Basic Physics

PET is an imaging modality derived from the unique physics of radioactive decay of positron-emitting radionuclides. Such non-natural isotopes, including ¹¹C, ¹³N, ¹⁸F, ⁶⁴Cu, and ¹²⁴I (Table 9-1), are unstable and decay through the conversion of a proton within the nucleus into a neutron, a neutrino, and a positron that is subsequently emitted from the atom. This decay process reduces the atomic number of the atom by 1 and ejects the positron from the nucleus. The emitted positron has a mean travel distance, dependent on the medium and its energy, in the range of 1 to 10 mm. After traveling some distance, the positron annihilates with its opposite particle the electron, and their energy is converted into two “annihilation” photons of 511 keV emitted in opposite directions.

To detect and localize such events, PET devices were developed as arrays of detectors that can convert the high-energy photons into optical photons through scintillation crystals, such as bismuth germanate and lutetium oxyorthosilicate, and subsequently into electrical pulses using photomultiplier tubes. Unlike similar devices for single-photon emission computed tomography that must rely on physical collimators to restrict the field of view of each detector to a defined spatial volume, PET scanners operate on the principle of coincidence detection. Because each positron emission event results in two photons traveling at 180 degrees to each other, PET detector arrays are designed to identify pairs of detectors that report events within a short (typically 6-20 nanoseconds) window of each other. These pairs of events are termed coincidence events. A typical PET acquisition involves the measurement of millions of coincidence events from a subject containing a distribution of a positron-emitting compound. This detection strategy is demonstrated in Figure 9-1. The concept of PET scanning was conceived in the late 1950s and early 1960s,⁴ resulting in the advent of a clinically useful scanning device in the early 1970s.^5,6 Modern clinical PET scanners achieve spatial resolutions of 4 to 6 mm, with sensitivities in units of counts per second for a 1 µCi point source at the center of the scanner of 200 to 2000.⁷

FIGURE 9-1 • Positron emission and its use in tomography. An unstable radionuclide (here, ¹⁸F) decays via the emission of a positron from the nucleus. The positron travels some distance before annihilating with an electron, resulting in the production of two 511 keV traveling in opposite directions. A detector array may be used to record the arrival of two “coincidence photons” within a short time of each other.

The performance of modern PET scanners is dictated by a number of factors. To conserve momentum in the positron-electron annihilation, the directions of travel of the annihilation photons are not perfectly collinear, with the divergence from collinearity approximately 1 degree or less. This slight noncollinearity between the emitted photons limits the resolution of PET, becoming more severe with increasing detector bore sizes. In addition, thick scintillator crystals are typically used in detectors to increase their photon-stopping ability. However, the thickness of the crystal can cause uncertainties in localizing events, particularly for detector pairs whose joining lines are at oblique angles to the detector face (such as detector pairs on the same side of the circular array). This causes a loss in spatial resolution for regions of the image away from the center of the scanner. Emerging “depth of interaction” detector technology has been developed to estimate where within a crystal an event occurred to mitigate this effect.^8,9 Scattered photons detected by the PET scanner as well as random coincidences occurring when photons from two separate positron events are recorded as a coincidence affect both the quantitative ability of PET as well as the spatial characteristics of resulting images. Strategies for random coincidence corrections and scatter rejection have been developed to combat these phenomena, and are discussed in the Image Reconstruction and Analysis section.

A recent achievement in PET scanner design has been the realization of time-of-flight (TOF) PET systems. From the inception of PET, engineers have striven to construct detectors with temporal responses fast enough to not only identify coincidence photons measured on separate detectors, but also to quantify the time lag between the arrival of each of the coincidence photons. This time can then be used to calculate where along the line connecting the two detectors the positron event occurred, increasing the quality of reconstructed PET images. One vendor now has a TOF PET-CT system available, the performance of which has been favorably compared with conventional PET scanners.¹⁰ In addition, combination of PET imaging technology with respiratory monitoring devices has resulted in the acquisition of motion-resolved, four-dimensional (4-D) PET data sets using methods analogous to 4-D CT.^11,12

Integration With Computed Tomography

Because the images obtained with PET are functional in nature and of spatial resolution significantly less than conventional modalities such as CT and MRI, interpretation of these data has traditionally been hindered by the lack of anatomic information. Although PET can be retrospectively registered to CT or MRI images acquired in subsequent patient examinations to provide an anatomic reference, an alternate solution to this problem has been to incorporate CT imaging hardware into a PET scanner, producing a hybrid modality that has been termed PET/CT. The architecture of these scanners generally involves the coupling of a helical CT scanner ring with a PET detector ring, offset along a coincident bore through which the patient table translates. A PET/CT acquisition typically involves first the acquisition of a whole-body helical x-ray CT scan, which can then be used to select a region for subsequent PET scanning. Therefore PET/CT is not a new imaging method by itself, but rather a scanning platform on which coregistered PET and CT data can be acquired in a single examination.

The principal advantage of PET/CT is the ability to interpret the PET images in the context of the anatomic reference provided by a high-spatial-resolution CT data set. This has been shown clinically to enhance the diagnostic interpretation of these data beyond that possible through consideration of the PET or CT in isolation.¹³ However, the integration of PET and CT has yielded other benefits as well. A key technical improvement afforded by PET/CT is the ability to correct the PET data for photon absorption within the subject, using the CT image to estimate the anticipated photon attenuation along the line joining each detector pair. Whereas attenuation correction was previously performed in standalone PET scanners using transmission images acquired with the PET detectors using a rotating rod source, the high signal-to-noise ratio and spatial resolution of CT images has significantly enhanced the performance of this correction. Furthermore, the development of hybrid PET/CT scanners has accelerated the inclusion of PET data in radiation treatment planning because it is possible using these devices to acquire coregistered PET data at the time of CT simulation.

Image Reconstruction and Analysis

The final acquired PET data set can be represented in two general forms: the total number of coincidence events measured for each detector pair, or a full list of every coincidence event along with the time at which is was recorded. The latter data type, termed list-mode, is useful for correlation with respiratory information to reconstruct 4-D PET data. However, after assignment of 4-D PET data into respiratory bins, the data are then integrated so that the total number of events measured for each detector pair is calculated. This information is calibrated for variations in detector response using normalization data acquired at regular quality assurance sessions. To reduce the influence of random coincidences, a “delayed window” technique is generally applied to estimate the count rate for each detector pair that is due solely to these spurious coincidence events. This procedure involves looking for coincidences triggered by the activation of a detector using a window of identical duration but that begins much later (50-100 nanoseconds). By calculating how many coincidences occur in this delayed window for a given detector pair, the random count rate may be estimated and subtracted from the actual count rate for the detector. Coincidences from scattered photons may potentially be rejected by using a narrow detector energy window; however, to maximize sensitivity, most PET scanners use broad windows (250-750 keV). The most common scatter correction involves estimation of the scatter contribution from measured projection data and subsequent deconvolution of this function from the acquired information. Attenuation correction as discussed previously is now typically implemented using the CT data acquired on a combined PET/CT scanner. By comparing the measured x-ray fluence for a given line-of-response (LOR) with and without a subject placed in the bore, the attenuation along that line can be estimated and used to correct the count rate measured for a detector pair sampling that LOR. Because the photon energy used in x-ray CT (≈ 150 keV) is significantly less than that of the annihilation photons measured in PET (511 keV), an adjustment must be applied to the attenuation calculated from the CT data to approximate the corresponding attenuation of the annihilation photons.

The traditional method of reconstruction for PET data has been the filtered backprojection technique, analogous to that used in x-ray CT. This method involves filtering the projection data using a ramp filter in the frequency domain, and uniformly distributing each filtered projection along the projection direction. This “backprojection,” when applied to a complete tomographic data set, then gives a reconstructed image. The sensitivity of this algorithm to high-frequency noise necessitates further filtering at the cost of spatial resolution. More computationally expensive iterative reconstruction techniques have evolved to address the shortcomings of analytic backprojection methods. In general, these methods operate through successive adjustments of the reconstructed parameter distribution, which is then subjected to a forward projection operator producing the corresponding expected measured data set that can be compared directly with the experimentally measured data. This strategy allows modeling of the acquisition process and can account for attenuation, scatter, noise, and other experimental factors. Examples of reconstruction methods of this type now in commercial use include maximum-likelihood expectation maximization,¹⁴ ordered-subset expectation maximization,¹⁵ and row-action maximization-likelihood¹⁶ algorithms. Following reconstruction of the data into spatial images, a system calibration factor determined from quality assurance procedures is then applied to convert the images from units of counts per second into units of activity per unit volume (µCi/mL).

Positron Emission Tomography Radiotracers

One of the strengths of PET is that the scanner technology is not specific to any radionuclide or compound; it can be used to map the distribution of any positron-emitting, radionuclide-containing compound with a subject. Therefore, in addition to the design and enhancement of PET scanner technology, a comparable research effort exists to design PET radiopharmaceuticals to image specific biologic and physiologic parameters. Because of the exquisite sensitivity of PET, radionuclide-containing compounds used with this imaging modality are generally termed radiotracers because they are given in trace doses so as not to perturb the physiology on which they are designed to report. PET radiotracers are typically delivered via systemic (intravenous) injection some time before scanning is to be performed, allowing distribution and specific localization of the agent prior to image acquisition.

The prevalent radiotracer in use in PET imaging is 2-deoxy-2-¹⁸F-FDG. FDG is an analog of glucose in which the hydroxy group typically found in the second position of the six-member ring has been replaced with ¹⁸F.¹⁷ Originally developed as an anticancer agent, 2-deoxyglucose (2DG) was designed to be cytotoxic to tumor cells that depend on glycolysis for energy production.¹⁸ This compound is taken up into cells by glucose transporters, and undergoes phosphorylation by hexokinase in the first step of glycolysis. However, because of the absence of the 2-hydroxyl group, the molecule cannot be further metabolized, and because at this stage the molecule has been phosphorylated and acquired a charge, it cannot easily cross the cell membrane and be cleared. This “trapping” of FDG within glycolytically active cells therefore generates a PET signal that can be imaged. Although 2DG as a cancer therapeutic has suffered largely because of the large doses required to out-compete endogenous glucose and the normal tissue toxicities thereby incurred, FDG has proven successful as a PET radiotracer because the trace doses administered avoid toxicity and because the uptake of this compound by tumors can be as much as 20 times that of surrounding normal tissue. A typical clinical dose for human imaging is up to 15 mCi of FDG at a specific activity of greater than 200 mCi/µmol, corresponding to a maximum mass dose of FDG of 75 nmol.

Although FDG PET now has proven clinical applications in a variety of oncologic sites, it has several shortcomings. Observation of elevated FDG uptake on a PET scan of a cancer patient could be associated with (1) increased tumor mass or cellularity, (2) upregulation of glucose transporters, (3) upregulation of hexokinase, (4) inflammatory responses, or (5) muscle activity. Therefore, although FDG PET can provide valuable clinical information, there have been subsequent efforts to develop more specific PET radiotracers that can report on specific aspects of an individual patient’s cancer and more fully characterize it so as to select and monitor the most effective course of treatment. ¹⁸F-fluorothymidine (FLT) is a thymidine analog that is taken up into cells and phosphorylated by thymidine kinases, resulting in a trapping similar to that of FDG.¹⁹ Although FLT is not incorporated into deoxyribonucleic acid, the association of thymidine kinase activity with S-phase cells suggests that FLT uptake gives a measure of cellular proliferation. FLT PET images acquired from human lung tumors have been correlated with proliferation measured using Ki-67 immunohistochemistry.²⁰

An area of PET tracer development with great relevance to radiation oncology is that of agents targeting hypoxic cells. A number of compounds employing the “cellular trapping” localization scheme have been synthesized and evaluated for their ability to accumulate in cells lacking oxygen. A number of these are based on 2-nitroimidazole–containing groups, which are reduced by cellular reductase enzymes to reactive species that bind covalently to intracellular macromolecules and are thus trapped. These reduction reactions can be reversed by molecular oxygen, resulting in radiotracer trapping occurring primarily in hypoxic cells.²¹ Radiotracers of this type that have been evaluated in the clinic include ¹⁸F-fluoromisonidazole,^{22 18}F-fluoroazomycin arabinoside,^23,24 and 2-(2-nitro-1H-imidazol-1-yl)-N-(2,2,3,3,3-pentafluoropropyl)-actamide (18F-EF5).^25,26 64Cu-Diacetyl-bis (N4-methylthiosemicarbazone) (64Cu-ATSM) is another hypoxia-selective PET radiotracer, employing a longer-lived ⁶⁴Cu radionuclide label, which employs bioreductive pathways causing the molecule to become charged to induce intracellular trapping.^27,28

In addition to radiotracers that become trapped within cells, a second class of PET agents are those that bind to specific proteins expressed on the surface of cells. Radiotracers of this type are common in neurologic studies of neuronal receptors; however, this localization method has frequently been applied to oncologic imaging as well. An emerging radiotracer of this type is arginine-glycine-aspartic acid peptides labeled with ¹⁸F or ⁶⁴Cu. These compounds have been shown to bind to α_ςβ₃ integrins expressed on endothelial cells of angiogenic vessels, allowing imaging of the molecular aspects of angiogenesis using PET.^29,30 In addition to small molecule probes, labeled antibodies against a variety of targets have been developed as PET radiotracers,³¹ providing a mechanism for translating antibodies developed for molecular biology applications into imaging agents. A key design constraint for receptor-binding PET radiotracers is that they must have high specific activity so as to avoid competition for binding sites between labeled (detectable) and unlabeled (undetectable) probe molecules within the radiotracer formulation.

A final class of PET radiotracers deserving mention are those that are used to measure vascular perfusion, such as ¹³N-NH₃ and ¹⁵O-H₂O. The short half-lives of these compounds as well as their biodistributions limit their use in PET experiments to short time frames after injection. Commonly, dynamic images are acquired rapidly after injection and fit to compartmental models to extract estimates of vascular volume or perfusion from the measured time-activity data.^32–34

Basic Physics

The properties that form the foundation for all nuclear magnetic resonance (NMR) techniques are (1) an observable macroscopic magnetization induced in populations of nuclei with the quantum property of “spin” when exposed to a magnetic field, (2) the frequency of precession of the magnetization vector is linearly related to the field experienced, and (3) the immediate chemical environment of a nucleus influences the magnetic field that it experiences. Although MRI leverages the large concentration of protons (nuclear spin ) in water in biologic tissues to encode spatial information and create an image, MRS exploits the latter property given previously to differentiate protons or other nuclei with quantum spin based on their molecular association. This technique forms the basis of the use of NMR techniques for the identification and measurement of chemical structure. Although MRS can be used to study any nuclei possessing quantum spin, including ¹H, ¹⁹F, and ³¹P, the proton (¹H) has the highest intrinsic sensitivity for nuclear measurement and has been studied in the most detail in vivo, and is the subject of this discussion.

Magnetic resonance pulse sequences used for the acquisition of spectroscopic data began with the simplest “pulse and measure” technique in which a 90-degree excitation pulse is applied to a sample under a strong static magnetic field, and the resulting free induction decay signal is recorded using an inductive coil. More sophisticated spin echo sequences employing a subsequent 180-degree refocusing pulse³⁵ were developed to avoid signal losses caused by field inhomogeneities and other static disturbances characterized by the relaxation rate T2*. To apply MRS to living subjects, a number of further technical issues must be resolved. Although it is feasible to obtain a homogeneous sample of a chemical for NMR analysis, for in vivo subjects, a mechanism of restricting the imaging volume is required. In the most straightforward sense, this can be achieved by using acquisition hardware with a limited sensitive volume. Surface coils have been applied in this regard as their transmission and reception sensitivity is very high near the coil aperture but falls off quickly.³⁶ More flexible methods have been developed to select volumes for spectroscopic measurement through software rather than hardware. This can be done in a manner identical to that done for slice selection in MRI, through the marriage of excitation pulses with a well-defined bandwidth and linear magnetic field gradients applied along a specific spatial dimension. Biologic spectroscopy acquisitions can apply three sequential selective excitations to excite orthogonal planes of spins, resulting in the final selection of a rectangular volume of interest. This technique can be implemented using either three 90-degree pulses, termed stimulated echo acquisition mode (STEAM),³⁷ or by one 90-degree and two refocusing 180-degree pulses, termed point resolved spectroscopy (PRESS).³⁸ PRESS is unable to achieve the short echo times (TEs) of STEAM because it uses multiple refocusing pulses, but it has the advantage of being a true spin echo technique and therefore insensitive to T2* signal losses.

In addition to the problem of suppressing signals from spins outside the volume of interest, in vivo MRS also faces the difficulty of removing signals from spins present in compounds other than those of biochemical interest. The largest of these is water, which is present in biologic tissues in molar concentrations and is several orders of magnitude more abundant than the metabolites of interest. Such an overwhelming signal can dominate the dynamic range of an MR system and prevent accurate quantitation of other less intense spin resonances. Strategies for suppressing the water signal from spectroscopic acquisitions focus on either destroying the coherent magnetization produced by water protons or on excluding the resonance frequency of water protons from the bandwidth of the excitation pulse. The former technique has been achieved through use of chemical shift selective saturation (CHESS) pulses, which consist of a 90-degree excitation pulse, specific to the resonance frequency of water protons, followed by a strong dephasing gradient pulse.³⁹ CHESS methods can be expanded to other potentially unwanted resonances such as lipid by modification of the pulse bandwidth or inclusion of additional CHESS pulses prior to measurement. The alternate approach, that of excluding the unwanted resonances from the bandwidth of the excitation pulses, can be achieved using frequency-selective techniques analogous to those described for STEAM and PRESS. However, the difficulty of simultaneously selecting both a restricted spatial volume (in conjunction with magnetic field gradients) and a restricted spectral range using a single pulse is apparent. This problem has been solved through the development of spectral-spatial selective pulses that incorporate complex varying gradient and pulse waveforms.⁴⁰

Although single-voxel spectroscopic acquisitions remain useful for a variety of diagnostic purposes, application of MRS in radiation oncology has been contingent on the development of methods for imaging spectroscopic signals. This goal is complicated by the dependence of spin resonance frequency and hence chemical shift information on the applied magnetic field, which prevents the use of frequency-encoding readout gradients like those used in MRI. The simplest solution to this problem is to apply a series of phase-encoding gradients along the spatial axes prior to the acquisition of an echo, and to step the intensities of these gradients so as to map out frequency space, or “k space,” to the desired extent and resolution. This procedure, termed chemical shift imaging (CSI) or magnetic resonance spectroscopic imaging (MRSI), has been applied to acquire MR spectra from two- and three-dimensional (3-D) grids of voxels. The time required for such acquisitions is given by the number of acquired voxels multiplied by the repetition time of the sequence. Faster spectroscopic imaging solutions have been pursued by applying time-varying gradients during readout and deconvolving the contribution of these gradients from the measured spectroscopic information during reconstruction,⁴¹ in an approach similar to that of echo-planar MRI. This technique, termed fast or spiral CSI, can reduce the acquisition time required for spectroscopic imaging at the expense of more pronounced susceptibility and aliasing artifacts also encountered with echo-planar MRI.

With the exception of surface coil techniques, MRSI does not require any specialized hardware beyond a standard clinical MR scanner. Both single-voxel spectroscopy and MRSI pulse sequences are now available on all commercial MRI platforms.

Image Reconstruction and Analysis

The raw data obtained from a spectroscopic acquisition must first be processed, corrected, and reconstructed. Spectroscopic imaging data must first be Fourier transformed along each phase encoding direction to separate the data into spatial voxels. Because of the nature of the frequency domain acquisition of MRSI data, the relatively coarse voxels can be retrospectively positioned and reconstructed within the excitation volume by multiplying the acquired data array by a complex exponential. This data is then Fourier transformed to generate a frequency spectrum, which is then subjected to a variety of corrections for spectral frequency and baseline and peak phase.⁴² When this reconstruction and correction procedure has been completed, the peaks of interest within the acquired spectra may then be identified and peak parameters (height, area, width) calculated.

A general difficulty in interpreting MRS relative to other molecular imaging modalities is that the measured data includes both spatial dimensions as well as a spectral, frequency dimension. A variety of classification techniques have been reported for discriminating normal from neoplastic tissues on the basis of spectroscopic information.^43–49 Of these, the work of McKnight and colleagues is particularly relevant to radiation oncology because it condenses the data contained in a spectrum into a single metric, or alternately translates a 4-D MRSI data set into a 3-D volume image more immediately familiar to modern radiologists and clinicians. This is accomplished for brain MRSI data by performing a regression analysis using the choline and N-acetyl aspartate (NAA) levels calculated for all voxels in an MRSI data set, and classifying each spectrum based on the distance its choline and NAA levels lie from the regression line calculated for an automatically selected population of “normal” spectra. This distance is called the z score or abnormality index, and serves as a measure of the deviation between a voxel’s spectrum and that of normal tissue.

Site Restrictions

Several phenomena limit the application of MRSI methods to several well-suited anatomic sites. First, the tissue of interest must contain a metabolic or molecular signature that makes it worthwhile to investigate with spectroscopy. To be detected by in vivo MRS, a metabolite must be present in appreciable concentrations to produce a detectable signal. In general, this requirement necessitates millimolar concentrations for the spin groups of interest.^50,51 Furthermore, the transverse relaxation time (T2) of the spin population must be sufficiently long relative to the TE used in the acquisition sequence to avoid complete relaxation of the magnetization before signal measurement. This precludes detection of macromolecules or compounds that are tethered to large complexes, such as membrane-associated molecules. Finally, the peaks of interest must be adequately removed from other peaks present in the spectrum so that they can be independently resolved and quantified. In addition, the macroscopic structure of the tissue must be amenable to the acquisition of high-quality spectroscopic data. This mandates that it be of relatively homogeneous magnetic susceptibility. In addition, because of the relatively long duration of MRS acquisitions, the tissue should not be subject to excessive motion.

The brain is particularly suited for interrogation with ¹H MRS techniques. It is a relatively static organ and is largely free of air–tissue interfaces that can introduce difficulties in producing a constant magnetic field across it. A long TE spin echo acquisition sequence can detect five basic types of metabolites in the brain: choline-containing compounds, creatine, NAA, lactate, and mobile lipids, as shown in Figure 9-2. Choline levels in glioma are generally increased relative to normal cerebrum^52–59 reflecting altered membrane phospholipid metabolism^54,59,60 or increased cellularity in neoplastic tissue.^54,59–61 High-grade (World Health Organization grade 4) gliomas do not show the same magnitude of choline elevation as lower grade gliomas; this is thought to be related to the micronecrotic nature of these lesions.⁵⁴ The appearance of the MRS creatine peak in brain neoplasms appears to be inversely correlated to tumor grade, with grade 4 lesions demonstrating the most reduced creatine levels relative to normal tissue.⁶⁰ NAA has been shown to be localized in neuronal cells,^62–64 allowing its use as an MRS marker of viable neurons. Dramatic reductions in the MRS NAA signal are observed in studies of gliomas, which are indicative of the death of neurons as they are encroached upon by the invading neoplasm, or the displacement of neuronal tissue by tumor-related mass effects.^52–59 The ability to simultaneously observe both a tumor-associated resonance (choline) as well as a marker associated with normal tissue (NAA) enhances the diagnostic ability of MRS. Lactate and lipid protons both resonate in the 1.0 to 1.5 ppm range. Lactate, the end product of anaerobic glycolysis, has a controversial status in clinical brain MRS; some have associated its presence with poor clinical outcome indicative of tumor hypoxia,⁶⁰ whereas others have noted that lactate accumulation is nonspecific and can be either an intrinsic property of the tumor or an effect of treatment-induced ischemia.⁵⁴ Lipid signals in brain MRS can either come from spectral contamination from subcutaneous fatty tissues included in the MRS volume, or from micronecrosis and cellular breakdown present in high-grade or post-treatment lesions.^56,65 Lipid resonances may be an indicator of membrane breakdown and necrosis, potentially reflective of the necrotic nature of glioblastoma multiforme. Increasing magnetic field strength and decreasing echo times and spatial resolutions have facilitated the detection of an increased number of metabolites from brain tissue, including glutamine, glutamate, taurine, inositol, and aspartate.⁶⁶

FIGURE 9-2 • A long echo (echo time = 144 ms) 1H magnetic resonance spectrum acquired from a 1-cc voxel in the cerebrum of a normal volunteer. The water peak is suppressed and clipped for clarity. Spectral resonances from protons in choline, creatine, and N-acetyl aspartate are visible. Protons in lactate and lipid resonate in the 1.0 to 1.5 ppm range and are not observed in normal brain tissue.

The use of transmission and reception coils inserted into the rectum has allowed the acquisition of high signal-to-noise ratio MRI and MRS data from the prostate. Long-echo MRS studies of the prostate allow quantitation of choline, creatine, and citrate as demonstrated in Figure 9-3. As in brain MRS, choline is a marker of altered neoplastic membrane metabolism, and the MRS peak it generates can similarly be used as a biomarker of cancer.⁶⁷ A peak containing contributions from polyamine compounds including spermine and spermidine is found at 3.1 ppm, but its detection can be complicated by broad line widths and by its proximity to the choline and creatine peaks, especially at 1.5 T. Decreased polyamine levels have been associated with malignant prostate cancer.⁶⁸ However, the principal marker of normal prostate metabolism in spectroscopic measurements is citrate, present in the MR spectrum as a quartet centered at 2.6 ppm. Citrate is produced, accumulated, and secreted by normal epithelial cells within the gland. During the process of malignant transformation, prostatic cells become dedifferentiated and correspondingly lose their capacity for citrate production and secretion. This results in levels of MRS-visible citrate being reduced in neoplastic tissue relative to normal tissue.

FIGURE 9-3 • A long echo (echo time = 130 ms) 1H magnetic resonance spectrum acquired from a 0.3-cc voxel in the prostate of a prostate cancer patient in a region of normal-appearing tissue. Spectral resonances from protons in choline, polyamines, creatine, and citrate are visible.

Several groups have attempted to study breast cancer using spectroscopic methods. The problems of water and lipid suppression are exacerbated when measuring within the breast; however, improved techniques have proven adequate for the robust detection of relative choline levels across the breast.⁶⁹ However, in breast spectroscopy there is no easily detectable normal marker to play the role that NAA does in brain MRS and citrate does in prostate MRS.

Tumor Staging

FDG PET and MRS are the two molecular imaging modalities that have achieved the highest degree of adoption in clinical oncology. The value of ¹⁸F-FDG PET and PET/CT has been studied extensively for a variety of cancers in several clinical scenarios. Currently, ¹⁸F-FDG PET and PET/CT scanning is used routinely for the evaluation of several tumors that are typically avid for FDG. These malignancies are lung cancer, melanoma, breast cancer, esophageal and colorectal cancers, head and neck cancer, thyroid cancer, and lymphoma. MRS has more severe technical restrictions to its use, and is currently most studied in staging and treatment monitoring of breast cancer, brain tumors, and prostate cancer.

Lung Cancer

In the evaluation of lung cancer, ¹⁸F-FDG PET and PET/CT is helpful for the evaluation of lung nodules, as well as for the staging and restaging of established non–small cell lung cancer (NSCLC). As demonstrated by a meta-analysis by Gould in 2001, PET has a sensitivity of greater than 96% for detecting and characterizing malignant lung nodules greater than 1 cm, while maintaining an overall specificity of 77%.^{70 18}F-FDG PET is an excellent modality for staging mediastinal and hilar lymph nodes, with an overall sensitivity in the range of 85% and specificity of 80%. A meta-analysis that directly compared patients staged by both ¹⁸F-FDG PET and CT showed that PET is superior, with a sensitivity range of 85% to 95% and specificity range of 81% to 100% versus a sensitivity of 64% and specificity of 62% for CT.⁷¹ In addition, PET is well suited for detecting distant metastases because it is routinely performed as a whole-body examination (from the base of the skull to the mid-thighs) and may find distant metastases in as many as 10% to 25% of lung cancer patients. PET is particularly useful for evaluating metastatic disease within the adrenal glands, because they pose a diagnostic challenge on CT.⁷²

Melanoma

In contrast to lung cancer, ¹⁸F-FDG PET has little role for the initial screening and detection of melanoma because of a significant decrease in sensitivity for detecting lesions smaller than 7 mm.^73,74 However, ¹⁸F-FDG PET does have a role at initial staging for patients at high risk for metastatic disease.⁷⁵ This includes those patients with extensive locoregional disease at initial biopsy, a clinical history and examination suspicious for metastases, and those patients with known metastatic disease (for whom an alternative therapy may be sought if additional metastases are found). In a retrospective review of 257 patients with stage III and IV melanoma by Bastiaannet et al., 21.8% of the patients were upstaged as a result of PET findings and in 17.1% the treatment was changed from surgery to systemic therapy.⁷⁶ For those patients with early stage melanoma at initial diagnosis, the added value of PET over conventional imaging is likely limited. In a prospective randomized trial of 144 patients (139 had primary cutaneous melanoma and five had local recurrent melanoma), Wagner et al. noted that PET scanning did not affect the care of patients with early stage melanoma already staged with standard techniques and thus is not recommended for the initial staging in this population.⁷⁴

Breast Cancer

¹⁸F-FDG PET has a limited role in the initial screening of breast cancer largely because of two limiting factors: low sensitivity of PET for lesions smaller than 7 mm and low FDG avidity for lobular carcinomas and carcinoma in-situ.⁷⁷ In addition, PET is not as sensitive as sentinel lymph biopsy for the detection of axillary lymph node disease.⁷⁸ Although PET has a limited role in the detection of primary breast cancer and staging the axilla, it can be important in detecting distant disease and in helping to plan surgical and medical treatment.^79–84

Jacobs et al. reported a preliminary study of 18 patients with histologically verified breast disease,⁸⁵ in which useable MRS imaging data was acquired for 83% of the patients. Partial suppression of the lipid and water resonances allowed relative quantitation of choline as well as lipid and water signals across the breast. Choline levels in patients with malignant lesions were on average 2.5 times larger than those in patients with benign lesions. Similar results have been reported by Bartella and Huang, who have observed that the choline level measured by MRS may be an early indicator of breast tumor response to chemotherapy.⁸⁶

Esophageal Cancer

In esophageal cancer, the initial diagnostic workup includes evaluation by a combination of endoscopy, endoscopic ultrasound (EUS) and CT, with ¹⁸F-FDG PET playing an adjunctive role. ¹⁸F-FDG PET is used primarily for initial staging and evaluating response to neoadjuvant treatment for both adenocarcinoma and squamous cell carcinoma.⁸⁷ Direct comparisons of the staging sensitivity and specificity of PET, CT, and combination CT and EUS have yielded mixed results. Locoregional lymph node staging appears to be most accurate with EUS, with one comparison study reporting an accuracy of 75% for EUS versus 63% for PET.⁸⁸ For distant nodal and organ metastases, PET appears to have greater sensitivity than CT and a greater effect in appropriately upstaging patients who would have otherwise undergone unwarranted surgery. Overall, the optimal initial staging strategy may be the decision-analysis model proposed by Wallace et al. in 2002, which demonstrated that the combination of ¹⁸F-FDG PET with EUS and fine-needle aspiration biopsy is most effective.⁸⁹

Colorectal Carcinoma

Because of the presence of nonspecific normal variant bowel activity and its relatively high cost, ¹⁸F-FDG PET is generally not recommended for colorectal cancer screening, with some reports suggesting that premalignant and malignant lesions cannot be reliably detected by this modality.⁹⁰ In addition, PET is not effective in evaluating tumor depth or locoregional nodal disease (T and N staging, respectively). ¹⁸F-FDG PET does have a high sensitivity for detecting distant metastatic disease, particularly in the liver, and may alter the intent of surgical procedures (from curative to palliative measures).

Head and Neck Cancer

The use of ¹⁸F-FDG PET for the identification and characterization of head and neck tumors has been described primarily for squamous cell carcinoma. The primary utility of ¹⁸F-FDG PET in head and neck cancer is for the detection of locoregional nodal disease and distant metastases. Numerous studies that directly compared ¹⁸F-FDG PET to CT and clinical examination showed that PET has a higher sensitivity and specificity than either CT⁹¹ or physical examination for the initial pretreatment evaluation of metastatic lymph nodes of the head and neck. Furthermore, several reports also demonstrate that ¹⁸F-FDG PET and MRI have comparable sensitivities, but PET has superior specificity in evaluating metastatic lymph nodes. PET is most commonly used for those patients with suspected local lymph node metastases in which the identification of additional contralateral or distant malignant lymph nodes needs to be elucidated. However, in those patients with negative nodal (N₀) disease by clinical examination and CT or MRI, the role of PET is uncertain. In current practice, N₀ patients undergo minimally invasive sentinel lymph node biopsy to determine the presence of locoregional nodal metastases and subsequent need for an elective neck dissection. At least one prospective report shows the sensitivity of ¹⁸F-FDG PET is unacceptably low in detecting occult metastases when compared with sentinel lymph node biopsy. In patients with more advanced disease, PET is useful and superior to CT for detecting distant sites of disease as well as synchronous tumors.

Thyroid Cancer

FDG PET has little to no role in the initial staging of differentiated thyroid cancer. However, it may be used adjunctively for initial staging of anaplastic thyroid cancer. The standard imaging management of differentiated thyroid cancer is initial and follow-up ¹²³I scintigraphy after surgery and radioiodine therapy. During the follow-up phase of clinical management, FDG PET is performed in the setting of discordantly high blood levels of thyroglobulin (Tg) and negative ¹²³I scintiscans. There are many reports of the value of PET in thyroid cancer, with the first one as early as 1987.⁹² Several reports deal in particular with iodine-negative, Tg-positive patients.^93–101 Stokkel et al. have tabulated the results in 18 published series. The sensitivity of PET varies from approximately 50% to 100%, and the range is likely due to the degree of differentiation of residual cancer and the volume of tissue. In general, poorly differentiated cancers demonstrate higher uptake of FDG, whereas well-differentiated cancers that trap iodine might fail to be imaged.¹⁰²

Lymphoma

¹⁸F-FDG PET and PET/CT are commonly used for initial staging of Hodgkin and non-Hodgkin lymphoma.^{103,104 18}F-FDG PET has shown high accuracy in the early prediction of response to chemotherapy.¹⁰⁵ Therefore, PET may play a role in tailoring the intensity of the treatment to the individual patient.

Brain Tumors

FDG PET has a limited role in the management of brain tumors because of strong FDG uptake by normal cerebral tissue. However, brain MRS has a solid biologic basis and has been applied in a number of studies to improve diagnosis and staging of tumors, including differentiation of active tumor from radiation necrosis,^106,107 and discrimination of low-grade and high-grade gliomas.^108,109 A study of 176 intracranial lesions using single-voxel spectroscopy by Möller-Hartmann et al. revealed that MRS improved diagnosis beyond MRI alone, increasing the number of correct diagnoses by 15.4% and reducing the number of incorrect diagnoses by 6.2%.¹¹⁰ A separate study by Murphy et al. of 100 patients demonstrated that MRS provided useful diagnostic information in 6 of 26 patients in which the MRI and clinical information were inconclusive.¹¹¹

Prostate Cancer

FDG PET is not part of the standard workup for prostate cancer patients, both because of difficulties imaging a structure adjacent to the bladder and because of limited FDG uptake by prostate tumors. MRS studies of prostate cancer, however, have demonstrated that this molecular imaging modality can significantly enhance the diagnosis and staging of prostate tumors. A specificity of 91% has been reported for combined MRI and MRS imaging examinations,¹¹² whereas incorporation of MRI, MRS imaging, and step-section pathologic examination has demonstrated a higher sensitivity, specificity, and positive predictive value relative to standard sextant biopsy of the prostate.¹¹³ In addition, staging of prostate tumors has been shown to be improved by inclusion of MRS imaging data, with a study by Bartolozzi et al. reporting sensitivities and specificities of a combined MRI and MRS examination in diagnosing capsular invasion of 95% and 82%, and in diagnosing seminal vesicle invasion of 80% and 93%, respectively.¹¹⁴

Target Volume Delineation

Advances in radiation therapy technology have greatly improved many aspects of the physical delivery of radiation. CT-based treatment planning for creating a 3-D model of electron density in the body, accurate algorithms for calculating dose in tissues with heterogeneous density, inverse planned intensity modulation for creating highly conformal plans, immobilization technology, 4-D imaging and radiation delivery technologies for organ motion management, and image-guidance tools integrated into the delivery systems all are designed to ensure that intricately shaped dose distributions are delivered with high spatial accuracy and small tolerances. The potential advantage of such precision is to increase the therapeutic ratio, improving tumor control while sparing normal tissues, and implies that there should be a sharp dose gradient between targets and adjacent organs. A premise underlying the use of conformal therapy is that the appropriate targets can be defined with a precision similar to that of the treatment technique. In fact, interobserver differences in CT-based target delineation across numerous anatomic sites and tumor types are substantially greater than the conformity achievable by these newer treatment delivery techniques. The ability to visualize the distribution of malignant cells is a key to optimizing curative radiation therapy.

Molecular imaging methods have had an important effect on the field of diagnostic imaging because of their higher sensitivity and specificity for tumor detection versus conventional anatomic imaging modalities. When applied to radiation treatment planning, the main effect of more sensitive imaging is inclusion of smaller tumor deposits in the volume receiving the highest radiation dose, or the gross tumor volume (GTV) in International Commission on Radiation Units and Measurements terminology. This should improve tumor control by avoiding geographic misses of detectable tumor deposits. When using radiation therapy for curative intent, an important concept for many tumor types is the definition of a target volume for occult or microscopic tumor spread, or the clinical target volume (CTV), particularly in the regional lymphatic system. For this, we must rely on knowledge or extrapolation from clinical experiences regarding the likelihood of nodal metastasis or patterns of tumor recurrence based on the known GTV distribution. However, to the extent that the higher sensitivity of molecular imaging allows smaller and smaller tumor deposits to be included in the GTV, the tumor cell burden in the CTV is reduced, potentially allowing a lower dose to be administered to the larger target volume and improving the therapeutic ratio. This is a concept yet to be tested clinically.

FDG PET–Based Treatment Planning

To date, the main molecular imaging modality used routinely at many centers for radiation treatment planning is PET using ¹⁸F- FDG. From a technical standpoint, because the majority of PET scanners currently are combined PET-CT scanners, PET has become a natural extension of CT-based treatment planning. It can be performed with the patient in the treatment position using appropriate immobilization devices, and the metabolic information is automatically coregistered with the higher resolution anatomic information from CT. FDG remains the only clinically approved PET imaging agent, is readily available commercially, and has proven the most useful for cancer diagnosis and staging.

For these reasons, the vast majority of the literature on metabolic imaging for radiation treatment planning has focused on the use of FDG PET, particularly for the malignancies in which it has demonstrated the greatest utility for staging, the most studied being NSCLC. Ultimately, the gold standard against which to measure the utility of FDG PET–based treatment planning is improved cure rates, decreased toxicity, and improved overall survival compared with treatment planning based on CT alone. Such data are presently lacking. However, studies of PET-based treatment planning have consistently demonstrated two interrelated findings: interobserver variability in target definition is significantly reduced when PET information is used in addition to CT; and the use of PET alters the treatment plan in clinically significant ways compared with CT alone. In addition, preliminary data on clinical outcomes using PET-based treatment planning suggest that PET helps to identify the appropriate target volumes for radiation therapy.

Consistency of Target Delineation and Changes in Treatment Plans with PET

Definition of targets for radiation therapy involves clinical judgment and is prone to subjectivity. This is reflected in the observation in numerous planning studies that target volumes defined by different practitioners on the same scan can vary substantially. A component of this variability is related to differences in expertise and training, including knowledge of anatomy and patterns of common patterns of lymphatic drainage and recurrence risks for given anatomic sites and histopathologic features. Consistency in this regard has been enhanced by improved education efforts, including the publication of standardized anatomic atlases, contouring courses sponsored by professional societies, and standardization of window and level display parameters, for example. However, there remain important sources of variability based on image interpretation, for example, when criteria other than size are used to determine tumor involvement, when contrast between tumor and normal tissues is low (e.g., tumor adjacent to obstructive atelectasis in the lung, or when CT contrast enhancement is subtle), and when translating physical examination findings such as the degree of mucosal extension into contours on the scan.

PET provides an additional contrast mechanism for distinguishing tumor from normal anatomy compared with CT alone, and as such can reduce the variability in image interpretation. A number of studies have compared tumor volumes contoured by multiple observers based on CT alone compared with PET-CT. Two studies in lung cancer^115,116 and a study in rectal cancer¹¹⁷ demonstrated substantial improvement in interobserver contouring consistency when fused PET-CT images were used. Of note, in the Memorial Sloan-Kettering Cancer Center study¹¹⁶ the comparison was between visual incorporation of PET information versus fused PET-CT acquired in the treatment position, highlighting the importance of accurate fusion and PET imaging integrated in the actual treatment planning scan. That study also found improved consistency in intraobserver recontouring with fused PET-CT. On the other hand, a study in head and neck cancer found high interobserver variability whether contouring on CT or PET-CT that improved when a standardized protocol was implemented, indicating the importance of consistent training and guidelines.¹¹⁸

Given the known differences in sensitivity and specificity between CT and PET for tumor detection, significant changes in tumor volumes can be expected when using PET in addition to CT for treatment planning. One of the most significant changes is identification of previously unsuspected distant metastases that would make definitive therapy unfeasible. In a study at Peter McCallum Cancer Institute, metastatic disease was found in 27 of 153 patients initially thought to be candidates for radical radiation therapy for NSCLC.¹¹⁹ With respect to changes in target volumes, a literature review of PET-CT planning studies across multiple tumor sites found that the incorporation of FDG PET imaging information resulted in altered target volumes in between 10% and 100% of patients.¹²⁰ The discordance between PET and CT resulted in both exclusion of CT-based targets and inclusion of new targets. To the extent that PET-CT target definition is more accurate, this would be expected to reduce geographic misses and decrease unnecessary normal tissue irradiation and toxicity (Figure 9-4).

FIGURE 9-4 • Three-dimensional model of a patient with stage IIIB non–small cell lung cancer undergoing involved field conformal radiation therapy. The red volume represents the planning target volume (PTV) based on computed tomography (CT) with contrast, whereas the magenta volume represents the additional PTV included based on positron emission tomography-CT, which identified hypermetabolic mediastinal nodes that did not meet CT size criteria for pathologic involvement.

Preliminary Clinical Outcomes Data

At present, there are no available results from prospective randomized clinical trials to document such improved outcomes from FDG PET–based treatment planning. However, a number of studies of patterns of tumor recurrence following radiation therapy in patients who had PET-based treatment planning provide insight into this question.

The Peter McCallum Cancer Institute conducted a retrospective comparison of patients who received radical radiation therapy for NSCLC, 80 of whom had pretreatment FDG PET staging and 77 of whom did not.¹¹⁹ The median survival of these groups was 31 versus 16 months, respectively. This largely reflects stage migration introduced by PET, which screened patients with distant metastatic disease detectable only by PET from receiving futile radical therapy, but may also represent an improved outcome from better targeting of the GTV. A prospective Dutch study of 44 patients with stage I–III NSCLC receiving definitive radiation therapy targeting only PET-positive nodes without inclusion of elective nodal regions demonstrated only a single recurrence isolated to an initially PET-negative nodal region outside the radiation field (and only two others with concurrent relapse either in-field or distantly) with a median follow up of 16 months.¹²¹ A retrospective analysis of patterns of locoregional recurrence in 35 patients treated at M.D. Anderson Cancer Center with FDG PET–based definitive radiation therapy for NSCLC using recursive partitioning analysis identified standardized uptake value (SUV) greater than 13.8 as the strongest factor predicting recurrence on a per-lesion basis, followed by tumor volume larger than 1.2 mL and SUV less than 13.8.¹²² In a series of patients from Memorial Sloan-Kettering Cancer Center who had locoregional recurrences after definitive radiation therapy for NSCLC, the recurrences at the primary tumor site were predominantly within the GTV when lower doses (<60 Gy) were used, but at the margin of the target volume with higher doses (>60 Gy), whereas the nodal recurrences were primarily marginal or geographic misses.¹²³ These observations were interpreted to indicate the importance of high radiation dose for tumor control, and the inadequacy of visual incorporation of PET data (used in this series) for GTV definition, particularly for nodal targets. A retrospective study of failure patterns in head and neck squamous cell carcinoma at the University of Michigan found that 9 of 61 patients had locoregional failure, all within the targeted GTV, 8 of which were in PET-positive regions and 1 of which was in a PET-negative region included in the GTV based on other criteria.¹²⁴

Together, these suggest that the areas at highest risk of locoregional recurrence after radiation therapy are those that were initially targeted with the aid of PET-based treatment planning. The only prospective randomized clinical trial of this concept in locally advanced NSCLC is being conducted by the Ontario Clinical Oncology Group. In the Positron Emission Tomography Imaging in Stage Three Non-Small Cell Lung Cancer: A Prospective Randomized Clinical Trial, patients were randomized between CT- and PET/CT–based radiation treatment planning. Accrual has been stopped early because interim analysis revealed that the PET arm demonstrated superior efficacy for identifying patients for whom radical therapy was inappropriate because of upstaging, the primary endpoint of the study. Information on the effect of PET-CT on radiation treatment planning, a secondary endpoint, is anticipated.

Practical Considerations

There are well-established guidelines and protocols for the optimal acquisition of FDG PET/CT images for diagnostic purposes, including appropriate fasting, timing and administration of diabetic medications, uptake time and environment, interval from injection to imaging, and instrument calibration procedures. In addition to these, the use of PET as part of radiation treatment planning requires consideration of additional practical aspects specific to radiation therapy.

One key to the accuracy of highly conformal radiation therapy is the use of patient positioning and immobilization aids, such as custom masks or body molds. Whereas with CT-based treatment planning these are typically made just prior to scanning with the patient on the CT couch, the optimal timing in relation to PET-CT scanning is to make them earlier. Reproducible timing should be observed between injection of FDG and initiation of the scan. If the positioning devices are made at the scanner after the patient has been injected, this subjects the therapists to increased radiation exposure because of their prolonged proximity to the patient, and leaves less time to optimize the positioning or remake the devices if necessary. Thus it is preferable to form the positioning aids before the patient is injected, ideally on a conventional simulator that allows imaging verification of appropriate positioning. This leaves ample time to position the patient in the pre-formed devices just prior to image acquisition, and results in minimal radiation exposure to the therapists.

It is important that the scanner be fitted with a flat couch top to correspond to the therapy couch top. Alignment lasers and indexing devices for the positioning aids are valuable additions. PET-based target delineation is highly sensitive to the quality of registration between the PET and CT components of the scan. Even in an integrated scanner, and even using immobilization aids, there is potential for misregistration because of patient motion between the CT and PET portions of the scan. When scanning a long field of view encompassing multiple PET bed positions, it is helpful to sequence the acquisition such that the most critical anatomy for planning purposes is scanned first. For example, when treating head and neck cancer, a head-first scanning sequence results in a shorter time between the PET and CT at the region of greatest interest than a feet first acquisition order, resulting in less motion-induced registration error at the tumor site.

The resolution of PET is significantly lower than CT. Thus the PET portion of the scan is more useful for identifying which targets to treat than for delineating the edge of the target. For example, PET can be used to determine which lymph nodes should be included in the GTV, but the size and location of the target should be delineated based on CT. However, when there is lack of contrast between the tumor and adjacent normal organs on CT, it may be necessary to define the extent of the target based primarily on PET alone. Examples include lung tumors adjacent to obstructive atelectasis, and esophageal or rectal tumors in which it is difficult to determine the tumor extent based solely on wall thickening. In such cases, the targets should be contoured based on the area of hypermetabolism on PET, with somewhat generous margins where the interface between tumor and normal tissues is unclear on CT, to account for the uncertainty in both PET- and CT-based edge definition.

It is a point of controversy whether it is appropriate to exclude from the target lymph nodes that are abnormal by CT criteria but not hypermetabolic on FDG PET. Necrosis in nodes is one factor that can result in false-negatives on PET. A conservative approach is to target nodes that are abnormal by either CT or PET criteria, particularly if they are located in high-risk areas (adjacent to or bridging clearly pathologic lymph node regions). However, in lung cancer, for example, in which there is an emerging trend to omit elective nodal irradiation to permit dose escalation to known macroscopic tumor deposits, some have advocated excluding PET-negative nodes on the basis of the high negative predictive value of FDG PET.

Particularly in the skull base, MRI can detect perineural enhancement or marrow space changes indicative of tumor extension along nerves or into bone, respectively, that FDG PET frequently fails to detect (Figure 9-5). It is possible that such changes represent reactive changes to microscopic tumor rather than the presence of macroscopic tumor, and pathologic confirmation generally is not available, but we recommend including such areas in the GTV. This illustrates the importance of considering all available information when determining target volumes.

FIGURE 9-5 • Positron emission tomography–computed tomography (PET-CT) (left panel) and magnetic resonance imaging (MRI) (right panel) of a patient with locally advanced nasopharynx cancer. Of note, sclerotic changes in the clivus on CT and marrow space enhancement on MRI indicate tumor involvement not detected by PET. The gross tumor volume (red contour) should be based on all available information on tumor extent, including other imaging modalities and physical examination.

Finally, because upstaging (or downstaging) by PET often implies important differences in overall prognosis, biopsy of relevant sites of PET positivity should be considered if the management would be dramatically altered on the basis of the PET findings. Particularly when a patient otherwise fit for definitive therapy is found to have a site of suspected distant metastatic disease, confirmation is warranted before deeming the patient to have incurable disease.

Magnetic Resonance Spectroscopy–Based Treatment Planning

Because of the common hardware used to acquire both MRI and MRS data, the role of MRS in radiation treatment planning is enhanced by the coregistration of functional spectroscopic data with anatomic MR images that are acquired as part of the staging or treatment-planning process for many cancer types. However, the widespread adoption of MRS as a component of radiation treatment planning has been slowed by several factors. It is generally known that irregularities in the magnetic field of an MR scanner can cause distortions in reconstructed MR images and correspondingly in MRS data. Although it does not significantly affect the diagnostic quality of these images, this warping can be problematic for radiotherapy planning in which precise target localization is needed for beam delivery. Furthermore, use of MRS data to define spatial treatment volumes is complicated by the poor spatial resolution of MRS relative to MRI as well as to PET, as well as the need to combine MRI and MRS data with the CT needed for dosimetric calculations during radiotherapy planning.¹²⁵ Nonetheless, preliminary investigations of the utility of MRS in planning radiation therapies for brain and prostate tumors have shown promising results.

A study of 36 patients with recurrent gliomas treated with gamma knife radiosurgery showed a significant association between change in tumor volume, progression-free survival, and overall survival with the extent of the pretreatment MRS-abnormal region relative to the radiosurgical target.¹²⁶ This suggested that MRS might be used to define the radiation target to treat regions of glioma not apparent on MRI. A subsequent planning study of 34 glioma patients indicated that the contrast-enhancing volume apparent on T1-weighted MRI is considerably smaller than the volume of abnormality visible with MRS imaging, and that although the abnormal volume seen on T2-weighted MRI is of similar size to the MRS abnormality, the MRS abnormality extended beyond the T2-abnormal volume in 88% of patients by a maximum extent as great as 28 mm.¹²⁷ Similar results have been obtained and correlated with patterns of failure for patients treated with 3-D conformal radiotherapy.¹²⁸ Incorporation of MRS into radiotherapy planning may therefore significantly alter the treatment plan for malignant gliomas, although results of prospective studies employing such planning methods have yet to be obtained.¹²⁹

In the prostate, MRS has been applied in the treatment planning context for the guidance of brachytherapy implants as well as intensity-modulated radiotherapy. A planning study of a patient with a Gleason 7 prostate cancer demonstrated that MRS imaging may be used to optimize the placement of brachytherapy seeds, allowing dose escalation to the dominant intraprostatic lesion and resulting in improvements in calculated tumor control probability.^130,131 The use of MRS imaging to detect extracapsular extension and to predict response to external-beam radiotherapy has been demonstrated.¹³² Several studies have demonstrated how the identification of abnormal voxels on a prostate MRS imaging examination can be used to direct boost doses to tumor-associated regions within the prostate.^133–135

Monitoring Tumor Response