review article from 3-d positron emission …...and pmt, such as found in the conventional gamma...

Molecular Imaging and BiologyVol. 6, No. 5, 275–290. 2004

Copyright � 2004 Elsevier Inc.Printed in the USA. All rights reserved.

1536-1632/04 $–see front matter

doi:10.1016/j.mibio.2004.06.003

REVIEW ARTICLE

From 3-D PositronEmission Tomography to 3-D

Positron Emission Tomography/Computed Tomography: What Did We Learn?

David W. Townsend, PhDDepartments of Medicine and Radiology, University of Tennessee Medical Center, Knoxville, TN

The recent introduction of combined positron emission tomography (PET)/computedtomography (CT) scanners is having a far-reaching effect on the field of medical imagingby bringing functional imaging to the forefront in radiology, oncology and other specialties.The PET/CT scanner is an evolution in technology combining two well-developed imagingmodalities: anatomical imaging with CT and functional imaging with PET. The first prototypePET/CT scanner was a consequence of a succession of steps that, in chronological order,included the development of the High Density Avalanche Chamber (HIDAC) PET camera,3-D PET methodology and the rotating partial-ring tomograph (PRT). The successfulcompletion of each step was a prerequisite to progress to the next phase, and the lessonslearned could then be applied to subsequent initiatives. This review will map themilestones from 3-D PET to 3-D PET/CT and assess the role each step played in thedevelopment of PET instrumentation over the past two decades. � 2004 Elsevier Inc. Allrights reserved.

Key Words: PET instrumentation; 3-D PET; Rotating tomograph; PET/CT scanners.

Introduction

The past 20 years in particular have seen dramaticadvances in the performance of imaging instru-mentation for positron emission tomography

(PET). From the low resolution, low sensitivity, singleslice designs of the early 1980s to the high resolution,multi-slice, scanners of today, key imaging parametershave in most cases improved by at least an order ofmagnitude. Current high performance clinical PETscanners comprise more than 20,000 individual detectorelements, with an axial coverage of 16 cm (or greater), a4.5 ns coincidence time window, and 20% (or better)energy resolution. With such specifications, modernPET scanners attain a measured spatial resolution ofaround 2 mm for brain research studies and 4 mm forclinical whole-body studies, a 3-D scatter fraction of 25%to 30%, and a peak noise equivalent count rate (NECR)for 3-D whole-body imaging that approaches 100 kcps.

Address correspondence to: David W. Townsend, PhD, Departmentof Medicine, University of Tennessee Medical Center, 1924 AlcoaHighway, Knoxville, Tennessee 37920-6999. E-mail: [email protected] support for the PET/CT development was provided by NCIGrant CA 65856

Also of note is a six-decade increase in active coinci-dence lines, from a few thousand in the early 1980s toover 109 today. This impressive progress is due to devel-opments in detector construction, new scintillators,better scanner designs, improved reconstruction algo-rithms, integration of application-specific electronics,and, of course, the vast increase in computer power, allof which have been achieved without a correspondingorder-of-magnitude increase in cost.

In the late 1980s, an important advance occurred withthe introduction of 3-D PET methodology for brainimaging with a multi-ring scanner1–3. Acquisition of PETdata in 3-D makes optimal use of the emitted radiation,improving sensitivity compared to 2-D acquisition by afactor of five even after accounting for the increase inscatter and randoms3. However, for whole-body im-aging, the successful implementation of 3-D methodol-ogy has had to await the appearance of faster scintillators,accurate scatter correction models and improved sta-tistically-based reconstruction algorithms. The recentavailability of scintillators such as gadolinium oxyorthos-ilicate (GSO) and lutetium oxyorthosilicate (LSO) withshort decay times, accurately modelled scatter distribu-tions, and attenuation-weighted ordered-subset EM re-construction algorithms have all helped to bring 3-D

275

mailto:[email protected]

mailto:[email protected]

276 Molecular Imaging and Biology, Volume 6, Number 5

whole-body imaging into the clinical arena. More re-cently, the importance of routinely imaging functionin conjunction with high resolution anatomy has beenrecognized and within the past three years, the com-bined PET/CT scanner has been introduced into clini-cal practice4–6. The device acquires accurately-alignedanatomical (CT) and functional (PET) images in thesame scanner during a single imaging session, overcom-ing many of the disadvantages of retrospective softwarefusion7. The combined PET/CT scanner thus providesphysicians with a powerful tool to diagnose and stagedisease, monitor the effects of treatment, and poten-tially design better, patient-specific therapies.

This paper maps the milestones from the early PETscanner designs and development of 3-D methodology tothe advanced combined PET/CT scanners of today.The topics will include 3-D reconstruction algorithms8,9

developed for the High Density Avalanche Chamber(HIDAC) positron camera10,11, the application of 3-Dtechniques to multi-ring tomographs1–3, the first multi-ring scanners with retractable septa12, and the develop-ment of the first partial-ring rotating tomograph(PRT)13,14. The PRT design in turn helped to launch thedevelopment of the PET/CT scanner4, the combinationof anatomical and functional imaging in the samescanner. From the acquisition of the first images in1998, there are now more than 350 combined PET/CTdevices installed worldwide and they represent up to80% of new PET scanner sales.

The past two decades of progress in PET instrumenta-tion have not been achieved without considerable effort,dedication and investment, both intellectual and finan-cial. While mapping the milestones of instrumenta-tion advances it is worth surmising whether suchprogress, impressive though it undoubtedly is, has en-abled PET to really achieve its full promise and po-tential. The lessons thus learned may help guide futuredirections and identify opportunities that are worthy offurther attention.

The HIDAC Camera and 3-D PET

Since the very early development of nuclear medicineinstrumentation, scintillators such as sodium iodide(NaI) have formed the basis for the detector systems.Thallium-activated NaI is an ideal scintillator to convertthe 140 keV photons from the decay of technetium(99mTc) into light of a wavelength matched to a photo-multiplier tube (PMT). The combination of scintillatorand PMT, such as found in the conventional gammacamera, has dominated nuclear medicine instrumen-tation since the 1950s. The challenge presented by posi-tron tomography is the detection of the higher energy,511 keV photons from the annihilation of positrons intissue. Early PET scanner designs, however, were based

on NaI (Tl) even though a typical thickness (3/8”)used in a gamma camera resulted in a very low efficiencyat 511 keV. To improve sensitivity, an increased depthof scintillator was adopted in some early PET scannersbased on individual crystal arrays15. An alternative ap-proach that was explored in the mid 1970s comprised apair of rotating gamma cameras16 in either a standardconfiguration or with thicker, 1” crystals for better sensi-tivity. Curiously, the rotating dual-head gamma cameraexperienced a short-lived revival for PET imaging in themid 1990s.

A real breakthrough for PET came with the introduc-tion of bismuth germanate (BGO)17 as an alternativescintillator to sodium iodide. BGO has only 15% ofthe light output of sodium iodide but is much denserand has a greater stopping power and photofraction.The first PET scanner designs based on BGO appearedin the late 1970s, and the scintillator eventually becamethe most widely-used detector material for PET, a situa-tion that has continued for more than 20 years. Evenso, PET scanner designs based on sodium iodidecontinued to be developed until recently and at onepoint a significant fraction of PET scanners in clinicaloperation were sodium iodide based18,19.

The late 1970s and early 1980s was a period of consid-erable innovation in PET instrumentation with anumber of different scanner designs under develop-ment. While the majority were scintillator-based, somegroups explored a completely different technology,such as the HIDAC10,11, a detector that originated fromthe field of high energy particle physics as the brainchildof Dr. Alan Jeavons, a physicist working at the EuropeanCenter for Nuclear Research (CERN) in Geneva, Swit-zerland. The HIDAC detector is a multi-wire propor-tional chamber (MWPC) that incorporates novel leadconverters to improve sensitivity for the detection of511 keV photons. The standard MWPC was originallydeveloped to detect high energy charged particles (� 1GeV) with high spatial resolution and such devices arerelatively insensitive to neutral gamma rays. The addi-tion of perforated lead converters within the chambersignificantly increases the intrinsic sensitivity due to thephotoelectric conversion in the lead of the incominggamma. The photoelectron is ejected from the lead intoone of the perforations (holes) and, being charged,ionises the gas within the chamber allowing the eventto be detected and accurately localized. Groups at theLawrence Berkeley Laboratory20, Queens University inKingston, Ontario21, University of Pisa in Italy22, Massa-chusetts Institute of Technology (MIT) in Boston23 andthe Royal Marsden Hospital in London, England24 alsoexplored the use of MWPC-based designs for PET, eachwith a different approach to increasing sensitivity; thedrilled lead converter was unique to the HIDAC detec-tor. Figure 1A shows the first HIDAC camera built for

From 3-D PET to 3-D PET/CT / Townsend 277

Figure 1. The development ofthe HIDAC positron camerafor medical imaging. (A) Dr. A.Jeavons makes an adjustment tothe first HIDAC camera used formedical imaging in 1980, (B) thefirst bone scan of a mouse in-jected with 5 µCi of 18F-fluoride,(C) the current design of thequad-HIDAC small animal scan-ner, and (D) a bone scan of a 27g mouse injected with 320 µCiof 18F-fluoride and scanned for20 minutes, one hour post-injec-tion (Data courtesy of Universi-tatsklinikum Munster)

medical imaging around 1980 and Figure 1B is the first18F-fluoride bone scan of a mouse imaged with theHIDAC; only 5 µCi of 18F were injected into the mouse.

Even with the lead converters, a HIDAC detector atthat time was, at best, only 10% sensitive to 511 keVphotons compared to an efficiency in excess of 80% forBGO. The advantage of the HIDAC approach is thepotential to achieve high spatial resolution in a wirechamber with closely-spaced wires. An intrinsic resolu-tion of 1.5 mm was measured with a radioactive linesource in air. A PET camera comprising a pair of 20cm × 20 cm HIDAC modules mounted on a rotatingsupport was assembled and the performance evaluatedwith some limited clinical imaging25 at Geneva Univer-sity Hospital, Geneva, Switzerland. Owing to the lowintrinsic sensitivity of the HIDAC modules, it was essen-tial to operate the camera in 3-D acquisition mode tomake maximum use of the available photon flux. Thedevelopment of fully 3-D image reconstruction algo-rithms was, therefore, an important aspect of this work.Since the BGO scanners in the early 1980s consisted ofonly one or two rings of small crystals with an axialcoverage of 1 cm, acquisition and reconstruction wereintrinsically 2-D. The 3-D acquisition demands of theHIDAC camera therefore provided an early impetus tothe development of fully 3-D reconstruction algorithms.A first step was to derive an appropriate filter to beused in 3-D filtered backprojection to reconstruct dataacquired at a number of discrete positions of the HIDACcamera9. Independently, Colsher, working on 3-D fil-tered backprojection for a dual head rotating gammacamera for PET, derived the general reconstruction

filter for continuously rotating area detectors26. Theresult was also applicable to a continuously rotatingHIDAC camera (where continuous in this contextmeans a large number of small steps), with the conditionthat the system point response function is spatiallyinvariant and appears the same for all points withinthe imaging field-of-view (FOV). To satisfy this conditionfor the HIDAC camera, the acceptance angle for annihi-lation events was limited to a value less than the max-imum, thus defining a smaller imaging volume withinwhich the point response function is spatially invari-ant27. The algorithm was implemented as 3-D backpro-jection followed by a 3-D filtering step, unlike themathematically-equivalent but more conventional im-plementation where the 2-D projections are first filteredand then backprojected in 3-D28.

In the early 1980s, a key advantage of the rotatingHIDAC camera was that the reconstructed spatial resolu-tion at the center of the FOV was isotropic and around3.5 mm, considerably better than the 8 mm to 10 mmfor scintillator-based scanners at that time; the corres-ponding volume resolution was 27 ml for the HIDACcompared with around 1000 ml for the scintillator-basedscanner. However, one lesson learned from this device isthat very high spatial resolution does not compensatefor low efficiency, except when imaging small animalsor small organs such as the thyroid. A coincidence timewindow (2τ) of 30 ns and the lack of effective energyresolution results in high background levels due to ran-doms and scatter that, combined with a low 511 keVsensitivity, is a challenging environment in which to


image patients. Some limited clinical imaging was never-theless attempted successfully, including 18F-fluoridefor bone, 124I for thyroid, 68Ga-colloid for liver, and55Co-bleomycin for brain tumors. While the originalHIDAC camera design never achieved prominence asa clinical scanner, improvements in the technology haveestablished the quad-HIDAC camera as a serious con-tender for animal imaging (Figure 1C). The most recentversion achieves sub-millimetre resolution at sensitivitiesadequate for imaging small animals; the image of a 27g mouse scanned on a quad-HIDAC after injection of320 µCi of 18F-fluoride is shown in Figure 1D. The quan-titative imaging potential of the quad-HIDAC has alsorecently been demonstrated29.

The HIDAC-related efforts in 3-D reconstruction andthe concept of a dual detector rotating scanner wereseminal to subsequent work on 3-D reconstruction formulti-ring scanners, the low-cost partial ring rotatingscanner and, ultimately, the combined PET/CT. TheHIDAC approach highlighted the necessity, in highquality clinical imaging, of achieving both good resolu-tion and high sensitivity while minimizing the acquisi-tion of randoms and scatter. Nevertheless, despite thelow sensitivity and high background levels, the HIDACcamera demonstrated the feasibility of fully 3-D imagingwith an area detector rotating scanner, including bothacquisition and reconstruction.

Multi-ring PET Scanners and 3-D Imaging

With the exception of the dual-head rotating gammacamera16,30 and the HIDAC camera11, most PET scannerdesigns in the 1980s were single or dual rings of crystalscovering an axial extent of 1 cm to 2 cm31. Such a smallaxial coverage was especially a limitation for PET studiesin neuroscience research where ideally the whole brainshould be imaged at the same time rather than withmultiple bed positions at different times. The architec-tural and financial difficulties of increased axial cover-age were largely solved by a breakthrough invention ofDrs. Ron Nutt and Mike Casey at CTI in Knoxville,Tennessee – the block detector in which a 5 cm × 5 cmblock of scintillator (BGO) is bonded to four 1″ photo-multiplier tubes (PMTs)32. The first blocks were cut into8 × 4 smaller crystals and the four PMTs localized theincident photon to one of the 32 crystals based onthe light sharing between the PMTs. The coupling of32 crystals to four PMTs significantly reduced cost (byreducing the number of PMTs relative to a 1-1 coupling)and a single ring of blocks covered 5 cm axially, subdi-vided into four rings of crystals. The first multi-ringscanners based on the block design appeared in the mid-1980s, and by 1987 scanners incorporating two rings ofblocks covering 10 cm axially with eight rings of crystalswere starting to appear at a small number of research

institutions33. With that coverage, much of the braincould be imaged in a single scan.

PET was originally conceived in the 1950s as a 3-D imaging modality with the physical collimation ofconventional single-photon emitters replaced by theelectronic collimation of coincidence imaging. How-ever, concerns over the high level of scattered photonsand the perceived absence of an effective fully 3-D re-construction algorithm resulted in the first multi-ringPET scanners being equipped with lead annuli, or septa,between each ring of crystals (Figure 2A). Septa shieldthe detectors from out-of-plane scatter, effectively subdi-viding the 3-D imaging volume into a set of independent2-D slices analogous to multiple CT sections (Figure 2B).Each slice was reconstructed with a 2-D algorithm suchas filtered backprojection that had been exhaustivelyvalidated for CT and single photon emission computedtomography (SPECT). The first multi-ring scannerstherefore imaged a 3-D positron-emitting distributionas a set of contiguous 2-D sections, a procedure thatcontinues to this day some two decades later eventhough it makes poor use of the available photon flux.

Following the installation of one of the early eight-ringPET scanners, the ECAT 931/08-1233 at HammersmithHospital, London in 1987, Dr. Terry Jones was amongthe first to propose removing the septa and operating thescanner fully in 3-D. With foresight, the eight-ring ECAThad actually been designed to acquire all possible coinci-dence lines-of-response (LORs) and collect a full 3-D data set of 8 × 8 sinograms. After physically removingthe septa, the first 3-D data sets were acquired for a multi-ring BGO PET scanner in early 1988 and reconstructedusing a backproject and filter algorithm34. This algo-rithm satisfies the condition that the point responsefunction must be spatially invariant by subdividing thereconstruction volume into smaller sub-volumes withineach of which invariance is preserved. The filter usedin this algorithm is that derived by Colsher26, the same asused with the rotating area detectors. However, thealgorithm is different to that for the HIDAC camerawhere spatial invariance was achieved by restricting theacceptance angle to a value less than the maximumpossible, ensuring a central volume of uniform re-sponse. Thisprocedure could not be appliedto the multi-ring scanner without restricting the acceptance angleto a small value equivalent to the acceptance angle in2-D (Figure 2B), eliminating all the benefits of 3-Dacquisition.

In general, a more efficient implementation of fil-tered backprojection, particularly in 3-D, is to filter the2-D projection data and then backproject in 3-D. Therequirement that the point response function is spatiallyinvariant is equivalent to a requirement that all projec-tions are fully measured. For a scanner of limited axialextent, this is obviously not the case since projections


Figure 2. (A) In the late 1980s, multi-ring PET scanners were designed with inter-plane lead or tungsten septa between eachdetector ring, (B) when extended the septa limited the acceptance of coincidence events to small incident angles subdividingthe imaging volume into a set of 2-D transverse sections, (C) from the early 1990s, multi-ring scanners incorporated retractablesepta that could be removed allowing all coincidence lines to be acquired and thus imaging the volume fully in 3-D, and (D)the NECR curves for the NEMA NU-2001 phantom acquired for an LSO PET scanner with septa extended (2-D) and septaretracted (3-D); the advantage of imaging in 3-D at low activity concentrations is evident.

at oblique angles are incompletely measured. A solutionto the problem of incompletely-measured projections,the reprojection algorithm35 was proposed by Kinahanand Rogers in 1989 while investigating image recon-struction for a volume PET scanner design. Instead ofrestricting the acceptance angle, all angles are allowedand the partially measured parallel projections are com-pleted by re-projecting the missing projection datathrough an initial estimate of the volume image. Theinitial estimate is obtained by reconstructing the com-plete set of 2-D sections and stacking them to form thevolume through which the missing LORs can be re-pro-jected. Volume images reconstructed with the reprojec-tion algorithm typically incorporated more than 90%of the full 3-D data set.

The results from these preliminary studies with thesepta removed were encouraging, particularly for thebrain, where a factor of 8 increase in count rate wasobserved. Although the random and scattered coinci-dence rates also increased, there was still a net gainof at least a factor of 2 in signal-to-noise for the 3-D acquisition, particularly at low activity concentrations.In addition, for the brain, the peak signal-to-noise in 3-D was achieved at an activity concentration of about10% that required to achieve a similar signal-to-noisewith 2-D acquisition. The lessons learned from this workclearly highlighted some of the advantages of 3-D PETimaging with a multi-ring scanner, at least for brain-sized distributions. However, physical removal of thesepta from this first multi-ring scanner was impracticalon a routine basis and a more efficient procedure to offerboth 2-D and 3-D acquisition capability was required.

Scanners with Retractable Septa

The results from these preliminary 3-D studies werepromising and in 1989 CTI PET Systems in Knoxville,Tennessee embarked on the design and constructionof the first multi-ring PET scanner with retractable septa,the ECAT 953B. With a patient port of 35 cm, the scan-ner was designed specifically for imaging the brain.Push-button control was provided to extend or retractthe septa within two minutes. The ECAT 953B was alsothe first 16-ring PET scanner comprising two rings of 5cm × 5 cm BGO blocks cut into 8 × 8 small crystals. Thefirst scanner of this design was installed in Hammer-smith Hospital London in early 199012. With septa re-tracted (Figure 2C), many more LORs are active and atotal of 256 (16 × 16) sinograms are acquired.

The capability of this system to acquire data in either2-D or 3-D offered an excellent environment in whichto assess the advantages and challenges of 3-D PET.Exhaustive phantom studies were conducted to com-pare the 2-D and 3-D performance of the ECAT 953B3,36.Studies with a 20 cm diameter uniform cylinder showedthat the system count rate increased by a factor of 7.8when the septa were retracted, reducing to a factor of4.7 after scatter subtraction. Other results confirmedthose found previously during the preliminary studies,such as a factor of 2 to 3 improvement in NECR at lowactivity concentrations despite a factor 3 increase inscatter fraction. Similar NECR improvement factors of3 to 5 were found for flow and ligand studies in the brain.

Procedures for 3-D normalization, attenuation andscatter correction were developed and image recon-struction was based on an implementation of the repro-jection algorithm with the Colsher filter35. At that time


a drawback to the adoption of 3-D methodology wasthat the reconstruction of a 128 × 128 × 31 voxel matrixtook several hours on a standard workstation. Accuratescatter correction was also a concern for quantitativestudies especially for receptor measurements. Despitethese and other concerns, by the early 1990s, 3-D PETmethodology for the brain had been developed to apoint where it was being applied in practice for acti-vation studies. Accurate quantitation was not essentialand 12 mCi of 15O-water could be injected in 3-D ratherthan the 50 mCi required in 2-D. Some brain recep-tor studies were also performed in 3-D since the highsensitivity was of benefit for measurements of low recep-tor populations, but here accurate quantitation is re-quired. Other PET scanner designs withretractable septafollowed the ECAT 953B and retractable septa quicklybecame standard on all scanners that were not al-ready intrinsically 3-D. Whole-body designs with retract-able septa appeared in the early 1990s for cardiac andtumor imaging although valid concerns about highlevels of random and scattered events limited the useof 3-D outside the brain. The lessons learned from thiswork firmly established 3-D PET methodology for brainstudies capitalizing on the increased sensitivity essentialfor ligand studies and enabling the injected activitylevel for activation studies to be significantly reduced.

The latest septa-retractable LSO-based PET scannerdesigns show a similar trend in 3-D (Figure 2D) as theearlier designs. The measurements of NECR in Figure2D are for the NEMA NU-2001 phantom in the ECATACCEL with septa extended (2-D) and septa retracted(3-D); the advantages of 3-D methodology at low activityconcentrations can clearly be seen. The success of 3-Dmethodology such as demonstrated in Figure 2D sug-gested a possible design for a low-cost or entry-level PETscanner. The design, completed in 1990 and commer-cially available even today, is based on dual rotatingarrays of block detectors, and the first prototype wasthe partial ring rotating tomograph (PRT-1)13.

The PRT

PET tomographs based on the rotation of a pair ofopposing 2-D area detectors were thoroughly exploredduring the late 1970s and early 1980s11,15,16,30. As de-scribed, most of these approaches had low sensitivityor poor count rate performance and generated littleclinical interest. With few exceptions, by 1988, rotatingdesigns had been abandoned for PET imaging in favorof the stationary multi-ring BGO scanners. In the late1980s, the capital outlay for the imaging equipment wasperceived as a major impediment to the wider accep-tance of PET and efforts to limit the financial investmentrequired to establish a PET center were considered im-portant. For a typical BGO scanner, more than half the

cost lies in the detectors – scintillator and PMTs – andeliminating a significant number of detectors can havea major impact on cost. Thus, increasing sensitivityby acquiring data in 3-D presented an opportunity todesign and build a lower cost scanner by removing overhalf the detector blocks from a multi-ring scanner androtating the remaining partial rings of detectors to ac-quire the full set of 3-D projection data (Figure 3A).This design, with two banks of block detectors, achievesa clinically-acceptable sensitivity by acquiring data fullyin 3-D. In fact, for a fixed number of detectors, it canbe shown that a higher sensitivity (a greater number ofactive LORs) is achieved by arranging the detectors astwo rotating arrays rather than a continuous stationaryring due to the greater axial coverage of the arrays37.

The first design of this type (Figure 3B, top), the PRT-1 consisted of two arrays of 48 (transaxial) × 2 (axial)blocks, each block cut into 8 × 8 crystals13. The blocksin this prototype were identical to those of the 953B12 andthe PRT-1 contained one third of the number of detec-tors in the full-ring scanner. The absolute sensitivityof the prototype was 0.51%, slightly less than the 0.63% ofthe ECAT 953B with septa extended38. The data wereacquired in step-and-shoot mode at six positions ofthe detectors for a 20 cm transverse FOV and a 10.8 cmaxial FOV. The minimum scan time including rotationwas 60 seconds and the prototype included both staticand dynamic imaging capability. The transmissionsource for attenuation correction in 3-D was a 6 mmthick solid germanium plane source containing 0.9 mCiof 68Ge; the source was 10.8 cm wide to cover the axialFOV and it could be attached to the front of one detec-tor array. Scatter correction was based on an iterativedeconvolution approach. Excellent results were ob-tained for imaging brain (Figure 3B, bottom) and heartwith 2-deoxy-2-[18F]fluoro-D-glucose (FDG), including a13-frame dynamic acquisition for the brain. The devicewas also capable of imaging 82Rb in the heart for myocar-dial flow studies, an impressive achievement with a rotat-ing scanner considering the 75 seconds half-life of thetracer13.

A second prototype14, PRT-2, was built in 1993 incor-porating dual arrays of ECAT EXACT block detectorscovering 16.2 cm axially (Figure 4A). Each array com-prised 10 (transaxial) × 3 (axial) blocks representing40% of the detectors in the equivalent full-ring scanner,the ECAT EXACT39. Data acquisition was again in step-and-shoot mode with the number of angular stepsdepending on the transaxial imaging FOV required,typically 18 positions for a 47 cm FOV. Similar to thePRT-1, the transmission source was a rectangular planarsource covering the 16.2 cm axial FOV; for this proto-type, the source was shaped to follow the curvature of thedetectors. The sensitivity of PRT-2 was approximatelydouble that of PRT-1 due to the additional axial cover-age and was actually 25% higher than that of the EXACT


Figure 3. (A) a more cost ef-fective design of PET scannerwas developed by removingdetectors from a full-ringscanner (top) and then rotat-ing the dual partial ring arrays(bottom) to acquire a full 3-D data set, (B) the first partialring rotating PET scannerPRT-1 built at the Universityof Geneva Hospital in collabo-ration with CTI PET Sys-tems (top) and (bottom) thefirst FDG brain scan acquiredwith the prototype in May,1991.

with septa extended; with septa retracted, the 3-D sensi-tivity of the EXACT was a factor 3.5 higher than that ofthe PRT-2. The scanner was evaluated clinically, showingpromise for imaging FDG in brain and heart (Figure4C, right), 13N-ammonia for myocardial flow (Figure 4C,left), and even 15O-water for blood flow activation stud-ies (Figure 4B). Whole-body FDG-PET studies of oncol-ogy patients were also acquired in step-and-shoot mode14.

The success of the two prototypes stimulated the de-velopment of a commercial version of the partial ring

tomograph, announced in 1995 (Figure 5a). The designof the advanced rotating tomograph (ART) (CPS Inno-vations, Knoxville, TN), was based on dual arrays of 11(transaxial) × 3 (axial) ECAT EXACT blocks corres-ponding to 46% of the detectors in the ECAT EXACT.The sensitivity was therefore similar to that of the PRT-2. The design eliminated cables by incorporating slip-ring technology, both mechanical for power and serialcommunications and optical for high-speed data trans-fer, and rotated constantly at 30 rpm. The early ART

Figure 4. (A) a second proto-type partial ring tomographPRT-2 still requiring the use ofcables, (B) cerebral blood flowimaged with 15O-water in thePRT-2; the images are summedover 12 injections each of 12mCi of 15O-water and the scantime is 90 s, and (C) a cardiacviability study with (left) 8 mCiof 13N-ammonia to image myo-cardial blood flow, and (right)10 mCi of FDG to imagemyocardial metabolism. Datacourtesy of the University ofPittsburgh.


designs incorporated dual 68Ge rod transmissionsources mounted at one end of each detector array,later replaced by collimated 137Cs point sources thatscanned axially during rotation; transmission acquisi-tion was in singles mode40. A number of these low-costscanners, the first dedicated PET scanner to be sold forunder $1 million, appeared at clinical installations fromthe mid to late 1990s and were used effectively for FDGimaging of brain, heart and whole-body. The ART wasalso used for brain activation studies with 15O-water,blood flow measurements in peripheral muscle, and82Rb for gated cardiac studies. Figure 5B shows theexperimental set-up for brain activation studies withlaser pain stimulation, and summed flow images in anormal volunteer subjected to pain stimulation areshown in Figure 5C. The ART played a particularly im-portant role in clinical whole-body imaging at severalinstitutions41,42.

The second and more significant impact of the PRT-1 design was the realization that, with dual arrays ofdetectors, there was potentially enough void betweenthe arrays (Figure 6A, arrowed) to mount an X-ray tubeand detectors and acquire anatomical images in addi-tion to the functional images from PET, the two datasets being accurately co-registered. Such a concept(Figure 6B), first envisaged in 1991, would address manyof the difficulties of software registration for the whole

body where patient positioning differences and internalorgan motion are particularly challenging issues.Although the combined PET/CT concept was proposedin 1991, it was another seven years before the first proto-type appeared in the clinic, as described in detail below.

Whole-body PET Imaging for Oncology

The success of 3-D PET methodology in the brain wasnot matched by a comparable situation for the rest ofthe body. Poor whole-body image quality was due toseveral reasons: randoms and scatter arising from activ-ity outside the imaging FOV could not be effectivelyshielded as they can for the brain; the uptake in thebrain is generally greater than for any other organ andhence in the rest of the body the true coincidence ratedecreases while the scatter and randoms increase; pa-tient movement is more difficult to control than for thebrain; patient body habitus varies considerably and, forthe larger patients, the transmission and emission scansare highly noisy; and the reprojection algorithm35, a 3-D version of filtered backprojection does not performoptimally in a high noise situation, introducing streakand other artifacts that interfere with image interpreta-tion (Figure 7A). The situation for whole-body imagingin 3-D was therefore not favorable despite extensiveeffort to produce an accurate scatter correction

Figure 5. (A) a commercial ECAT ART scanner with the front cover removed; the data and power cables required in the twoprototypes are replaced by mechanical and optical slip rings, (B) the ART scanner being used for brain activations studies; thedevice to the left of the scanner is a laser for stimulating a response to differing levels of pain, and (C) summed blood flowimages acquired with 15O-water injections during pain stimulation. (Activation studies performed in collaboration with Drs.Anthony Jones and Stuart Derbyshire, Manchester University, UK.)


Figure 6. (A) the proposal fora combined PET/CT scan-ner arose from the observa-tion of the empty space(arrowed) between the detec-tor arrays in the PRT-1, and(B) the initial design envis-aged for the combined PET/CT with the X-ray tube anddetectors mounted in thegaps between the PET detec-tors. The final design actuallyhad the PET detectorsmounted on the rear of theCT support.

model43,44. At that time, most institutions with retract-able septa scanners elected to perform whole-body scan-ning in 2-D since the septa shielded the detectors fromthe high level of randoms and scatter and a fast 2-Dreconstruction algorithm could be used.

Nevertheless, from 1995 onwards an increasingnumber of 3-D only, dedicated PET scanners, includ-ing the ART scanner, appeared in clinical practice eventhough it was not until 1998 that Medicare approvedreimbursement in the US for PET in lung and a fewother cancers. In addition to the ECAT ART, designssuch as the QUEST (UGM Medical, Philadelphia, PA)based on the PENN-PET18, and later the C-PET19 were3-D only and used the superior energy resolution ofsodium iodide to limit the 3-D scatter fraction. Neverthe-less, for all but the lighter patients, whole-body imagequality in 3-D was poor, further degraded by attenuationcorrection due to the magnitude of the correction fac-tors. Many physicians insisted on reviewing the less-noisynon-corrected images. Even so, for a typical patient, the

NECR averaged over all body positions is about a factorof 2 higher in 3-D than 2-D45. Owing to the success (andlow capital outlay) of the sodium iodide-based designs,and with a small contribution from the installed baseof ART scanners, by the late 1990s a significant fraction oftomographs in clinical operation for oncology were 3-D only. However, for the larger patients, the imagequality was far from optimal and whole-body BGO scan-ners with septa extended were generally superior. Thelesson to be learned is that BGO is not the best scintil-lator for whole-body imaging in 3-D and filtered back-projection is not the optimal algorithm to reconstructthe images.

The situation began to change with the introductionof new fast scintillators for PET. For the first time in overtwo decades, an alternative to BGO appeared in theform of lutetium oxyorthosilicate (LSO), a scintillatorwith five times the light output and seven times fasterthan BGO. Discovered around 199046, LSO first ap-peared in commercial PET scanners some ten years

Figure 7. A coronal sectionfrom an FDG-PET whole-bodyscan acquired in 3-D modewith septa retracted and recon-structed using (A) 3-D filteredback-projection algorithm withreprojection, (B) Fourier rebin-ning (FORE) and attenuation-weighted OSEM (AWOSEM).The noise streaks that are acharacteristic feature of filteredbackprojection are eliminatedby the statistical approach (re-constructions courtesy of DrDavid Brasse).


later. LSO offers better positioning accuracy, shortercoincidence time window, better energy resolution, andreduced detector dead time due to the superior physicalcharacteristics. The scintillator was introduced by CTI,Inc (Knoxville, Tennessee) and was incorporated intothe ECAT ACCEL, the first LSO-based clinical scan-ner47. Compared to the ECAT EXACT39, the BGO-equivalent of the ACCEL, the impact of LSO on perfor-mance includes higher spatial resolution, reduced scat-ter and randoms rates and better count rateperformance. The NECR curves for the ACCEL in 2-Dand 3-D are shown in Figure 2D. At about the sametime, Philips Medical introduced the ALLEGRO, a gado-linium oxyorthosilicate (GSO) based scanner48. Firstdescribed in 198349, GSO is less sensitive than LSO, hasabout 50% longer decay time, and only 36% of thelight output; however, it outperforms BGO in all aspectsexcept stopping power.

In additiontothese improvements indetector technol-ogy and hardware, significant software advances havetaken place over the pastdecade. The considerable effortinvested in scatter correction has converged to an accu-rate image-based scatter correction model50, and withimprovements in computing power and algorithm devel-opment, a statistically-based reconstruction algorithmhas emerged as a successor to filtered backprojection. Inorder to reduce the computational burden, the 3-D ac-quired data set is rebinned into an equivalent 2-D set51

and then reconstructed using the attenuation-weightedordered-subset EM (AWOSEM) algorithm52. Currently,Fourier rebinning (FORE)51 is performed to reduce the3-D sinogram set to 2-D for reconstruction by AWOSEMin which a coincidence LOR is assigned a weight relatedto the attenuation factor of the line – LORs with largefactors are assigned the smallest weight. The combina-tion of FORE with AWOSEM has been implementedefficiently and a reconstructed image can be pro-duced within a couple of minutes. The same data setas that used for Figure 7A has been reconstructed withFORE � AWOSEM; the significantly improved imagequality is shown in Figure 7B. A fast implementation of3-D AWOSEM is under evaluation and initial resultsshow improved clinical image quality compared to the2-D rebinning approach.

The most obvious lesson to be learned from thisdecade of focussed effort is that it is a challenge toobtain high quality 3-D whole-body images. The combi-nation of BGO, filtered backprojection reconstructionand a simple scatter correction model failed to producewhole-body images of sufficiently high quality thatwere diagnostically useful and acceptable to physicians,especially in large patients. Fortunately, these lessonshave been well-learned and the situation is now verydifferent with 3-D whole-body imaging becoming anestablished methodology in many institutions.

The Combined PET/CT Scanner

The significance of anatomical landmarks in functionalimages has long been recognized, at least since the daysof hand-drawn neck outlines on the early nuclear medi-cine thyroid scans. With the advent of digital imaging,a more rigorous approach to combining anatomy andfunction became possible by using software to align thetwo image sets. Nevertheless, most image alignment isstill performed visually by physicians with CT and PETimages presented together on adjacent displays. Despitereceiving little attention clinically, over the past decadeor so software registration algorithms for aligningimage sets acquired by two different modalities haveevolved from simple matching procedures to complexnon-linear warping techniques7,53–55. Motivated by thebelief that the combination of anatomical and func-tional images is beneficial and complementary to bothmodalities, sophisticated software tools have been devel-oped to perform registration. Outside the brain, how-ever, despite the sophistication, software registrationhas had only limited success56 and is still not widely usedclinically. Difficulties include allowing for inconsistentpatient positioning between the different scanners, anduncontrolled internal organ movement. While reason-ably accurate localized alignment is possible for specificregions such as the thorax, co-registration of the wholebody remains problematic and validation of the tech-niques remains a challenge. Nevertheless, limited studiesthat have been made using software-based registrationhave highlighted,despitepractical difficulties, the intrinsicadvantages of combining anatomical and metabolicinformation56.

The general concept of fusing anatomy and functionin the same device is not new. Historically, one of thefirst dual modality systems reported in the literature wasa combination of CT and SPECT with work in the late1980s by Hasegawa et al.57 resulting in a device that usedthe same detector material, high purity germanium,for both modalities. However, to avoid compromisingeither modality by the choice of a common detectormaterial, Lang, Hasegawa and coworkers later devel-oped a device based on two separate scanners – a CT anda SPECT scanner58. In patenting the concept Hasegawanoted that it could equally well apply to CT and PET59,although to progress from concept to realization of afunctioning PET/CT scanner required funding, ade-quate engineering facilities, and a motivated team ofpeople. While the PET/CT concept emerged indepen-dently in 1991 from a proposal by Townsend and Nuttafter observing the void within the PRT-1, the completedprototype did not appear in the clinic until 1998.Funded in part by a grant from the National CancerInstitute in 1995, the final prototype design4 was a con-siderable departure from the initial concept.


Figure 8. A schematic of the original PET/CT prototype design combining a Siemens Somatom AR.SP spiral CT scanner withan ECAT ART PET scanner. Acquisition and reconstruction were kept separate and the CT images were imported into thePET environment to generate CT-based attenuation correction factors and for the fused image display on the PET console.

The prototype design (Figure 8) incorporated a Sie-mens Somatom AR. SP spiral CT scanner with an ARTscanner. The high packing density of components onthe CT rotating assembly precluded the mounting ofthe ART detectors as originally envisaged (Figure 6B).Instead, a second annular support was attached to therear of the CT assembly, set back from the slip rings,and the PET detectors mounted on the additional sup-port. The complete assembly rotated at 30 rpm and toavoid any cross interaction the imaging componentswere operated consecutively and not concurrently. Thedata acquisition systems and image reconstruction com-puters were not integrated and the CT and PET imageswere acquired and processed separately (Figure 8). TheCT images were imported into the PET environmentand used to generate the PET attenuation correctionfactors to be applied to the PET emission data. The CT,PET and fused images were displayed on the PET con-sole for review and interpretation. The device wasinstalled at the University of Pittsburgh PET Facility inApril 1998 for clinical evaluation. During the subse-quent three-year evaluation program, more than 300cancer patients were scanned on the prototype60–63.Some important lessons were learned from the six-yearPET/CT development program, covering scientific, engi-neering and clinical aspects. The scientific and engi-neering experience influenced the design of the firstcommercial PET/CT scanners, while the initial clinicalexperience clarified some of the strengths and challenges

of the approach, identified the application areas wherePET/CT could bring the greatest benefits, and definedprotocols appropriate for these applications.

Following the promising clinical results, the majorvendors moved rapidly to bring a commercial design tothe marketplace. A common feature of the designs thateventually emerged is the low level of hardware integra-tion – all essentially comprise a CT scanner and a PETscanner in tandem64. The integration of the two modal-ities is more focused on the computer systems in orderto present a unified operation for acquisition and imagereconstruction with the underlying computational com-plexity transparent to the user. Interestingly, of the fourmajor distributors, three opted for 3-D-only PET scan-ners and eliminated the septa and two of them opted notto provide standard PET transmission sources and basethe attenuation correction on the CT images. It is lessthan three years since the first commercial PET/CTscanner was installed and the technology is still evolving,and will doubtless continue to evolve in the future. CTin particular is experiencing a period of rapid changeas multi-slice CT scanners, in only three years, have pro-gressed from 2 to 16 slices, with 32 and 64 slices recentlyannounced. There is now a realistic prospect of 2-Darea detectors and cone-beam CT. The appropriate con-figurations for PET/CT have still to be established andmay well depend upon the application area targeted: on-cology, cardiology or neurology. High performance CTmay not be required for imaging cancer, and PET/


Figure 9. (A) the latest high-performance 16-slice LSO PET/CT scanner at the University of Tennessee Medical Center in Knoxville,Tennessee. The unique design of the patient couch eliminates any vertical deflection due to patient weight as the bed movesinto the scanner, and (B) a typical PET/CT scan acquired with this system; the patient is a 155 lb, 50 year-old male undergoingtreatment for colon cancer with bladder metastases. The PET scan (left) was acquired 148 minutes post-injection of 10 mCiof FDG for four minutes per bed position, and the scan demonstrated a focal area of elevated uptake in the right lowerquadrant; the PET/CT scan (right) accurately localizes the uptake anatomically within the abdomen.

CT protocols are still under development, addressingissues such as respiration65–67 and the use of CT oralcontrast media68.

One of the latest PET/CT designs (Figure 9A) com-prises a Sensation 16 CT scanner (Siemens Medical So-lutions, Forchheim, Germany) with a high resolution,LSO-based PET scanner (CPS Innovations, Knoxville,TN). This combination is now distributed by Siemensas the biograph 16. The spiral CT with 5 mm slice spacingtakes less than 25 s to scan from base of brain to upperthigh, and the combination of fast scintillator and highcount rate electronics ensures a PET imaging time thatcan be as short as six to seven minutes for small patients.This is a significant improvement over the 60-minuteimaging times that were the norm just a few years ago.Other commercially-available designs include an opensystem where the CT and GSO-based PET scanner can bemoved apart (Gemini, Philips Medical), and a BGO-based PET scanner with 6 mm detectors and a 4, 8 or16-slice CT (Discovery ST, GE Medical Systems).

An illustration of the extent of progress in PET instru-mentation is demonstrated by the quality of the scansin Figure 9B; the images show a coronal section of apatient scanned recently on the 16-slice LSO PET/CTat the University of Tennessee, Knoxville. The PET scanidentified a focal area of increased uptake in the rightlower quadrant. To date, many thousands of cancerpatients have been imaged in PET/CT scanners in-stalled at more than 400 institutions worldwide. Theimpact of this evolution in imaging technology has beenprofound within radiology and nuclear medicine espe-cially, and sales this year are predicted to exceed $1billion. Many lessons have already been learnt from

combined PET/CT imaging, the most important onesbeing the added confidence in reading PET scans withthe co-registered CT anatomy routinely available, theability to accurately localize focal tracer uptake, andthe potential to distinguish between normal accumula-tion of tracer and pathology. Many more lessons are tobe learned as the technology matures and becomesmore widely available.

Discussion

In the past two decades, it is evident that there havebeen dramatic advances in imaging instrumentation forPET. Numerous scientists and engineers have contrib-uted to these advances at universities and corporateinstitutions worldwide. Much of the pioneering work hasbeen funded through research grants and with financialsupport from national agencies and private foundations.The unrelenting quest for higher sensitivity, better spa-tial resolution, improved signal-to-noise and lower costhas resulted in PET scanners that achieve 2 mm spatialresolution throughout the brain, can acquire data atsystem count rates up to 4 × 107 singles per second, andattain a peak NECR of up to 100 kcps. To meet theselevels of performance, the number of active LORs hasincreased from a few thousands in 1980 to more than109 today. Fortunately, the progress in computationaland data storage capacity has kept pace with demand andthe resources available today largely satisfy clinical needs.

Until reimbursement was approved for a limitednumber of oncology studies in 1998, PET was primarilya research tool for neuroscience. Consequently, in thedecades preceding reimbursement the main driving


force for instrumentation development was brain im-aging, specifically for studies of psychiatric and neuro-degenerative disorders. These studies demanded highspatial resolution to resolve small brain structures,high sensitivity to measure low receptor populations, dy-namic scanning capability to follow rapid tracer kinetics,and accurate quantitation of both specific and non-specific ligand binding. As the emphasis shifted tooncology in the latter part of the 1990s, the demands onthe instrumentation also changed. For clinical oncologyimaging, of greater importance are short scan times,high throughput capability, ease of operation, high re-liability, fast image reconstruction and application-specific image display and analysis. Obviously, thereare some demands that apply equally to clinical andresearch applications such as high sensitivity and lowscatter and randoms rates. Research activities generallyrequire access to the raw data, to scanner operatingprocedures, detector setup, data acquisition settings,reconstruction parameters and image analysis tools.Clinical operation has a preference for standard proto-cols with fixed parameter settings to reduce the opportu-nity for operator error, fast reconstruction times forrapid physician feedback, and efficient display toolsfor reading and interpreting images. The lessonslearned in one domain obviously impact progress in theother domain.

From the early beginnings with HIDAC and otherfully 3-D PET scanners, the man-years of effort that hasbeen devoted to 3-D methodology, first in the brain andnow for the whole-body, is a major investment. SincePET is intrinsically 3-D, in these days of heightenedsensitivity to radiation exposure, it is an obligation tomake maximum use of the available photon flux fromthe patient while ensuring diagnostic image quality. Tolimit the potentially high skin exposure from CT, consid-erable effort has been devoted to making efficient useof the X-ray dose. Although radiation exposure from apositron-emitting tracer like FDG is much lower thanthe CT skin dose, there is nevertheless an increasingtrend to reduce the injected dose to levels that willalmost obligate the use of 3-D acquisition. Fortunately,the achievements for the past 15 years have finally estab-lished the role of 3-D methodology for both brain andwhole-body.

The addition of anatomical imaging capability to PETscanners has had a profound effect, particularly on theclinical operation by essentially eliminating the time forthe transmission scan, improving image quality and in-creasing patient throughput. Initially, from the perfor-mance of the original prototype where five minutes wasrequired for CT and 45 minutes for PET, the PET/CTapproach was often criticized for making inefficient useof the CT scanner. As a consequence of the dramaticadvances in both CT and PET technology, the high

performance PET/CT scanners of today require 30 sec-onds for CT and 10 minutes or so for PET. From theviewpoint of the patient, the actual scan time is now asmall fraction of the time required for the completestudy, including patient preparation, FDG uptake, andpositioning in the scanner. Much has been learnedand much has been achieved towards improving im-aging technology for PET.

From its inception, one of the great strengths of posi-tron tomography was the richness of the compoundsthat could be labelled with short-lived positron-emittingbiomolecules such as 15O, 13N and 11C. The first imagesof the human brain with FDG, were acquired in 1976at a time when the HIDAC camera was imaging the firstmouse with 18F-fluoride. Now, 28 years later, FDG is stillthe only radiopharmaceutical widely used and reim-bursed for clinical imaging, with 82Rb on a much smallerscale for myocardial flow imaging. Whole-body imagingwith FDG can hardly be said to challenge the perfor-mance of current instrumentation. There are, of course,a much wider range of PET radiopharmaceuticals avail-able in the research domain, primarily for brain studies.It is important for the future of PET that the currentsituation evolves so that other tracers such as 31[18F]flu-oro-31-deoxythmidine (FLT), 11C-acetate, 18F-cholineand maybe a hypoxia imaging agent such as 18F-misoor 64Cu-ATSM become available clinically. There aremany aspects of tumor physiology more representative ofcancer than elevated glucose metabolism that could pro-vide an earlier indication of a primary or recurrentmalignancy at a time when effective treatment is stillpossible. PET is clinically under-utilized and even withFDG has not yet achieved the promise of the 1980sdespite these major advances in imaging instrumenta-tion. Progress in tracer development and regulatory ap-proval of PET radiopharmaceuticals is now essential tomatch the advances described here that have occurredin PET instrumentation over the past two decades.

It is a great honor, although with considerable sadness, toaccept this award in memory of my dear friend and colleagueDr. Peter Valk who passed away December 16, 2003. Peterhimself made numerous contributions to the advances de-scribed here, particularly to the evaluation of PET scannerdesigns, to 3-D PET methodology, and to clinical applicationsof PET in which he was instrumental in obtaining the firstMedicare reimbursement for PET oncology studies in 1998.He was a great supporter of PET/CT and he will be sadlymissed. It is also a privilege to recognize the many individuals,both friends and colleagues, who have contributed to this workover more than two decades. The author is particularly in-debted to Dr. Alan Jeavons inventor of the HIDAC camera

This paper reflects the content of the 2003 AMI Distinguished ScientistAward Lecture presented by Dr. David Townsend at the AnnualMeeting of the Academy of Molecular Imaging in Orlando, Floridain March 2003.


and the first to introduce the author to the challenges of PETin 1975; toProfessor GaborHerman for teachingthe author thebasic mathematics of image reconstruction; to Professor AlfredDonath who had the foresight to see the promise of clinicalPET from as early as 1980; to Professor Terry Jones for hisenthusiasm, encouragement and motivation for PET, espe-cially 3-D PET and PET/CT; and to Dr. Ron Nutt for hisinsight, leadership and dedication to the development ofPET instrumentation for over twenty years, and especially forhis contributions to the work chronicled in this review. Theauthor offers special thanks: to Dr. Thomas Beyer, who ob-tained his PhD on the PET/CT project and who made numer-ous innovative contributions; to Drs. Benno Schorr, MichelDefrise, Paul Kinahan and Rolf Clackdoyle for their manyoriginal contributions, both theoretical and practical, to thefield of image reconstruction; to Drs. Antoine Geissbuhler,Terry Spinks, Dale Bailey and Sylke Grootoonk for their nu-merous contributions to 3-D PET methodology; to Dr. HenriTochon-Danguy, Martin Wensveen, Tony Brun, RaymondRoddy, Larry Byars and Anne Christin who help to design,build and operate the first rotating PET scanner; and to Drs.David Brasse and Claude Comtat for their contributions tothe PET/CT development project while at the University ofPittsburgh; and finally to Drs. Jeffrey Yap, Jonathan Carneyand Nathan Hall from the Departments of Medicine and Radi-ology, University of Tennessee Graduate School of Medicine,Knoxville for their friendship and continuing contributionsto the development and application of PET/CT.

The development of the Partial Ring Tomograph was sup-ported by the Swiss government Commission for ScientificResearch (CERS). The PET/CT development project is sup-ported by National Cancer Institute grant CA 65856.

References1. Townsend, D.W.; Spinks, T.J.; Jones, T.; et al. Three-di-

mensional reconstruction of PET data from a multi-ringcamera. IEEE Trans. Nucl. Sci. 36:1056–1065; 1989.

2. Cherry, S.R.; Dahlbom, M.; Hoffman, E.J. Three-dimen-sional positron emission tomography using a conventionalmultislice tomograph without septa. J. Comput. Assist.Tomogr. 15:655–668; 1991.

3. Townsend, D.W.; Geissbuhler, A.; Defrise, M.; et al.Fully three-dimensional reconstruction for a PET cam-era with retractable septa. IEEE Trans. Med. Imaging 10:505–512; 1991.

4. Beyer, T.; Townsend, D.W.; Brun, T.; et al. A combinedPET/CT scanner for clinical oncology. J. Nucl. Med.41:1369–1379; 2000.

5. Lardinois, D.; Weber, W.; Hany, T.F.; et al. Staging of non-small-cell lung cancer with integrated positron emissiontomography and computed tomography. N. Engl. J. Med.348:2500–2507; 2003.

6. Makhija, S.; Howden, N.; Edwards, R.; et al. Positron emis-sion tomography/computed tomography imaging for thedetection of recurrent ovarian and fallopian tube carci-noma: a retrospective review. Gynecol. Oncol. 85:53–58;2002.

7. Hawkes, D.J.; Hill, D.L.; Hallpike, L.; et al. Coregistrationof structural and functional images. In: Valk, P.; Bailey,

D.L.; Townsend, D.W.; Maisey, M.; eds. Positron EmissionTomography: Basic science and clinical practice. Springer-Verlag, New York; 2003:181–198.

8. Schorr, B.; Townsend, D. Filters for three-dimensionallimited-angle tomography. Phys. Med. Biol. 26:305–312;1981.

9. Schorr, B.; Townsend, D.; Clack, R. A general methodfor three-dimensional filter computation. Phys. Med. Biol.28:1009–1019; 1983.

10. Jeavons, A.P.; Hood, K.; Herlin, G.; et al. The high densityavalanche chamber for positron emission tomography.IEEE Trans. Nucl. Sci. 30:640–645; 1983.

11. Townsend, D.W.; Frey, P.; Jeavons, A.P.; et al. High densityavalanche chamber (HIDAC) positron camera. J. Nucl.Med. 28:1554–1562; 1987.

12. Spinks, T.J.; Jones, T.; Bailey, D.L.; et al. Physical perfor-mance of a positron tomograph for brain imaging withretractable septa. Phys. Med. Biol. 37:1637–1655; 1992.

13. Townsend, D.; Wensveen, M.; Byars, L.; et al. A rotatingPET using BGO block detectors: design, performance andapplications. J. Nucl. Med. 34:1367–1376; 1993.

14. Townsend, D.W.; Byars, L.G.; Defrise, M.; et al. Design andperformance of a rotating PET scanner PRT-2. IEEE Medi-cal Imaging Conference Record 2:1058–1062; 1994.

15. Brownell, G.L.; Burnham, C.A.; Chesler, D.A.; et al. Trans-verse section imaging of radionuclide distributions inheart, lung and brain. In: Reconstruction tomographyin diagnostic radiology and nuclear medicine. (Ter-Pogos-sian, M.M.; Phelps, M.E.; Brownell, G.L.; Eds) Baltimore:University Park Press; 1977:293–307.

16. Muehllehner, G.; Atkins, F.; Harper, P.V. Positron camerawith longitudinal and transverse tomographic capabilities.In: Medical radionuclide imaging, volume 1 STI/PUB/44C. Vienna: IAEA; 1977:391–407.

17. Cho, Z.H.; Farukhi, M.R. Bismuth germanate as a poten-tial scintillation detector in positron cameras. J. Nucl.Med. 18:840–844; 1977.

18. Karp, J.S.; Muehllehner, G.; Mankoff, D.A.; et al. Continu-ous-slice PENN-PET: a positron tomograph with volumeimaging capability. J. Nucl. Med. 31:617–627; 1990.

19. Karp, J.S.; Muehllehner, G.; Geagan, M.J.; et al. Whole-body PET scanner using curve-plate NaI(Tl) detectors. J.Nucl. Med. 39:50P; 1998[Abstract].

20. Lim, C.B.; Chu, D.; Kaufman, L.; et al. Initial characteriza-tion of a multiwire proportional chamber positroncamera. IEEE Trans. Nucl. Sci. 22:388–394; 1975.

21. McKee, B.T.A. Towards high-resolution positron emissiontomography for small volumes. In: Positron annihila-tion. Proceedings of the Sixth International Conferenceon Positron Annihilation, University of Texas at Arlington,955–964; April 1982.

22. Del Guerra, A.; Lum, G.K.; Perez-Mendez, V.; et al. TheHISPET Project: state of the art. In: Positron annihilation.Proceedings of the Seventh International Conferenceon Positron Annihilation, New Delhi, India: 810–812,1985.

23. Lanza, R.C.; Osborne, L.S.; Holman, B.L.; et al. Meshchambers for positron emission tomography. IEEE Trans.Nucl. Sci. NS-33:482–485; 1986.


24. Ott, R.J.; Marsden, P.K.; Flower, M.; et al. Clinical PET witha large area multiwire proportional chamber PET camera.Nuclear Instruments and Methods in Physics Research.A269:436–442; 1988.

25. Frey, P.E.; Townsend, D.W.; Flattet, A.; et al. Tomo-graphic Imaging of the human thyroid using 124I. J. Clin.Endocrinol. Metab. 63:918–927; 1986.

26. Colsher, J.G. Fully three-dimensional positron emissiontomography. Phys. Med. Biol. 25:103–115; 1980.

27. Clack, R.; Townsend, D.; Jeavons, A. Increased sensitivityand Field of View for a Rotating Positron Camera. Phys.Med. Biol. 29:1421–1431; 1984.

28. Brookes, R.A.; Di Chiro, G. Principles of computer assistedTomography (CAR) in Radiographic and RadioisotopicImaging. Phy. Med. Biol. 21:689–732; 1976.

29. Honer, M.; Bruhlmeier, M.; Missimer, J.; et al. DynamicImaging of Striatal D2 Receptors in Mice using Quad-HIDAC PET. J. Nucl. Med. 45:464–470; 2004.

30. Paans, A.M.J.; Vaalburg, W.; Woldring, M.G. A rotatingdouble-headed positron camera. J. Nucl. Med. 26:1466–1471; 1985.

31. Williams, C.W.; Crabtree, M.C.; Burgiss, S.G. Design andperformance characteristics of a positron emission com-puted axial tomograph, ECAT II. IEEE Trans. Nucl. Sci.26:619–627; 1979.

32. Casey, M.E.; Nutt, R. A Multicrystal two-dimensional BGOdetector system for positron emission tomography. IEEETrans. Nucl. Sci. 33:460–463; 1986.

33. Spinks, T.J.; Jones, T.; Gilardi, M.C.; et al. Physical perfor-mance of the latest generation of commercial positronscanners. IEEE Trans. Nucl. Sci. 35:721–725; 1988.

34. Defrise, M.; Kuijk, S.; Deconinck, F. A new three-dimen-sional reconstruction method for positron cameras usingplane detectors. Phys. Med. Biol. 33:43–51; 1988.

35. Kinahan, P.E.; Rogers, J.G. Analytic three-dimensionalimage reconstruction using all detected events. IEEETrans. Nucl. Sci. 36:964–968; 1990.

36. Bailey, D.L.; Jones, T.; Spinks, T.J.; et al. Noise equivalentcount measurements in a neuro-PET scanner with re-tractable septa. IEEE Trans. Med. Imag. 10:256–260; 1991.

37. Townsend, D.W.; Byars, L.; Defrise, M.; et al. Rotatingpositron tomographs revisited. Phys. Med. Biol. 39:401–410; 1994.

38. Bailey, D.L.; Spinks, T.J.; Jones, T. A method for measur-ing the absolute sensitivity of positron emission scanners.Eur. J. Nucl. Med. 18:374–379; 1991.

39. Wienhard, K.; Eriksson, L.; Grootoonk, S.; et al. Perfor-mance evaluation of the positron scanner ECAT EXACT.J. Comput. Assist. Tomogr. 16:804–813; 1992.

40. deKemp, R.A.; Nahmias, C. Attenuation correction in PETusing single photon transmission measurement. Med. Phys.21:771–778; 1994.

41. Bailey, D.L.; Young, H.; Bloomfield, P.M.; et al. ECATART - a continuosly rotating PET camera: performancecharacteristics, inital clinical studies, and installation con-siderations in a nuclear medicine department. Eur. J.Nucl. Med. 24:5–15; 1997.

42. Townsend, D.W.; Beyer, T.; Meltzer, C.C.; et al. The ECATART scanner for Positron Tomography: 2. Research andClinical Applications. Clin. Posit. Imaging 2:17–30; 1999.

43. Watson, C.C.; Newport, D.; Casey, M.E. A single scattersimulation technique for scatter correction in three-di-mensional PET. In: Grangeat, P.; Amans, J.L.; eds.Three-dimensional image reconstruction in radiology and nu-clear medicine. Dordrecht, The Netherlands: Kluwer,255–268, 1996.

44. Ollinger, J.M. Model-based scatter correction for fully3-D PET Phys. Med. Biol. 41:153–176; 1996.

45. Townsend, D.W.; Isoardi, R.A.; Bendriem, B. VolumeImaging Tomographs. In: Townsend, D.W., Bendriem, B.;eds. The Theory and Practice of 3-D PET. Dordrecht, TheNetherlands: Kluwer Academic Publishers. 111–132, 1998.

46. Melcher, C.L.; Schweitzer, J.S. Cerium-doped lutetium ox-yorthosilicate: a fast, efficient new scintillator. IEEE Trans.Nucl. Sci. 39:502–505; 1992.

47. Halama, J.R.; Sajdak, R.A.; Henkin, R.E. Performancemeasurements utilizing the NEMA NU-2 protocol onthe ALLEGRO GSO 3-D PET scanner for 17F and 82Rb.J. Nucl. Med. 44:111P; 2003[Abstract].

48. Valk, P.E.; Casey, M.E.; Bruckbauer, T.; et al. Clinical eval-uation of an LSO PET system for 3-D whole-body imaging.J. Nucl. Med. 42:309P; 2001[Abstract].

49. Takagi, K.; Fukazawa, T. Cerium-activated Gd2SiO5 singlecrystal scintillator. Applied Physics Letters. 42:43–45; 1983.

50. Watson, C.C. New, faster image-based scatter correctionfor 3-D PET. In: IEEE Nuclear Science Symposium andMedical Imaging Conference. Toronto, Canada, Novem-ber 1999:1637–1641.

51. Defrise, M.; Kinahan, P.E.; Townsend, D.W.; et al. Exactand approximate rebinning algorithms for 3-D PETdata. IEEE Trans. Med. Imaging 16:145–158; 1997.

52. Comtat, C.; Kinahan, P.E.; Defrise, M.; et al. Fast recon-struction of 3-D PET data with accurate statistical model-ing. IEEE Trans. Nucl. Sci. 45:1083–1089; 1998.

53. Pelizzari, C.A.; Chen, G.T.; Spelbring, D.R.; et al. Accu-rate three-dimensional registration of CT, PET, and/orMR images of the brain. J. Comput. Assist. Tomogr. 13:20–26; 1989.

54. Pietrzyk, U.; Herholz, K.; Heiss, W.D. Three-dimensionalalignment of functional and morphological tomograms.J. Comput. Assist. Tomogr. 14:51–59; 1990.

55. Woods, R.P.; Mazziotta, J.C.; Cherry, S.R. MRI-PET regis-tration with automated algorithm. J. Comput. Assist.Tomogr. 17:536–546; 1993.

56. Wahl, R.L.; Quint, L.E.; Cieslak, R.D.; et al. Anatometa-bolic tumor imaging: fusion of FDG PET with CT or MRIto localize foci of increased activity. J. Nucl. Med. 34:1190–1197; 1993.

57. Hasegawa, B.H.; Stebler, B.; Rutt, B.K.; et al. A prototypehigh-purity germanium detector system with fast photon-counting circuitry for medical imaging. Med. Phys. 18:900–909; 1991.

58. Lang, T.F.; Hasegawa, B.H.; Soo Chin, L.; et al. Descrip-tion of a prototype emission-transmission computed to-mography imaging system. J. Nucl. Med. 33:1881–1887;1992.

59. Hasegawa, B.H.; Gould, R.G.; Lang, T.F.; et al. Emission-transmission imaging system using single energy and dual


energy transmission and radionuclide emission data. USPatent No. 5376795, 1994.

60. Charron, M.; Beyer, T.; Bohnen, N.; et al. Image analysisin patients with cancer studies with a combined PET andCT scanner. Clin. Nucl. Med. 25:905–910; 2000.

61. Meltzer, C.C.; Martinelli, M.A.; Beyer, T.; et al. Whole-bodyFDG PET imaging in the abdomen: value of combinedPET/CT. J. Nucl. Med. 42:35P; 2001[Abstract].

62. Meltzer, C.C.; Snyderman, C.H.; Fukui, M.B.; et al. Com-bined FDG PET/CT imaging in head and neck cancer:impact on patient management. J. Nucl. Med. 42:36P;2001[Abstract].

63. Kluetz, P.G.; Meltzer, C.C.; Villemagne, M.D.; et al. Com-bined PET/CT imaging in oncology: impact on patientmanagement. Clin. Posit. Imaging 3:1–8; 2001.

64. Townsend, D.W.; Beyer, T.; Blodgett, T.M. PET/CT Scan-ners: A Hardware Approach to Image Fusion. Sem. Nucl.Med. XXXIII(3):193–204; 2003.

65. Goerres, G.W.; Kamel, E.; Heidelberg, T.-N.H.; et al.PET/CT image co-registration in the thorax: influence ofrespiration. Eur. J. Nucl. Med. 29:351–360; 2002.

66. Goerres, G.W.; Burger, C.; Schwitter, M.R.; et al. PET/CTof the abdomen: optimizing the patient breathing pattern.Eur. Radiol. 13:734–739; 2003.

67. Beyer, T.; Antoch, G.; Blodgett, T.; et al. Dual modalityPET/CT imaging: the effect of respiratory motion on com-bined image quality in clinical oncology. Eur. J. Nucl.Med. 30:588–596; 2003.

68. Carney, J.P.; Beyer, T.; Brasse, D.; et al. Clinical PET/CTscanning using oral CT contrast agents. J. Nucl. Med.45:57P; 2002[Abstract].

review article from 3-d positron emission …...and pmt, such as found in the conventional gamma...

Documents