optimization of rb-82pet acquisition and · pdf file2008 ieee nuclearscience symposium...

2008 IEEE Nuclear Science Symposium Conference Record

OPTIMIZATION OF Rb-82 PET ACQUISITIONAND RECONSTRUCTION PROTOCOLS FOR

MYOCARDIAL PERFUSION DEFECTDETECTION

ling Tang, Member, IEEE, Arman Rahmim, Member, IEEE, Riikka LautamaId, Martin A. Lodge,Frank M. Bengel, and Benjamin M. W. Tsui, Fellow, IEEE

M10-44

Abstract- The purpose of this study is to optimize the dynamicRb-82 myocardial perfusion (MP) PET acquisition/reconstructionprotocols for maximum perfusion defect detection using realisticsimulation data and task-based evaluation. Time activity curves(TACs) of different organs at both rest and stress conditions wereextracted from dynamic Rb-82 PET images of 5 normal patients.Combined SimSET-GATE Monte Carlo simulation was used togenerate nearly noise-free (NNF) MP PET data from a time seriesof 3D NCAT phantoms with organ activities modeling difTerentpre-scan delay times (PDTs) and total acquisition times (TATs).Poisson noise was added to the NNF projections and the OS-EMalgorithm was applied to generate noisy reconstructed images.The channelized Hotelling observer (CHO) with 32x32 spatialtemplates corresponding to 4 octave-wide frequency channelswas used to evaluate the images. The area under the ROCcurve (AVC) was calculated from the CHO rating data asan index for image quality in terms of MP defect detection.The 0.5 cycle/cm Butterworth post-filtering on OS-EM (with21 subsets) reconstructed images generates the highest AVCvalues while those from iteration number 1 to 4 do not showdifferent AVC values. The optimized PDTs for both rest andstress conditions are found to be close to the cross points ofthe left ventricular chamber and myocardium TACs, which maypromote individualized PDT for patient data processing andimage reconstruction. Shortening the TATs for <"-'3 minutes fromthe clinically employed acquisition time does not affect the MPdefect detection significantly for both rest and stress studies.

Index Terms-Rb-82 myocardial perfusion PET, acquisition/reconstruction protocol optimization

I. INTRODUCTION

Myocardial perfusion (MP) PET contributes to the diagnosisof cardiovascular disease by providing perfusion and leftventricle function information in a single study. It has seen asignificant recent gain in the clinical acceptance as a result ofthe developments in PET scanner technology and availability,along with its reimbursement and the commercial distributionof Rb-82 generators [1], [2]. The available evidence suggeststhat myocardial perfusion PET provides an accurate meansfor diagnosing obstructive CAD, which appears superior toSPECT especially in the obese and in those undergoingpharmacologic stress [3].

Despite its increased clinical popularity, there has been limited information on statistical optimization of the Rb-82 PET

Manuscript received November 14, 2008.The authors are with the Department of Radiology, The Johns Hopkins

University, Baltimore, MD 21287 USA, e-mail: [email protected].

acquisition or reconstruction protocols for perfusion study. Anempirical pre-scan delay time (PDT) of "-'90 second has beensuggested and employed in the clinic. It is widely consideredoptimal to leave out early phases with high activity in the bloodpool when reconstructing MP images for diagnosis [4]. Ideallythe longer the acquisition time (the shorter PDT), the higherthe detected counts in the images and thus higher signal-tonoise ratio, which would lead to better detectability of MPabnormality. However, the earlier the PDT, the more spilloverfrom the blood chamber to the myocardium, which woulddeteriorate the contrast between normal and defect regions.The availability of list mode Rb-82 PET myocardial data in theclinic allows optimization of the acquisition and reconstructionprotocols. The purpose of this study is to investigate theeffects of Rb-82 PET myocardial image acquisition and reconstruction protocols on the detection of perfusion abnormalitythrough realistic data simulations and task-based evaluationwith a mathematical observer. The results suggest optimizeddata acquisition and reconstruction parameters for maximumdetectability of myocardial defects in the Rb-82 PET studies.

II. METHODS

A. Patient Data Collection

To perform realistic simulation, we sought through clinicalPET/CT Rb-82 MP records of patients at the Johns HopkinsHospital and located five patients with healthy myocardia.These patients had normal cardiac function, no perfusiondefects, normal CT angiograms, and zero calcium scores. Inthe clinic, the MP PET data acquisition starts at the same timeas the injection of Rb-82 Chloride tracer and the total scan timeinterval is 8 minutes [5]. For each patient, the time activitiesof different organs including heart, liver, lungs, spleen, andstomach were measured from the dynamic images taken duringthe entire data acquisition process at both the rest and stressconditions. For heart, the time activities of the myocardiumand the left ventricular (LV) chamber were extracted.

The organ time activity curves (TACs) of individual patientswere cubic-spline smoothed and then averaged to generate aset of organ TACs representing the Rb-82 biodistribution in anaverage normal patient. The average TACs of the myocardiumand LV chamber at rest and stress conditions are plottedin Fig. 1. Note that the time axis starts at the time when

978-1-4244-2715-4/08/$25.00 ©2008 IEEE 4629

Authorized licensed use limited to: Johns Hopkins University. Downloaded on August 25, 2009 at 13:32 from IEEE Xplore. Restrictions apply.

activity appears, which is ",0.5 minute after the injection (notdisplayed).

Fig. 1. Smoothed TACs of the LV chamber and myocardium of an averagenormal patient at (a) rest and (b) stress conditions. The activities appear I"VO.Sminute after tracer injection (not displayed).

Wall Fraction

(b)

Inferior1800(a)

00Anterior

2700Septal

measured cumulated organ activity over the 8-minute scan.The simulated dose (decay number) for each organ was 20times of the measured cumulated organ activity times theNCAT organ volume. The generated NNF organ sinogramswere ready to be scaled and summed to simulate differentcombinations of organ activities corresponding to any givenpre-scan delay time (PDT) and total acquisition time (TAT).

c. Data Generation and Image Reconstruction

As described in the last subsection, we used the NNF organsinograms to generate simulated MP datasets corresponding todifferent PDTs and TATs. For a specified PDT, the cumulatedactivity for each organ was calculated by integrating theactivity from the time acquisition starts through the end ofthe TACs. The NNF organ sinograms were scaled based onthe cumulated organ activities and then summed together.A Poisson noise generator was used to create 100 noiserealizations of the simulated NNF sinogram for each PDT.Besides the normal MP sinograms described above, we alsogenerated sinograms corresponding to a myocardium withperfusion defect on the LV wall. This perfusion defect is atransmural defect spanning 40: over the anterior-lateral regionand 1.5 em over the long-axis direction (Fig. 2). It has anactivity that is 10% less than the normal activity. Similar tothe sinograms corresponding to a normal heart, 100 noiserealizations of defect sinograms were generated for each PDT.

87

: - LV Chamber: •• -Myocardium

2 3 4 5 6Time (min)

(a)

1 2 3 456 7 8Time (min)

(b)

~ -LV ChamberI.•• -Myocardium

l

>\o. 0 00

• .~I : -......: ~ -. "' .. -.. _-

B. Monte Carlo Simulation

We applied a combined SimSET-GATE simulation tool [6Jto generate the nearly-noise-free (NNF) Rb-82 myocardialperfusion sinogram dataset of individual organs using the3D NCAT phantom [7]. The photon propagation inside thevoxelized NCAT phantom was simulated using the SimSETfrom which the history file generated was passed to the GATEfor the GE Discovery RX PETtCT scanner geometry anddetector circuitry simulation. The GE RX scanner uses LYSOcrystals of dimensions 4.2 x 6.3 x 30 mm3 in the tangential,axial, and radial directions and the LYSO crystals are arrangedinto 9 x 6 blocks. The scanner contains 24 rings and 630crystals per ring, and is operated in the 20 data acquisitionmode. The combined SimSET-GATE tool saves the simulationtime by over 20 times as compared to using GATE alone,making the simulation for NNF datasets accomplishable.

To generate an NNF sinogram for each organ, the number ofdecays applied in the simulation was calculated based on the

Fig. 2. Schematic diagrams showing the (a) short-axis and (b) long-axisviews of the LV with the region of perfusion defect shaded, Ocemer = 45°,I}.H = 40::::, I}.z = I.S em.

The sinogram datasets were reconstructed using the OS-EMalgorithm we developed, which incorporated the deterioratingfactor corrections to achieve performance comparable to thereconstruction software built on the GE RX scanner. Thefollowing procedures were included in the reconstruction process: (1) normalization; (2) attenuation correction: attenuationcoefficient measured via forward projection of the 511 keVequivalent mu-maps, which were generated from the simulatedattenuation images via bilinear scaling [II J; (3) randomscorrection: randoms estimated from the crystal single rates andknowledge of the scanner's timing coincidence window [12J;(4) scatter (in the object/patient) correction: using the singlescatter simulation technique [13J; (5) decay correction; and(6) deadtime correction: deadtime estimated as a global scalingfactor from the average 'block busy' information provided with

4630


(2)

Fig. 3. CHO templates in (a) frequency and (b) spatial domains.

prediction of a human observer performance in a myocardialdefect detection task [14J.

Each of the reconstructed images was reoriented and theshort-axis slice covering the centroid voxel of the perfusiondefect region was cropped to the channel template size, withthe centroid voxel at the center of the cropped image. The pixelvalues in the cropped short-axis images were then windowedby scaling the image so the maximum value in the heartwas mapped to 255 and the resulting floating values wererounded to integers. This scaling and rounding was performedto duplicate what should be done to images used in a humanobserver study.

The leave-one-out strategy was applied in training andtesting the observer [9]. For each pair of normal/abnormalimage noise realization, the rest of all other noise realizationswere used in training the CHO. This process was repeatedfor all the noise realizations in each testing ensemble. Theresulting ratings acquired from CHO were used to estimateROC curves with the LABROC4 program [10J. This programestimates the parameters of the ROC curve, the AUC value,and the standard deviations of these parameters.

(b)(a)

III. RESULTS

A. Iteration Number and Cutoff Frequency Optimization

For the rest condition, we calculated the NSD and contrastfor several PDTs from unfiltered images at iteration numberfrom 1 to 4 of the OS-EM (with 21 subsets) reconstruction.The NSD vs. contrast curves for different PDTs are plotted inFig. 4.

The images reconstructed with two iterations of the OSEM algorithm (clinically employed) were Butterworth filtered at cutoff frequencies ranging from 0.1 cycle/cm to1 cycle/cm (0.1 cycle/cm increments) before ROC analysiswas performed. The AUC values in the MP defect detectiontask increase significantly as the cutoff frequency increasesfrom 0.1 to 0.3 cycle/cm and they plateau when the cutofffrequency reaches 0.5 cycle/cm. The AUC values for imagesfrom different PDTs with cutoff frequency being 0.1, 0.5, andI cycle/em are plotted in Fig. 5.

With the 0.5 cycle/cm cutoff frequency filtering, the AUCvalues were calculated from images reconstructed with different iteration numbers. It was found that the AUC values didnot change from I to 4 iterations then decreased at higheriterations. In the studies performed hereafter, iteration number

XN-XDContrast == -=----=-

XN+Xo

where XN and X0 are the average value from the normal anddefect regions, respectively. Similar to the NSD calculation,voxels close to edges of the regions were not included to makesure the voxels included have relative uniform values.

where x{ is the ith voxel value of the jth noise realization,Xi is the ith ensemble mean value of the voxel i, n is thenumber of voxels in the region and m is the number ofnoise realizations. The contrast was calculated from noisefree sinogram reconstructed images with perfusion defects. Itwas calculated from the defect region and a normal region ofsimilar area:

1 ~n . / 1 ~m (j -)2n 6i=1 V m-1 6j=1 Xi - XiNSD == (1)

Xi

D. Image Evaluation Criteria

Before performing task-based evaluation, mathematical criteria were used to evaluate the reconstructed images. Thenormalized standard deviation (NSD) and contrast were calculated for different PDTs. The NSD was computed fromthe reconstructed images with normal MP activity. It wascalculated from a region over the LV wall of the noisyreconstructed images where the voxel intensity was relativelyuniform:

E. ROC Analysis

Receiver operating characteristic (ROC) analysis was performed for every testing ensemble, i.e., normal and abnormalnoisy MP images from a specific PDT and TAT, reconstructioniteration number, and Butterworth filter cutoff frequency. Toperform ROC analysis, ratings for the defect-present anddefect-absent images need to be generated. A channelizedHotelling observer (CHO) with four octave-wide rotationallysymmetric frequency channels was applied to the reconstructedimages (processed as described below) to generate the ratings[8]. The start frequency and width of the first channel wereboth 1/64 cycles per pixel and the size of the channels was32 x 32. This channel model was previously found to give good

the scanner. Normalization and attenuation corrections wereincorporated in the system matrix, while estimated randomsand scattered events were incorporated in the denominator ofthe ordinary Poisson OS-EM algorithm formula [12], with theimages finally scaled by global decay and deadtime factors.The number of subsets was set to 21 (as done in the clinic)and cubic voxels of size (3.27 mm)3 were utilized in the reconstructions. The reconstructed images were post filtered usingButterworth low-pass filters with different cutoff frequencies.

After optimizing the PDT using methods described in thefollowing sessions, normal and defect sinograms were generated for the optimized PDT but with different TATs throughshortening the acquisition time from the end of the averagepatient TACs. These data were then reconstructed and postfiltered for TAT optimization.

4631


1.....---,---..,--~--,.======:::;-]

Fig. 9. AVC values for images with different POT at stress condition.

0.7

o .25.5.751 1.5 2 2.5 3 3.5Pre-scan Delay Time (min)

00.9::)<tU)

m0.8~...

(J)

0.6

0.9o::)c( 0.8

o.25.5.751 1.5 2 2.5 3 3.5Pre-scan Delay Time (min)

for a direct impression. More PDT samples are tested at thebeginning of the TACs as the activities of the LV chamber andthe myocardium fluctuate more. Two peaks are found for therest study with results shown in Fig. 8 and one peak for thestress study results in Fig. 9. Reviewing Fig. 1, we notice thatthe second AVC peak for the rest study (at 45 sec) and thepeak for the stress study (at 38 sec) occur around the crosspoints of the LV chamber and myocardium TACs. This providevaluable information for optimizing the PDT for individualpatients. Counting the 30 sec prior to first appearance ofdetected counts, the optimized PDT we arrived at is shorterthan the 90 sec PDT generally applied in the clinic.

..U)CDa: 0.7

1

Fig. 8. AVC values for images with different POT at rest condition.

•••IXI,,

" X,•

"'2.5 min---1.5 min• 1 min

'.'0.5 min""'0 min

••""t'"

~,'

o 0.5 1 1.5 2 2.5 3 3.5Pre-scan Delay Time (min)

""'" 0.1 cycle/em-0.5 cycle/em

-l ---1 cycle/em,

'f

111",1'1 ·1"""''''"",,1 ..,\\ ". "\\\\\\ """'',,,,"",,

20

50

60

0.5

c 40U)z

30

0.9

g 0.8c(

1;) 0.7CDa:

0.6

10~--'---------'----------'------'

.03 .035 .04 .045Contrast

Fig. 4. NSD vs. Contrast plot for unfiltered images corresponding to differentPOTs (0, 0.5, 1, 1.5, and 2.5 min), reconstruction iteration number (Iter)increasing from 1 to 4.

2 and the optimized cutoff frequency of 0.5 cycle/cm areemployed for image reconstruction and post filtering.

Fig. 5. AVC values for rest images from different POTs with two-iterationsof reconstruction, Butterworth filtering cutoff frequencies at 0.1, 0.5. and 1cycle/em.

From Fig. 4, it reads that the sequence in terms of betterNSO vs. contrast tradeoff (curve closer to the lower rightcorner) is PDT being 0, 1, 1.5, 0.5, and 2.5 minute. Viewingthe Ave values corresponding to these POTs in Fig. 5 (the0.5 cycle/cm curve), we notice that in general the better NSOvs. contrast tradeoff, the higher the resulting AVe values. Thisconfirms that NSD vs. contrast tradeoff as a reasonable preliminary mathematical criterion for image quality evaluationin terms of the MP defect detection task.

B. Pre-scan Delay TIme Optimization

For both the rest and stress conditions, we perfonned ROCanalysis and estimated AVC values for images correspondingto different POTs (while for each PDT, the acquisition coverstill the end of recorded activities). For each PDT, four noiserealizations of the processed images with defects (for ROCanalysis) are shown in Fig. 6 (at rest) and Fig. 7 (at stress)

C. Total Acquisition Time Optimization

For the optimized POTs of rest (at 0.75 min) and stress(at 0.63 min) studies, we calculated the AVCs for differentshortened TATs. The reason we use the second peak of restAVC (Fig. 8) for TAT analysis is that the LV in imagesat earlier POTs is almost unifonn (Fig. 6), which has beenavoided by physicians. The TATs were shortened from theend of TACs. From Figs. 10 and 11, it is obvious that the

4632


omin, AUC =0.90 .25 min, AUC =0.93 .38 min, AUC =0.93 .5 min, AUC =0.85 .63 min, AUC =0.89

.75 min, AUC = 0.89 1 min, AUC = 0.87 1.5 min, AUC =0.85 2.5 min, AUC =0.80 3.5 min, AUC =0.65

Fig. 6. Processed reconstructed rest images with centroid of defect in the centroid, 4 noise realizations shown for each PDT.

omin, AUC =0.92 .25 min, AUC =0.93 .38 min, AUC = 0.93 .5 min, AUC =0.97 .63 min, AUC =0.97

.75 min, AUC =0.92 1 min, AUC =0.90 1.5 min, AUC =0.89 2.5 min, AUC =0.85 3.5 min, AUC = 0.71

Fig. 7. Processed reconstructed stress images with centroid of defect in the centroid, 4 noise realizations shown for each PDT.

Aves are not significantly affected when reducing the TATsfor < I"V 3 minutes for both the rest and the stress studies.

IV. DISCUSSION

As discussed in subsection 3.2, two peaks are found in theAve vs. PDT curve for the rest study (Fig. 8) while one isfound for the stress study (Fig. 9). As the images from whichthe AVe values were calculated carry cumulated activitiesfrom different POTs, we plotted the cumulated activity vs.PDT in Fig. 12. For the rest study the cumulated activity ofthe LV chamber is higher than that of the myocardium at thebeginning and falls lower as PDT increases (Fig. 12(a)). Forthe stress study, the cumulated activity of the LV chamber isalways lower than that of the myocardium for all the POTs(Fig. 12(b)). Bearing this in mind, we observe that there areapparently two types of reconstructed rest images (Fig. 6),one with the LV looking like an almost uniform disk at thebeginning and the other more like the donut shape later on.This may explain the two peaks (Fig. 8), one for each type ofimages, in the AVe curve for the rest study.

As for the stress condition, the LV always has chamberactivity lower than the myocardium activity. The one peakhappens where the noise level and the contrast between thechamber and myocardium activity reaches the best tradeoff.As known, the shorter a PDT, the more cumulated countsand therefore higher signal-to-noise ratio in the reconstructedimages. At the same time, there is also more spillover fromthe LV chamber to the myocardium tissue as a result of partialvolume effect as well as scattering. This would affect thecontrast between the normal and defect tissue and therefore thedetectability of the defect. Improved partial volume correctionthrough resolution modeling [15] and accurate scatter correction may help better take advantage of the higher statistics atvery early POTs.

In practice, physicians are not used to viewing LV imageswith higher chamber activity than tissue activity. However,we notice that the first peak value in Fig. 8 is higher than thesecond one and more importantly that the AVe value withPDT as 0 min is also higher than the second peak Ave value.It may be then meaningful to consider viewing the imageswith earlier POTs although the chamber activity is higher,

4633


0.8

....en~ 0.85

........E 2.......~........>-~ 1.5;:;uII( 1OSE~

0 0.5UiCDa: 0 L...--J.----L..---J....----l.....---1....-_..1--------.l_----'-_--L-_--L.....-.J

o.25.5.75 1 1.5 2 2.5 3 3.5Pre-Scan Delay Time (min)

(a)

7::=:::-03X 10E ........~........>-....:~ 2....u<COS51oenenCD

b OL..--l---I..--I..--I..----L..-----L..-----L..-----L..----'------"---'

en 0 .25.5.75 1 1.5 2 2.5 3 3.5Pre-scan Delay Time (min)

(b)

1r-----.-------:;:::===:::::r:::::::===:::::r:::::::====::r:::=::::;"]il--PDT =0.63 mini

0 0.95::)<C= 0.9CD......en

0.85

o::)<C

12345Time Shortened (min)

0.95

Fig. 10. AVC values for images with optimized PDT and different TATs atrest condition.

1 r:------.-----i=====:::::!:=:===:::::r:::::::===:::::r:::::::~

--PDT =.0.75 ~inl

o 1 2 3 4 5Time Shortened (min)

Fig. 11. AVC values for images with optimized PDT and different TATs atstress condition.

Fig. 12. Cumulated activity of the LV chamber and myocardium of theaverage normal patient with different TATs at rest (a) and stress (b) conditions.

which have not been favored in the clinic. It is expected tobe more the case with advanced scatter and partial volumecorrection and suggest consideration of change in paradigm.Further studies, especially human observer studies, will beperformed to validate the findings here.

v. CONCLUSIONS

Through realistic simulation of Rb-82 MP PET data, wehave studied the effect of acquisition and reconstruction process on the MP defect detection using ROC analysis. It wasfound that the iteration number of OS-EM algorithm (with 21subsets) varying from 1 to 4 does not affect the AVC values,while the selection of the cutoff frequency of the Butterworthfilter for postproceessing is critical. The cutoff frequency at'"'-'0.5 cycle/em produces images for best MP defect detectionbased on ROC analysis. For both the rest and stress conditionswe optimized the PDT and found that the optimal PDT areclose to the cross points of the TACs of the LV chamberand the myocardium. This information may be critical forimplementing individualized imaging in the clinic for best MPdefect detection. Through studying the effect of shortening theTAT for the optimized PDT, we conclude that shortening the

TAT for <'"'-'3 minutes from the total 8 minutes applied inthe clinic does not affect MP defect detection significantly forboth the rest and stress studies.

REFERENCES

[1] M. F. Oi Carli, "Advances in positron emission tomography", J. Nucl.Cardiol., vol. 11, no. 6, pp. 719-32, 2004.

[2] M. F. Oi Carli, and R. Hachamovitch, "Should PET replace SPECT forevaluating CAD? The end of the beginning", J. Nucl. Cardiol., vol. 13,no. 1, pp. 2-7, 2006.

[3] M. F. Di Carli, S. Oorbala, 1. Meserve, G. EI Fakhri, A. Sitek, and S. C.Moore, "Clinical Myocardial Perfusion PETtCT", J. Nucl. Med., vol. 48,no. 5, pp. 783-93, 2007.

[4] S. L. Bacharach, J. 1. Bax, 1. Case, O. Oelbeke, K. A. Kurdziel, W. H.Martin, and R. E. Patterson, "PET myocardial glucose metabolism andperfusion imaging: Part I Guidelines for patient preparation and dataacquisition", 1. Nucl. Cardiol., vol. 10, pp. 543-54, 2003

[5] A. Chander, M. Brenner, R. Lautamaki, C. Voicu, J. Merrill, and FrankM. Bengel, "Comparison of measures of left ventricular function fromelectrocardiographically gated 82Rb PET with contrast-enhanced CTventriculography: A hybrid PETtCT analysis", J. Nucl. Med., vol. 49,pp. 1643-50, 2008.

[6] M. A. Shilov, E. C. Frey, W. P. Segars, J. Xu and B. M. W. Tsui,"Improved Monte-Carlo simulations for dynamic PET", J. Nucl. Med.,vol. 47, pp. 197,2006.

[7] W. P. Segars, O. S. Lalush and B. M. W. Tsui, "A realistic spline-baseddynamic heart phantom", IEEE Trans. Nucl. Sci., vol. 46, pp. 503-06,1999.

4634


[8] E. C. Frey, K. L. Gilland, and B. M. W. Tsui, "Application of taskbased measures of image quality to optimization and evaluation of threedimensional reconstruction-based compensation methods in myocardialperfusion SPECT", IEEE Trans. Med. Imag., vol. 21, no. 9, pp. 1040-50,2002.

[9] C. E. Metz, "ROC methodology in radiologic imaging", Invest. Radiol.,vol. 21, no. 9, pp. 720-33, 1986.

[10] C. E. Metz, J. H. Shen, and B. A. Herman, "New methods for estimatinga binormal ROC curve from continuously distributed test results", Annu.Meeting Amer. Statistical Assoc., Anaheim, CA, 1990.

[11] C. Burger, G. Goerres, S. Schoenes, A. Buck, A. Lonn and G. vonSchulthess, "PET attenuation coefficients from CT images: experimentalevaluation of the transformation of CT into PET 511-keV attenuationcoefficients", Eur. 1 Nucl. Med., vol. 29, pp. 922-27, 2004.

[12] A. Rahmim, J. C. Cheng, S. Blinder, M.-L. Camborde and V. Sossi,"Statistical dynamic image reconstruction in state-of-the-art high resolution PET", Phys. Med. BioI., vol. 50, pp. 4887-912, 2005.

[13] C. C. Watson, "New, faster, image-based scatter correction for 3D PET',IEEE Trans. Nucl. Sci., vol. 47, pp. 1587-94, 2000.

[14] S. D. Wollenweber, B. M. W. Tsui, E. C. Frey, D. S. Lalush, and K. J.LaCroix, "Comparison of human and channelized Hotelling observers inmyocardial defect detection in SPECT", J. Nucl. Med., vol. 39, p. 771A,1998.

[15] A. Rahmim, J. Tang, M. A. Lodge, S. Lashkari, M. R. Ay, R. Lautamtiki,B. M. W. Tsui, and F. M. Bengel, "Analytic system matrix resolutionmodeling in PET: an application to Rb-82 cardiac imaging", Phys. Med.BioI., vol. 53, pp. 5947-65, 2008.

4635


optimization of rb-82pet acquisition and · pdf file2008 ieee nuclearscience symposium...

Documents