lightguitlc thickness is 1.9 cm 5 rnrn tlccp slots cut in both tlic trarisvcrsc arid axial tlirncrlsiorls. The optimization of the lightguitlc emthccs tlic best crystal tliscrirniriatiori using a local wciglitctl ccntmitl calculatior~ [I]. Imrt of the scarmcr is 3 cm in tliarnctcr. a 2.5 cm thick lcatl sitlc shicltl usctl to rcjcct cvcnts from activity outsitlc the FOV. An q-amp basctl ydsc shapin, D circuit has bccri tlcvclopctl to rnakc the GSO signal symmetric. resulting in irn~mnml samplirig (digitization) of tlic signal by tlic flash analog-to-tligital-convcrtc~s (ADCS) running at a clock spcctl of 5 MHz. Optimal sampling of the signal is rquird to acliicvc good cricrgy rcsolutiori at iritcgmtiori tirncs of ris. Wc use a triggcririg sclicrric to first tlctcrmiric if an iritcractiori is a coiricitlcrit cvcrit. bcforc ~~crformirig any Imsition calculatior~s. To keep the clcctronics sirnplc. PMT signals arc groupctl into small trigger charmcls which liavc bccri optirnizctl to rcthcc tlic tri ggcririg tlcatltirnc by using a high trigger cncrgy tlmdmltl [4]. Each trigger elmml currently cornpriscs of three acljaccrlt colurrms of PMTs a oric colurriri ovcrlap bctwccri acljaccrit trigger charincls. Overall. the G-PET scarmcr has 18 trigger charmcls 24 PMTs ~)cr trigger charmcl. With this ammgcmcr~t /2/$ IEEE 119

2 Fig. 2. Schematic of G-PET scanner (transaxial cross-section), detector diameter of 42 cm and axial field-of-view of 25.6 cm. we can set the trigger channel threshold at 4 kev, allowing a significant reduction in the number of non-photopeak events triggering the electronics. Additionally, we have built in the capability to divide each trigger channel axially into two equal halves (five rows each an overlap of two rows) which would result in a doubling of the total number of trigger channels to 36. Once implemented in the future, this could potentially reduce the scanner deadtime even further. B. Data Processing and Image Reconstruction Three types of corrections are performed before the data are binned into a sinogram. These are: PMT gain matching, energy correction, and distortion removal. Before energy and position calculations, the digitally integrated PMT signals are corrected for PMT gain variations using a lookup table. The PMT gain matching is essential in order to achieve optimum crystal separation as well as good energy resolution. Once the position and deposited energy of an interaction have been calculated, a position-dependent energy correction is performed via a lookup table. Energy correction compensates mainly for systematic variations in the light collection from crystals near a PMT center compared to those crystals near the PMT edge. In principle, this correction also accounts for variations in the light output of individual crystals. However, our measurements have shown that the standard deviation of the light output of the 1856 crystals used in G-PET is only 6.8%, and during assembly, care was taken to use crystals of similar light outputs next to each other. The final data correction involves removal of distortion in the calculated position due to the use of a local centroid algorithm, and the eventual placement of the event in a real crystal position on the scanner. For this purpose, we developed an automated search algorithm which detects the minima in a flood histogram, defines the crystal boundaries in it, and eventually assigns a real position to all events which occur in each boundary region. After all these corrections have been performed, the acquired data from the scanner are rebinned into 32 angles (32 crystals/row), rays, upto 15 out-of-plane tilts, and 115 axial slices. The radial samples are 4.4 mm apart in the center of the scanner and closer together towards the edge (4.4 mm is the crystal-to-crystal spacing along the scanner circumference). The axial pitch of the 58 crystal rows is also 4.4 mm; thus the slice separation is 2.2 mm. Transverse interleaving is performed to improve the radial sampling; the 32 angles rays are resorted into 16 angles 125 rays. The radial bins are now 2.2 mm apart in the center of the scanner (closer together at the edge). In order to remove the non-uniform radial sampling and to generate a sinogram compatible existing software, data interpolation is performed prior to further data corrections (normalization, scatter, and attenuation) and reconstruction. During interpolation, the 125 radial bins (unevenly sampled) are interpolated into 256, 1 mm (evenly sampled) bins; 16 angles are interpolated into 192 angles and 115 slices are interpolated into 128, 2 mm slices. This step may be eliminated in the future to avoid loss of resolution associated interpolation. III. Scanner Performance Measurements Performance measurements were done on the G-PET scanner following the procedure outlined in the new NEMA NU2-21 standards. A. Spatial Resolution Spatial resolution measurements were performed using a point source of 18 F activity in a thin glass capillary tube an inner diameter of less than 1 mm. The axial length of the point source was also kept to less than 1 mm. According to NEMA requirements, measurements were performed at radial positions of 1 and 1 cm, as well as an in-between radial position of 7 cm. The acquired sinograms were reconstructed using Fourier rebinning followed by filtered back projection a ramp filter out any smoothing. The image was reconstructed into a field-ofview array where the pixel size is 1 mm in the transverse and 2 mm in the axial directions. For comparison purposes we also reconstructed the image into a field-of-view (pixel size of 2 mm) using 3D RAMLA iterative reconstruction [5]. Figure 3 summarizes the resolution results. From Figure 3 we clearly see that the spatial resolution of the G-PET scanner is 4 mm in the transverse direction after Fourier rebinning and FBP. This resolution worsens by about 12% due to depth-of-interaction effects at radial distance of 1 cm. The results from 3D RAMLA reconstruction are generally worse by about 1 mm mainly due to the optimization of the reconstruction parameters for clinical imaging situations (and not point sources in air), where reduced noise in the image is desired. Only one iteration was performed in the RAMLA reconstruction a lambda value of.24 and a blob radius of 2.5. The RAMLA results clearly show that the significant deterioration of axial resolution at radial distance of 1 cm (after FBP), is an artifact arising due to the Fourier rebinning 111

3 FWHM [mm] FWTM [mm] Transverse (FBP) Transverse (3D RAMLA) Axial (FBP) Axial (3D RAMLA) Radial position of source [cm] Transverse (FBP) Transverse (3D RAMLA) Axial (FBP) Axial (3D RAMLA) Radial position of source [cm] Fig. 3. Spatial Resolution for a point source in the G-PET scanner. The Top and Bottom figures show the measured fwhm and fwtm values, respectively, in the reconstructed images. The 2D FBP projection was performed a ramp filter out any smoothing after Fourier rebinning of the data. process and does not represent the true performance of the scanner. B. Scatter The scatter fraction for the G-PET scanner was measured using a line source filled 18 F and placed at three radial positions (, 45, and 9 mm) in a water-filled 2 cm diameter and 19 cm long cylindrical phantom. The data were acquired at low count-rates and rebinned using single slice rebinning (SSRB). The energy gates for these measurements were set at EW1 ([435, 665]keV) and EW2 ([41, 665]keV). The measurement was performed the line source at three different radial positions (, 45, and 9 mm) and data analyzed according to the NEMA standard. The maximum scatter fraction (SF) value in the central slices varies from 44% for source radial position of mm to about 42% and 3% when the source is placed 45 and 9 mm radially respectively. The average scatter fraction for the scanner after using the NEMA weighting scheme is measured to be 34.5% (over central 17 cm of the scanner axial length) for EW1. Using a lower energy threshold (EW2) the average scatter fraction increases to about 39%, thus showing the importance of using a high threshold. C. Sensitivity As described in the NU2-21 standard, we measured the absolute sensitivity of the G-PET scanner using a 7 cm long line source at the center of the scanner, out and four different metal sleeves representing varying attenuation coefficients. This measurement technique is based on work described previously by Bailey, et. al. [6]. The measured count rate each metal sleeve was corrected for activity decay and the log of the results plotted as a function of sleeve thickness. Linear regression was employed to fit the data and obtain an extrapolated value for the absolute sensitivity of the scanner. We performed this measurement for two different energy windows: EW1 ([435, 665]keV) and EW2 ([41, 665]keV). The absolute sensitivity was measured to be 4.57 and 4.79 cps/kbq, respectively, showing only a 4.8% difference between the two energy windows. Figure 4 shows a plot of the axial sensitivity profile obtained the EW1 energy window. The axial sensitivity profile peaks at.8 cps/kbq (3 kcps/mci). However, this sensitivity measurement is normalized to the total activity present in the 7 cm long line source. Since the scanner axial FOV is 25.6 cm, the absolute sensitivity for activity in the field-of-view will be about three times higher than the value obtained by the NEMA measurement. Sensitivity [cps/kbq] Fig Slice number NU2-21 axial sensitivity profile for the G-PET scanner. The scanner sensitivity was also measured using the 2 cm diameter by 19 cm long NEMA cylinder. This cylinder was uniformly filled water containing a small amount of 18 F activity dissolved in it. Data were collected at a low count-rate so that random coincidences are negligible. The scatter fraction as measured in Section III B was used to correct for the scattered events in the collected sinograms, and thus estimate the true coincidences (Trues = Total (1-SF)). Measured this way, we obtained a sensitivity of about 2 kcps/kbq/ml (68 kcps/µci/cc). Since the length of this phantom (19 cm) is shorter than the axial FOV of the G-PET scanner (25.6 cm), the measured sensitivity is lower than the true value. The slice sensitivity (every 2. mm) for this phantom in G-PET is about.4 kcps/kbq/ml/2 mm (16 kcps/µci/cc/2 mm) near the center of the FOV. D. Count-rate tests For count-rate measurements, as recommended in the appendix of the NU2-21 standards, we used the 2 cm 1111

4 Count-rate [kcps] Count-rate [kcps] [kbq/ml] True Scatter Random Total Activity concentration [µci/cc] [kbq/ml] True NEC Activity concentration [µci/cc] Fig. 5. Count-rate curves for the G-PET scanner as measured a water-filled 2 cm NEMA cylinder. NEMA cylinder uniformly filled water and 18 F activity dissolved in it. We started about 111 MBq (3 mci) of activity and acquired dynamic scans over several halflives of the 18 F isotope. The tail fitting method using a parabola was used to estimate the background (scatter+random) fraction (background counts/total counts) for each acquisition frame. The background fraction at low activity levels gives the scatter fraction, while at higher activity concentrations it can be used to estimate the Random counts. The Noise Equivalent Count (NEC) rate was then calculated using the formula NEC = T T T+Sc+R, where T is the True, Sc is Scatter, and R is Random count-rate. Figure 5 summarizes the results of this countrate measurement. The energy gates for this measurement were set at EW2 ([41, 665]keV) and the NEC rate peaks at 75 kcps for an activity concentration of kbq/ml (.35 µci/cc). Better results could potentially be achieved the smaller energy window EW1, which reduces scatter in the collected events as well reduces deadtime. E. Image Contrast Image contrast measurements were performed using six small s in the 2 cm NEMA cylinder. The s had internal diameters (d o ) of 37, 28, 22, 17, 13, and 1 mm and were placed so that their centers lie in the same axial plane close to the central slice in the scanner. This measurement was performed in analogy to the prescribed image quality measurement in the NU2-21 standards for whole-body scanners the IEC recovery phantom. A total starting activity of MBq (1.5 mci) of 18 Fwas Fig. 6. A central slice from the reconstructed image for the contrast measurements. All s, filled an 8:1 activity concentration respect to the background, are clearly visible. used the six s containing a similar activity concentration of eight times that of the background cylinder. We imaged this phantom for 5 and 3 min. and used 3D RAMLA reconstruction of fully corrected data. TABLE I Summary of results obtained the G-PET scanner for the contrast measurements a 2 cm NEMA cylinder and six hot s placed in it. The ratio for the to background activity concentrations is 8:1. Sphere dia. Percent contrast (%) Background d o (mm) D=d o D=d o /2 Variability (%) 5 minutes acquisition 5 Mcts minutes acquisition 26 Mcts For analysis, we measured the count density for circular regions diameters (D) similar to the physical dimensions, and centered over the six s in the central slice. Counts in the volume of the are lost due to the partial volume effect. Therefore, we repeated this analysis by reducing the diameter of the circular regions to half that of the size. The background count density was obtained by drawing 12 regions in each of the central and ±1 and ±2 cm slices, leading to an average over 6 background regions for each size. As described in the NU2-21 contrast measurement, the percent contrast for each is defined as 1 C H,j C B,j 1 a H ab 1, where C H,j is the count density in the region-of-interest (ROI) for j, C B,j is the average count density in the background ROI s for j, and a H and a b are the activity concentrations in the s and background respectively. Figure 6 shows the central slice through one of these acquired images which had a total of 26 Mcts. Table I summarizes these results. 1112

5 Fig. 7. Selected slices from the reconstructed image of a 3D Hoffman Brain Phantom measured the G-PET scanner. Images are reconstructed fully 3D RAMLA iterative reconstruction algorithm. F. 3D Hoffman Brain Phantom Figure 7 shows representative slices acquired for a 3D Hoffman Brain Phantom the G-PET scanner. The data were acquired over 45 minutes about 5 million collected events. The image reconstruction was performed using 3D RAMLA reconstruction algorithm. These images show the high spatial resolution and image quality attained by this scanner a good delineation of substructures in the brain phantom. G. Patient Studies Figure 8 shows transverse, sagittal, and coronal views of selected slices from a patient study after an 18 F-FDG injection and scan time of about 3 minutes. High image contrast and spatial resolution lead to good visual quality in routine clinical scans. IV. Discussion and Conclusion A high resolution, high sensitivity, and high count-rate brain scanner using the GSO Anger-logic detector has been developed. Our performance measurements show that this scanner has a transverse resolution of 4 mm (8 mm fwtm) at the center out significant deterioration at a radial distance of 1 cm. The 55% solid angle coverage achieved by this scanner in 3D imaging mode, together the use of 1 mm long GSO crystals, results in an absolute sensitivity of 4.57 kcps/kbq. Good spatial resolution and high sensitivity are essential for imaging certain receptor systems in the brain which may also have low specific activity. In addition, brain perfusion studies 15 O-water, as well as studies using other short-lived isotopes such as 11 C, will require this scanner to operate at high count-rates in order to achieve good signal-to-noise ratio for imaging rapidly changing processes. With the fast signal decay time of GSO, together the restricted light spread in the detector design, the G-PET scanner is capable of operating at significantly higher count-rates little deadtime. The NEC rate for this scanner peaks at 75 kcps (for a 2 cm diameter by 19 cm long cylinder) for an activity concentration of about kbq/ml (.35 µci/cc). The true coincidence rate at this activity concentration was measured to be about 2 kcps. The count-rate results presented here were measured a low energy gate of 41 kev, and could potentially increase when the energy gate is raised to 435 kev. We have also measured the percent contrast achieved by this scanner by imaging six hot s in a uniform warm background. These measurements show that the small s are easily detectable, but even high spatial resolution, the partial volume effect is significant. For clinical 18 F-FDG imaging we currently scan for 3 minutes which provides high image quality. The results we show for the 3D Hoffman brain phantom, as well as the patient studies, illustrate the degree of structure which is visible in these scans. 1113

6 Fig. 8. Transverse, sagittal, and coronal views of selected slices from a patient study. The patient was scanned for 3 minutes after injection of 18 F-FDG. Acknowledgments We would like to thank Dr. Richard Freifelder for help the scanner construction, Dr. Margaret Daube- Witherspoon for data rebinning and reconstruction, and Mr. Chris Cardi for software and data acquisition support. The support we received from the ADAC UGM engineering staff during the completion of this scanner is greatly appreciated. This work was supported by the Counter Drug Technology Assessment Center (CTAC), an office in the Office of National Drug Control Policy, and by the U.S. Dept. of Energy grant No. DE-FG2-88ER6642. References [1] S. Surti, J. S. Karp, R. Freifelder, and F. Liu, Optimizing the performance of a PET detector using discrete GSO crystals on a continuous lightguide, IEEE Trans. Nucl. Sci., vol. 47, no. 3, pp , 2. [2] J. S. Karp, S. Surti, R. Freifelder, M. E. Daube-Witherspoon, C. Cardi, L.-E. Adam, B. Chase, P. Vaska, and G. Muehllehner, Performance of a GSO Brain PET Camera, in IEEE MIC Conference Record, 2. [3] Joel S. Karp, Richard Freifelder, Michael Geagan, Gerd Muehllehner, Paul E. Kinahan, Robert Lewitt, and Lingxiong Shao, Three-Dimensional imaging characteristics of the HEAD PENN-PET scanner, J. Nucl. Med., vol. 38, pp , [4] S. Surti, A Model of Scintillation Detector Performance for Positron Emission Tomography, Ph.D. thesis, University of Pennsylvania, December 2. [5] M. E. Daube-Witherspoon, S. Matej, J. S. Karp, and R. M. Lewitt, Application of the row action maximum likelihood algorithm spherical basis functions to clinical PET imaging, IEEE Trans. Nucl. Sci., vol. 48, no. 1, pp. 24 3, 21. [6] D. L. Bailey, T. Jones, and T. J. Spinks, A method for measuring the absolute sensitivity of Positron Emission Tomographic scanners, European Journal of Nuclear Medicine, vol. 18, pp ,