ositron Emissionomography/Computed Tomographyavid W. Townsend, PhD
Accurate anatomical localization of functional abnormalities obtained with the use ofpositron emission tomography (PET) is known to be problematic. Although tracers such as18F-fluorodeoxyglucose (18F-FDG) visualize certain normal anatomical structures, the spa-tial resolution is generally inadequate for accurate anatomic localization of pathology.Combining PET with a high-resolution anatomical imaging modality such as computedtomography (CT) can resolve the localization issue as long as the images from the twomodalities are accurately coregistered. However, software-based registration techniqueshave difficulty accounting for differences in patient positioning and involuntary movementof internal organs, often necessitating labor-intensive nonlinear mapping that may notconverge to a satisfactory result. Acquiring both CT and PET images in the same scannerobviates the need for software registration and routinely provides accurately alignedimages of anatomy and function in a single scan. A CT scanner positioned in line with a PETscanner and with a common patient couch and operating console has provided a practicalsolution to anatomical and functional image registration. Axial translation of the couchbetween the 2 modalities enables both CT and PET data to be acquired during a singleimaging session. In addition, the CT images can be used to generate essentially noiselessattenuation correction factors for the PET emission data. By minimizing patient movementbetween the CT and PET scans and accounting for the axial separation of the two modal-ities, accurately registered anatomical and functional images can be obtained. Since theintroduction of the first PET/CT prototype more than 6 years ago, numerous patients withcancer have been scanned on commercial PET/CT devices worldwide. The commercialdesigns feature multidetector spiral CT and high-performance PET components. Experiencehas demonstrated an increased level of accuracy and confidence in the interpretation of thecombined study as compared with studies acquired separately, particularly in distinguish-ing pathology from normal, physiologic tracer uptake and precisely localizing abnormalfoci. Combined PET/CT scanners represent an important evolution in technology that hashelped to bring molecular imaging to the forefront in cancer diagnosis, staging and therapymonitoring.Semin Nucl Med 38:152-166 2008 Elsevier Inc. All rights reserved.
istorically, instrumentation for tomographic imaging offunction (single-photon emission computed tomogra-hy [SPECT], positron emission tomography [PET]) evolvedlong a path somewhat different from that of anatomical im-ging devices (computed tomography [CT] and magnetic res-nance imaging [MRI]) and the corresponding clinical stud-
epartments of Medicine and Radiology, University of Tennessee MedicalCenter, Knoxville, TN.
inancial support for the original PET/CT development was provided byNCI Grant CA 65856.
ddress reprint requests to David W. Townsend, PhD, Departments of Med-icine and Radiology, University of Tennessee Medical Center, 1924Alcoa Highway, Knoxville, TN 37920-6999. E-mail: [email protected]
52 0001-2998/08/$-see front matter 2008 Elsevier Inc. All rights reserved.doi:10.1053/j.semnuclmed.2008.01.003
es were performed and interpreted separately in differentlinical services, ie, nuclear medicine and radiology, respec-ively. Despite this segregation, the usefulness of combiningnatomical and functional planar images was evident to phy-icians even in the 1960s, preceding the invention of CT. Thelignment of tomographic images is a complex procedurewing to the large number of degrees of freedom and, with-ut some common features, such coregistration, may beroblematic. In addition to simple visual alignment or the usef stereotactic frames that are undesirable or inconvenient in diagnostic setting, sophisticated image fusion software waseveloped from the late 1980s onwards.1 For (relatively)igid objects, such as the brain, software can successfully
lign images from MR, CT, and PET, whereas in more flexible
mailto:[email protected]:[email protected]
Positron emission tomography/computed tomography 153
nvironments, such as the rest of the body, accurate align-ent is more difficult because of the large number of possibleegrees of freedom. Software fusion is also dependent onatching common features that are extracted either from the
mages or from markers placed on the patient. Functionalmaging modalities such as PET and SPECT often lack reli-ble anatomical correlates and have coarser spatial resolutionnd greater noise levels than CT or MR.
One way to address the problems of software fusion is byombining devices (emission and transmission) rather thanusing the images post hoc, an approach that has now coinedhe term hardware fusion. A combined, or multimodality,canner such as PET/CT can acquire coregistered structurend function in a single study. The data are complementary,llowing CT to accurately localize functional abnormalitiesnd PET to highlight areas of abnormal metabolism. A furtherdvantage of combined instrumentation is that the anatomi-al images from CT can be used to improve quantitation ofunctional images through more accurate attenuation, scatternd partial-volume corrections. This is important in achiev-ng accurate and objective assessment of functional parame-ers such as myocardial perfusion, tumor uptake values andosimetry for treatment-planning and monitoring response.Since the commercial introduction of PET/CT in 2001,
doption of the technology has been rapid, particularly inncology. Advances in CT and PET instrumentation haveeen incorporated into the very latest PET/CT designs. In thisrticle, we briefly describe some of the early work that led tohe commercial exploitation of PET/CT and subsequently tots current designs. The impact of recent advances in CT andET performance on these designs will be discussed. An al-orithm for CT-based attenuation correction (CT-AC) will beescribed in addition to the challenges that must be ad-ressed by any implementation of the algorithm in practice.
istorical Conceptshe origins of tomographic imaging in medicine date from
he 1960s or even earlier, but fusion of tomographic imagesas not explored systematically until the late 1980s.1 Follow-
ng the earlier superposition of planar images, in the 1990s 2rincipal approaches have emerged to image fusion: softwarend hardware. The software approach attempts to align 2mage sets post hoc after they have been acquired on differentcanners at different times. In contrast, the hardware ap-roach combines the instrumentation for 2 imaging modali-ies and thus acquires both image sets within the same refer-nce frame and thereby ensures as accurate alignment asossible.
mage Fusion With Softwarelthough a complete discussion of the topic is beyond thecope of this chapter, it is instructive to briefly review some ofhe basic principles of software fusion; a thorough review ofoftware fusion methods can be found in Hawkes et al.2 Fu-ion of 2 image sets is achieved either by identifying common
andmarks or fiducials that can then be aligned or by opti- c
izing a metric based on image intensity values. Whateverhe method, the number of possible degrees of freedom be-ween the 2 image volumes defines the complexity of theubsequent transformation. For distributions that do not in-olve a change in shape or size, rigid-body transformationsre adequate. When shears (or a nonisotropic dilation with-ut shear) are involved, an affine transformation comprisinglinear transformation and translation is indicated. When
here are no constraints on the deformation, a nonlinearransformation (warp) is used. Although methods involvinghe alignment of extracted features or fiducials have shownome success, at least for the brain, most current methods arentensity-based and images are coregistered by assessing thentrinsic information content. Metrics include intensity ra-ios3 and mutual information.4 Although such techniquesave shown great success in aligning images of the braincquired with CT, PET, SPECT, and MR, they have been lessuccessful for other parts of the body. Earlier clinical assess-ent in the lung5 was disappointing, demonstrating a local
egistration accuracy of 5 to 8 mm, compared with an accu-acy of 2 mm for the brain.6 A recent review7 suggests thatoftware fusion can achieve an accuracy of 2 to 3 mm forome studies.
Commercially available software has improved consider-bly during the past several years both in the accuracy of theegistration algorithms and in the sophistication of the usernterface and display. As an example, Hermes Medical Solu-ions (Stockholm, Sweden) offers advanced fusion softwareor many clinical applications, including correction of mis-lignment errors for PET/CT scans, registration of PET/CTcans with MR scans, registration of longitudinal PET/CTtudies, alignment of PET and MR scans in Alzheimers dis-ase and other forms of dementia, and registration of SPECTr PET myocardial perfusion studies with CT or MR scans ofhe heart. However, despite considerable progress, fusionoftware will probably never compete with the simplicity andonvenience of coregistered studies acquired on a combinedET/CT scanner.
ultimodality Prototypeshe pioneering work of Hasegawa and colleagues in the late980s8,9 set the stage for the hardware solution to imageusion. The aim of this work was to design a device that coulderform emissio