Principles of Operation

CT is an x-ray imaging technique used to visualize thin slices of the body. The first commercial CT scanner was developed by Sir Godfrey Hounsfield of the EMI Corporation in 1972. Alan Cormack developed an accurate mathematical technique to reconstruct images from x-ray projections. Hounsfield and Cormack received the 1979 Nobel Prize in medicine for their efforts. Modern scanners employ a pure rotation motion and require only 0.4 to 1.0 second per rotation to acquire the data (Fig. 8-1). Multislice detector arrays now allow as many as 320 slices to be acquired in one 0.35 second rotation.

FIGURE 8-1 • Schematic drawing of a third-generation scanner showing the x-ray tube, collimator, detector array, and data acquisition system. The four subsystems are rigidly attached and go through a single rotational movement during a scan. Their weight typically is on the order of 2000 pounds. Also shown are outline drawings of the patient couch and gantry.

(From Barnes GT, Lakshminarayanan AV: Computed tomography: physical principles and image quality consideration. In Lee JKT, Sagel SS, Stanley RJ, [eds]: Computed body tomography with MRI correlation, ed 2, New York, 1989, Raven Press, p. 7.)

A thin beam of x-rays of about 120 kilovolts (kV), in the fan-beam geometry, is incident on the patient transversely. The transmitted x-rays are detected by an array of many solid-state detectors. Concomitantly, the x-ray source and the detector array are rotated through an angular range from 180 to 360 degrees. The detected x-ray transmission factors are input data to a computer program, the so-called reconstruction algorithm, which produces an output matrix (usually 512 × 512) of digital values. Current scanners using workstations and the filtered-back projection algorithm can reconstruct an image in a fraction of a second. Each element of the matrix represents a small area, or pixel; the product of a pixel area and the slice thickness constitutes a volume element, or voxel (Fig. 8-2). The digital value reconstructed for each pixel (or voxel) represents the linear attenuation coefficient (approximately equivalent to the electronic density) of that volume. The digital values are converted to gray-scale levels for display on a video monitor screen or are used as input to a PACS workstation or to a laser imager for hard-copy output on film. The contrast of the image may be adjusted by window and level settings that determine the range of attenuation values displayed within the gray-scale range of the output device.

FIGURE 8-2 • Pixel value in a computed tomographic image represents the linear attenuation coefficient of a right square prism volume element, or voxel (i.e., pixel area ∞ slice width).

(From Barnes GT, Lakshminarayanan AV: Computed tomography: physical principles and image quality considerations. In Lee JKT, Sagel SS, Stanley RJ, [eds]: Computed body tomography with MRI correlation, ed 2, New York, 1989, Raven Press, p. 4.)

The linear attenuation coefficients of the pixels can be expressed as CT numbers and defined in Hounsfield (H) units as

where µ_t is the linear attenuation coefficient of the tissue element in that pixel, and µ_w is the linear attenuation coefficient of water at the effective energy of the x-ray beam. With this definition, water has a CT number of 0 H, air of −1000 H, and dense bone of more than 1000 H. In general, CT values for soft tissues range from −100 to +100 H.

For therapy-planning dosimetry, the value of 1000 is not always subtracted; then the CT value scale runs from 0 (air) to more than 2000 (dense bone), and negative values are avoided. CT images can show attenuation differences of less than 0.5% with high-contrast resolution of 10 line pairs per centimeter or more. Each image typically consists of a 512 × 512 matrix of numbers, each of which can be displayed at 256 Gy levels.

Uses in Radiation Oncology and Limitations

Computed tomographic images initially gained widespread use in various areas of radiation oncology including (1) the delineation of the target volume, (2) the determination of the relative geometry of critical structures, (3) the optimal placement of beams and the shaping of apertures, (4) the calculation of dose distribution, (5) treatment verification, and (6) follow-up evaluation of treatment outcome. Currently, the process of CT simulation has received increased clinical adoption; see the section on CT Simulation.

Computed tomographic images are usually acquired with a 120 to 140 kV x-ray beam (an effective energy of 70 to 80 keV), which undergoes a significant number of photoelectric interactions, particularly in bone. Therefore, a nonlinear conversion table must be used to obtain electron densities for dose calculations relevant to the much higher energy treatment photons that undergo mainly Compton interactions. Several authors have developed an empiric relationship between CT numbers in H units and electron densities normalized to 1.0 for water.⁵ This relationship is expressed as two or three linear regions and stored in a lookup table (LUT). Phantoms that contain multiple objects of a known electron density for converting CT numbers are commercially available.

CT provides cross-sectional images, which are ideal for radiation oncology treatment planning; however, there are several limitations. First, although CT images show exquisite cross-sectional anatomy, in some cases they do not allow one to differentiate diseased from normal tissue. Typical commercial scanners have a gantry opening of less than 70 cm in diameter, which can be restrictive for setting up patients with immobilization devices. To simulate the patient treatment position, a flat insert is needed for the standard imaging couch that has a concave surface in the transverse plane.

Artifacts arising from various sources can have a significant effect on the accuracy of CT numbers. The boundary of an object occupying only a portion of a voxel may not be accurately represented in the CT image, an effect known as partial volume averaging. Motion from respiration can affect the apparent shape and extent of organs. Streak artifacts can occur as a result of highly attenuating objects such as dense bone or metallic implants; there may also be darkened regions between or behind highly attenuating objects. In addition, when contrast agents with high attenuation coefficients are used to provide improved visualization of certain tumors or structures, the CT numbers (H units) in the affected portions of the image are artificially high and will introduce errors in dose calculations. Editing of the image (e.g., by substituting soft-tissue values in the affected region) is usually performed, but distortions in the organ shapes are difficult to correct.

The modern CT gantry is capable of tilting up to 30 degrees from the vertical, although the majority of CT images for radiotherapy application are in the transverse plane. Therefore, sagittal or coronal images, or DRRs (see sections on CT Simulation and Image Reconstruction), if desired, must be reconstructed from sequential and preferably contiguous axial images. The resolution of such reconstructed images depends on the slice thickness and spacing. The availability of helical scanners and, more recently, multislice scanners has made practical the acquisition of larger data sets in a few minutes (i.e., several hundred axial images spaced 1 mm to 2 mm apart), thus improving the quality of images reconstructed along nonaxial planes.

Recent Developments

Faster and more accurate CT examinations have become possible with the introduction first of the single-slice and then of the multi-slice spiral, or (more properly) helical, scanners. Volumetric scan data is acquired by having the table move continuously relative to the gantry, concomitant with the continuous rotation of the x-ray tube and detector array around the patient. Multislice units may acquire the data for as few as 4 or as many as 320 slices during a single 0.35 second rotation.⁶ State-of-the-art multislice CT scanners can obtain isotropic data sets with voxels of dimension as small as 0.5 mm. These scanners can acquire the data for a full set of images in times well below 1 minute, usually during one breath hold. For radiotherapy application, however, one must consider the possibility that patient anatomy during a breath hold may not be in its “average” location during radiation treatment.

To address the limitations of CT gantry aperture on the ability to set up patients with immobilization devices, commercial scanners with 80 to 85 cm apertures have become available. In connection with respiration-gated radiation therapy, in which dose delivery from the treatment accelerator occurs at a preselected interval in the respiratory cycle while the patient breathes normally, different means of synchronizing CT acquisition with respiration have been introduced. In one method, a respiration monitor system triggers the scanner to acquire a single axial image while the patient breathes normally, followed by indexing of the table to the next position; the process is repeated until the volumetric set of images is obtained.⁷ A more widely used approach is the acquisition of volumetric data concurrently throughout the respiration cycle, for the purposes of obtaining information on anatomical motion and more accurately accounting for it in the treatment plan.^8,9 In this approach, axial images are acquired continuously while the respiration signal is simultaneously recorded; the images are then retrospectively sorted according to the respiration phase to form volumetric sets at different phases.

The use of CT for treatment verification is discussed in the section on In-Room CT-Based Systems.

Use and Limitations of Computed Tomographic Scout View

Digital projection images, the so-called scout view, can be obtained from CT scanners by translating the patient relative to the x-ray tube and detector array (Fig. 8-3). Although the x-ray tube may be oriented at any angle, the most common arrangement is in either the anterior-posterior (AP) or the lateral direction. The data obtained are a two-dimensional (2-D) matrix of relative transmission values in an arbitrary numeric range that may differ from one type of scanner to another. Adjustment of the window and level controls provides an image of appropriate brightness and contrast. The scout view is often useful in determining the extent and spacing of axial scans. In addition, in the hard-copy output, the positions of the axial images can be annotated on the scout views, relative to a reference position selected during the patient setup.

FIGURE 8-3 • Schematic drawing illustrating the acquisition of a localizer radiograph, or scout view.

(From Krestel E [ed]: Imaging systems for medical diagnostics: fundamentals and technical solutions, Berlin, 1990, Siemens, p. 434.)

The CT scout view differs from conventional projection radiographs in that the x-ray beam diverges only in the direction across the patient table, but not in the direction along the patient table. In addition, the shape of the CT detector array is concave. Thus, CT scout views cannot be directly compared with conventional projection images used in radiation oncology such as the simulator images. Furthermore, although window and level adjustment allows the scout view images to have excellent contrast, the spatial resolution of the CT scout view is significantly less than that of simulation images, by almost an order of magnitude.

Principles of Operation

The phenomenon of nuclear magnetic resonance (NMR) was first described by Purcell and Bloch, the 1952 winners of the Nobel Prize in physics. Early work on the use of NMR for imaging was done by Lauterbur, Damadian, and others. Nuclei with a magnetic moment, usually those with odd numbers of protons or neutrons, when placed in a strong magnetic field, can be excited from their ground or lowest energy state to a higher state by a pulse from a second and weaker magnetic field that is varying in radiofrequency. When these nuclei return to the ground energy state, they emit radiofrequency energy that can be detected by very sensitive wire coils and appropriate electronics. Many commercial MRI units use a 0.5 to 1.5 tesla (T) magnetic field produced by a superconducting magnet and radiofrequency signals of 20 to 65 MHz for studies of the distribution of hydrogen nuclei (protons) in the body. Additional electromagnets are used to localize the region where resonance takes place (x-, y-, and z-gradient coils) and remove inhomogeneities in the main magnetic field (shim coils) (Fig. 8-4).

FIGURE 8-4 • Typical magnetic resonance imaging system. RF, Radiofrequency.

(From Sprawls P: Physical principles of medical imaging, ed 2, Madison, WI, 1993, Medical Physics Publishing.)

Typical magnetic resonance (MR) images contain a matrix of 256 × 256 values, a factor of 2 less than CT images in each dimension. Each pixel contains a complex combination of information about hydrogen nuclei density (ρ), spin-lattice relaxation time (T₁), and spin-spin relaxation time (T₂). The relaxation times are a measure of the time required for the disappearance of the signal caused by nuclei returning from an excited to an unperturbed energy state, or for dephasing of the initially aligned and precessing nuclear spins. Radiofrequency pulse sequences can be designed to enhance the dependence of the image on the values of any of the three parameters and improve the contrast for various tumors or other abnormal tissue variations. The appearance of MR images may be further affected by blood flow or by injection of contrast agents such as gadolinium compounds. MR image data may be obtained for a three-dimensional (3-D) volume or for any arbitrary plane within the volume. MRI offers the possibility of excellent discrimination of certain tumors with high contrast, the ability to select arbitrary planes for imaging, and good resolution (Fig. 8-5).

FIGURE 8-5 • Physical characteristics of tissue that are displayed in the magnetic resonance image. MR, Magnetic resonance; RF, radiofrequency; T₁, spin–lattice relaxation time; T₂, spin–spin relaxation time.

(From Sprawls P: Physical principles of medical imaging, ed 2, Madison, WI, 1993, Medical Physics Publishing.)

Uses and Limitations

MR images often provide better soft-tissue contrast than CT, thus improving discrimination between tumor and normal tissue in some disease sites. Small differences in T₁ and T₂ can be exploited by manipulating imaging parameters to enhance the contrast among different tissues. In addition, by using the three orthogonal field gradients independently or in combination, the image plane can be oriented in a direction that best displays the extent of a particular tissue. Functional MRI of the brain can identify speech, visual, and sensory areas to avoid when treating intracranial lesions.¹⁰

Several characteristics have limited the applicability of MR images alone for treatment planning. One is the lack of signal from cortical bone: Although bone location can be inferred in some disease sites, it is not possible to distinguish bone-air interfaces such as sinuses. Pixel intensities do not correlate with electron density as they do in CT; thus they cannot be used directly in dose calculations. Intensity variations occur across the images, such as those arising from falloff in sensitivity toward the ends of the receiver coil. The images may be distorted by variations in local magnetic fields caused by imperfections in the machine itself and by the presence of metal objects in the environment and within the patient. Because of these limitations, MR images are usually registered to CT, and the information MRI provides is transferred to the CT study for treatment planning purposes. The registration of CT and MR images is described in another section. MRI has been effectively used for treatment planning of brain, head and neck, liver, pelvis, and prostate tumors (Fig. 8-6). Other disadvantages include the high cost of equipment and site preparation, examination times that are 1.5 to 2 times those required for CT, a limited diameter of the patient tunnel opening in the magnet, and magnetic and radio-frequency shielding problems.

FIGURE 8-6 • Left, Axial magnetic resonance image of the brain. Malignant tissue appears as brighter areas in the upper left. Gross target volume and planning target volume are outlined in light blue and yellow, respectively. Right, Axial computed tomographic image of the same patient. The computed tomographic and magnetic resonance image sets have been registered by alignment of the patient surface and the target volume contours transferred to the computed tomographic image.

(From AcqSim, Philips Medical Systems, Bothell, WA.)

MRI alone can be used in treatment planning of sites that do not have large tissue inhomogeneities by substituting a bulk density correction for dose calculation in the absence of an electron density map.¹¹

MRSI of ¹H nuclei in combination with MRI is receiving increased clinical adoption for evaluating tumor extent and necrosis in brain tumors and potential staging and evaluation of prostate cancer, and thus as an aid to external beam and brachytherapy planning.^12,13 The relative concentrations of metabolites choline, creatine, and citrate, specifically elevated values of the ratio (choline + creatine)/citrate, have been hypothesized to be an indicator of malignant prostatic tissue (Fig. 8-7).^14,15

FIGURE 8-7 • A single plane of ¹H magnetic resonance spectroscopy imaging (MRSI) data from a prostate cancer patient (Gleason score 7, PSA 8 ng/ml). Left, MRSI grid shown on the T2-weighted magnetic resonance image. Right, Spectra corresponding to the voxels shown on the grid. H, Healthy peripheral zone; SC, suspicious for cancer.

(From Zaider M, Zelefsky MJ, Lee EK, et al: Treatment planning for prostate implants using magnetic-resonance spectroscopy imaging, Int J Radiat Oncol Biol Phys 47:1085, 2000.)

MRI studies of perfusion, diffusion, and contrast enhancement in recent years have shown potential for assessment of therapeutic response and as an early predictor of treatment outcome.¹⁶ These investigations indicate a potential use of MRI as an aid to optimizing therapy for individual patients.

An advantage of MRI is that rapid-sequence images can be acquired without the radiation dose associated with CT. Applications of cine MRI to radiotherapy have evaluated the influence of rectal and bladder filling on prostate motion and respiration-induced motion of anatomic structures in the thorax.^17,18

Recent Developments

Several recent developments may increase the use of MRI, MRS, and MRSI in radiation oncology. Self-shielded magnets and very specialized, low-cost permanent magnet systems will make MRI more available in the radiation oncology clinic.

The introduction of new fast-pulse sequences and greater computer power for data manipulation make it possible to track the uptake of gadolinium diethyl enetriamine pentaacetic acid, Gd-DTPA, dynamically to measure tissue perfusion quantitatively with high resolution. Dynamic, contrast-enhanced imaging can also be used to measure cerebral blood volume, blood-brain barrier permeability, necrotic fraction, extracellular space, and permeability surface area product. This technique is being used to study osteosarcoma and to attempt to predict necrotic fraction after treatment.

The in vivo mobility of water molecules depends on the properties of the diffusing medium and the presence of physical barriers. Diffusion-weighted imaging is being used to provide information on bone marrow cellularity and changes in hematopoiesis resulting from chemotherapy in leukemia patients.

Rapid imaging techniques such as echo-planar imaging that allow images to be obtained in a fraction of a second are allowing functional MRI (fMRI) studies of sensory and motor stimulation, vision, and language. These techniques are also being used to study angiogenesis in tumors and healing of wounds.

The availability of high field (≥3.0 T) clinical MRI systems will provide images with increased signal-to-noise ratio (SNR), increased separation of spectrum peaks, and enhanced susceptibility effects. The increased SNR can reduce acquisition time or enhance resolution for fMRI studies. The improved image spectroscopic resolution might also provide the opportunity to obtain new biochemical information.

There is also a new class of MRI contrast agents, which are capable of gene transfection into cells. These agents will allow gene therapy to be monitored by MRI through cotransported MRI contrast agents. Additional refinements include MRI contrast agents that will enable the MRI signal to be greatly amplified through association of the site-directed contrast agent with its target.¹⁹

Principles of Operation

Medical ultrasonography is a direct descendant of sonar techniques developed during World War II to search for submarines. Short bursts of sound are sent out into the ultrasound transmission medium at regular intervals. Between bursts, sound echoes return from reflecting objects or interfaces. A knowledge of the speed of sound in the medium and the time of the roundtrip travel to the reflector enable one to calculate the distance from the sound source and detector to the reflecting object.

Medical ultrasound imaging requires the production and detection of mechanical pressure waves within the tissues of the body. It normally uses sound frequencies in the 3 to 10 MHz range. Ultrasound beam production and detection are accomplished with piezoelectric materials. When placed in a varying electric field, such substances vibrate and produce ultrasound. And when they are deformed by mechanical pressure, for example, because of the reflected ultrasound, they produce electric fields that are detected as the signals in ultrasound scanners. Manufactured ceramic piezoelectric crystals are incorporated in the transducer, the component of the ultrasonographic system that is placed against the patient. The transducer produces the initial ultrasound energy and detects reflected energy. Some of the ultrasound energy sent into the body is reflected from tissue interfaces and returns to the transducer. The total time between the production and the detection of the sound waves, which move at an approximately constant velocity of 1540 m/s through soft tissue, is a measure of the roundtrip distance from the transducer to the tissue interface. Ultrasonographic image resolution depends on the ultrasound frequency, transducer size, and focusing characteristics of the system. In general, resolution increases with increasing frequency and decreasing transducer size. Medical ultrasound-imaging systems have resolution capabilities of 1 to 10 mm. At the higher frequencies, imaging becomes more difficult at depth, because of increasing absorption of the ultrasound energy in the tissue. The ultrasound energy is not ionizing.

The images produced show interfaces and variations in acoustic impedance. The acoustic impedance of a given tissue is equal to the product of the tissue density (mass/volume) and the velocity of sound in that tissue. The images can show tomographic views in almost any orientation.

Uses and Limitations in Radiation Oncology

Early medical ultrasound B-scan units provided some of the first cross-sectional images for treatment planning; however, current CT and MRI units, although 5 to 10 times more expensive, produce images with better resolution and contrast that are more desirable for treatment planning. The major exception is the use of high-resolution images of the prostate obtained with an ultrasound imager that employs a special rectal probe. Transrectal ultrasound has become the predominant image modality for planning and delivery of prostate implants using transperineal template-guided insertion. Intraoperative ultrasound has also been applied to evaluate high-dose-rate prostate brachytherapy implants.²⁰ Application of ultrasound in breast cancer occurs most commonly in the initial diagnostic evaluation: in addition to detecting abnormalities, it is useful in assessing the presence of multicentric tumor.²¹ Ultrasound can serve to define involved primary and nodal sites for breast-cancer radiotherapy and postlumpectomy radiotherapy. A commercially available device allows ultrasound-guided patient positioning for external-beam radiation therapy; several centers have reported on its application to daily localization and correction of prostate position.²²

Ultrasound beams cannot penetrate gas-filled cavities or bone. This limits the use of ultrasonography. Furthermore, ultrasonographic images are much more difficult to interpret than those from CT or MRI (Fig. 8-8).

FIGURE 8-8 • Example of a sagittal CT image (left) and ultrasound image (right) of the same patient in the treatment position. The images were taken within minutes of each other. Orange, yellow, and green curves are outlines of the bladder, prostate, and rectum from the planning CT.

(From Kuban DA, Dong L, Cheung R, et al: Ultrasound-based localization, Semin Radiat Oncol 15:180, 2005.)

Recent Developments

Imaging systems consisting of multielement transducers can produce ultrasonographic images in a slice at rates fast enough to show smooth motion on a video screen (≈30 pictures per second). All modern ultrasonographic units use a digital scan converter to produce digital images. Individual pictures, or frames, can be transferred to PACs or to film with a laser imager. Recent ultrasound-guided patient positioning systems produce 3-D ultrasound images, thereby improving image interpretation and alignment relative to earlier 2-D systems.²³

Recent Developments

A modern commercial simulator has replaced the image intensifier with an amorphous silicon flat-panel detector (see the section on Electronic Portal Imaging Devices), providing distortion-free radiographic and fluoroscopic digital images. The detector is mounted on a retractable arm that allows imaging of a larger range of patient setup positions compared with an image intensifier unit. The system also provides a 3-D image set through cone-beam image reconstruction; see the section on In-Room CT-Based Systems.

Portal Radiographs

Portal radiographs are a fundamental part of the clinical quality-assurance program. They provide a beam’s eye view record of the patient treatment volume, radiographically revealing the relationship between anatomical structures and the treatment field. There are three types of imaging systems available for acquiring portal radiographs: film, electronic portal imaging devices (EPIDs), and CR; each system is described in later sections. The American Association of Physicists in Medicine (AAPM) Task Group 28 defines a portal radiograph as “a radiograph produced by exposing the image receptor to the radiation beam which emanates from the portal of a therapy unit.”²⁴ The report describes three types of radiographs (within the context of acquisition with film):

Localization: A portal radiograph produced by an exposure that is short compared with the daily treatment of that field (generally called localization films, beam films, or port films).

Verification: A film exposed for an entire treatment field.

Double exposure: A localization film composed of two sequential exposures; the first is that of the shaped treatment field, and the second of a larger rectangular field, superimposed on the first. The double-exposure portal makes it easier to relate the treatment field to the surrounding anatomy and is particularly useful for relatively small field sizes.

The importance of portal radiography as a quality-assurance tool is supported by numerous published studies, which have made it clear that portal radiographs are essential to accurate radiotherapy and that frequent imaging is desirable for difficult patient setups or highly conformal treatments.²⁵

Portal Films

Portal films have been increasingly replaced by EPIDs (described in the following text). The films traditionally used for portal imaging are the therapy localization film and the verification film, both manufactured by Kodak. The former has a high silver content and is properly exposed by only a few centigray. The latter is a relatively slow industrial film, making it more useful for a full-treatment exposure. Several different films by other manufacturers have been successfully used for portal imaging with optimal exposure techniques determined through trial and error.

The image quality of either type of film is reduced in part by high-energy secondary electrons that exit from the patient onto the film. Portal film quality is improved with specialized radiotherapy cassettes that provide metal front screens in uniform and close contact with the film and that is thick enough to absorb the shower of secondary electrons exiting the patient. The thickness of these screens is determined by measurements of scatter-to-primary ratios for large fields.²⁴ Thicknesses are generally 1 to 4 mm, depending on the beam energy and on the screen material (usually lead or copper). A rear metal screen is also often used to improve the overall sensitivity, or speed, of the system. The use of rear screens may cause some reduction in image resolution; however, in practice no significant degradation of image quality is observed. A second important function of the film cassette is to provide good film–screen contact.

To improve the readability and accessibility of portal films, various film digitizing and image processing systems are available. The most commonly used film digitizers are either charge coupled device (CCD)-based or combination laser–photomultiplier tube. The resultant images can be stored, accessed over a computer network, and enhanced and processed similarly to electronic portal or CR images; see section on Use of Portal Radiographs.

Computed Radiography

Photostimulable phosphors (PSPs), also called storage phosphors, are compounds that are capable of absorbing and storing energy from x-rays. They can be stimulated subsequently to release energy in the form of visible light. The term computed radiography (CR) refers to the process of image acquisition and readout of PSP plates, then conversion to a digital image. Absorbed x-ray energy excites electrons in the PSP plate to semistable higher energy levels, producing a latent image. The plate is read out with a helium-neon (red) laser, which stimulates the semistable electrons to drop down to their ground energy state while emitting visible (blue) light. The light is sensed by a photomultiplier tube and converted to an electrical signal, which is digitized. In a commercial system designed specifically for simulator and portal radiographs, the PSP plate is exposed in a standard film cassette and processed by a CR reader. The exposure requirements are similar to those for conventional simulator or portal localization films; however, the exposure flexibility reduces the likelihood of under- or over-exposure, thus reducing the number of retakes.

Electronic Portal Imaging Devices

EPIDs have largely replaced film, owing to advances in detector technology and hence image quality, and to the availability of commercial PACs specifically designed for radiation oncology. Several features make EPIDs attractive for portal radiography. One is the ability to acquire and display an image in seconds. When combined with a computer-controlled accelerator fitted with an MLC, field setup and image acquisition can be performed remotely, obviating the need to re-enter the treatment room each time. This allows for more frequent treatment verification, or to acquire a series of images during a single treatment for examining patient movement. Another powerful feature is that the images are in digital form, which allows for application of software tools to extract information relevant to treatment verification.

The EPID is attached to the gantry of an isocentric radiation treatment machine so that it may intercept the exiting photon beam at all gantry angles. The ideal support system must be sturdy and capable of precise repositioning, yet easily retracted or removed to facilitate patient access. First-generation commercial EPIDs have been either video-based or matrix ionization chamber-based systems.²⁷ In video-based systems, the image is formed by the interaction of the radiation with a fluorescent screen assembly consisting of a metal (e.g., copper or tungsten) plate onto which the phosphor is deposited. The metal plate is upstream relative to the phosphor and absorbs scattered electrons emanating from the patient, as well as interacting with the photons to generate high-energy electrons that, in turn, produce the fluorescent image in the phosphor. A front-silvered mirror, placed diagonally, reflects the fluorescent light by 90 degrees into a video camera. In the matrix ionization chamber system, electrode strips on two printed circuit boards form an array of ionization chambers; the strips run horizontally on one board and vertically on the other. A 1 mm gap between the boards is filled with iso-octane liquid, which serves as an ionization medium for the x-rays. An image is obtained by switching on the high voltage to one row, collecting the currents from each of the column electrodes, and repeating the process for each of the rows. The first-generation systems, although more convenient than film, have a number of limitations. The camera-based systems are bulky and image quality is reduced because of the poor optical transfer system, and the matrix ionization chamber system requires long irradiation times and is especially sensitive to small changes in dose rate, resulting in line and band artifacts.

The current commercially available EPIDs take advantage of the large industrial development of flat-display panel technology, and provide faster acquisition and superior image quality than their predecessors. Variously referred to as active-matrix flat-panel imagers or amorphous silicon flat-panel image detectors, the array of electronic circuitry comprising the pixels is fabricated on panels of hydrogenated amorphous silicon. Each pixel comprises a light-sensitive photodiode attached to a thin-film transistor acting as a switch. The physical buildup of the detector array consists of a metal plate (typically 1 mm copper) and phosphor screen to convert x-rays to visible light; the metal layer also serves to attenuate scattered radiation. The light is converted to electron-ion pairs in the photodiode, and collected by application of a bias voltage onto a storage capacitor. The charges stored in the pixels are transferred to the readout electronics by activating the pixel switches row by row and reading out all columns simultaneously. The arrays are capable of 10 frames per second or higher, and thus are applicable to fluoroscopy as well as radiography. For portal imaging applications, several frames are averaged together to reduce image noise. Panel sizes of 30 × 40 cm² or 40 × 40 cm² are available in arrays of 512 × 384, 1024 × 768, or 1024 × 1024 pixels, respectively (Fig. 8-10).

FIGURE 8-10 • Example of digitally reconstructed radiograph (left) and electronic portal image (right) of a patient treated for nasopharynx cancer. Portal image was acquired on an amorphous silicon flat-panel image detector; green contour indicates irradiated area of an intensity-modulated treatment field delivered with a multileaf collimator. Note the projected grid scale (graticule) indicating the field central axis and calibrated in 2 cm intervals in the isocenter plane.

(From Varian Medical Systems.)

Use of Portal Radiographs in Radiation Oncology

Image Registration

Portal radiography is a primary tool for quality assurance in radiation delivery. The portal image from the first treatment is compared with the corresponding reference image (a digitized simulator film or digital image, or digitally reconstructed radiograph from CT) to ensure correct patient setup and proper block construction and placement or MLC programming. Since portal (and simulator) radiographs of oblique fields are confusing and difficult to interpret because of their nonstandard view of radiographic anatomy, AP and lateral portal films are generally used to verify the isocenter location. Side-by-side comparison of the simulator and portal images can be facilitated with the use of a graticule, a device that projects a cross-hair shadow and centimeter scale in the portal film; the device is supplied by several vendors.^24,28 The use of such a device is recommended because the cross hairs of the linear accelerator or cobalt-60 fields do not appear on the portal image. If field adjustments are needed, this tool facilitates the quantification and communication of the desired changes (see Fig. 8-10).

Besides the side-by-side manual procedure, various algorithms have been developed to enable computer-assisted comparison of images for detection of field placement errors. This is usually accomplished in several steps: (1) information about the positions of the patient anatomy and radiation field are extracted from both portal and reference images; (2) a common reference frame for both images is established by registration of the radiation fields; and (3) setup error is determined by registering the patient anatomy in both images, then measuring the resulting displacement between the portal and reference radiation field edges.

Two of the most common methods of interactive, or manual, image registration are the line drawing or template technique, and point-pair registration. Point-pair registration involves identifying the positions of corresponding anatomical landmarks in the reference and portal images; the portal image is then transformed (i.e., translated, rotated, and scaled) such that the selected points align with those in the reference image. Although the methodology is simple, point-pair registration is error prone, because of the difficulty in correctly identifying corresponding fiducial points, particularly when reference and portal images are of different image quality (i.e., kV vis-à-vis megavoltage beam quality). The template method involves drawing (graphically, by means of a mouse or trackball) lines or curves indicating the field edge and patient anatomy on the reference image, which is then overlaid and aligned with the portal image (Fig. 8-11). Semiautomatic extensions of the template method are those that automatically extract anatomic features in the portal image and align them with the reference template. A third category of registration methods uses automated pixel-by-pixel, grayscale intensity–based image correlation. This approach assumes similar quality images for matching; thus, it requires registration of a first-day portal image to the reference image by some other means, then uses the first-day portal image as reference for subsequent treatment sessions.

FIGURE 8-11 • Example of template-based portal image registration as applied to treatment of a paraspinal tumor. Left, Digitally reconstructed radiograph in which a template is drawn around a surgical metallic (titanium) implant; localization field edge is represented by the outer squared lines. Right, Portal image (from an amorphous silicon flat-panel detector) in which the template has been aligned to the observed implant; differences between the position of the software-detected field edge in the portal image and reference field edge from the digitally reconstructed radiographs indicate the patient setup error.

(From Portal Vision, Varian Medical Systems, Palo Alto, CA.)

As previously indicated, portal radiographs are part of the overall quality-assurance program for ongoing verification of treatment accuracy. Portal images can reveal errors in the patient setup position, field size and orientation, or placement and shaping of shielding blocks. Thus they should be taken during the initial treatment setup and weekly thereafter, as recommended by the AAPM Task Group on Radiotherapy Portal Imaging Quality and the AAPM Task Group on Comprehensive Clinical Quality Assurance.²⁸ Regular review takes place at the weekly chart (film) rounds; the current week’s portal films are compared side by side with the appropriate reference image. Alternatively, electronic images can be displayed by means of a computer monitor or liquid crystal display projector. Approved portal images should be signed (on the film electronically via a radiotherapy PAC) and dated for medicolegal documentation purposes, and discrepancies should be investigated and rectified. Note that altering the patient setup position or skin marks based on a small discrepancy (e.g., <0.5 cm) observed in a single portal image may not be appropriate. Rather, it may be more appropriate to take a repeat image on the subsequent day before effecting a change. However, by using online imaging with implanted markers, treatment accuracy can be improved such that changes of less than 0.5 cm can be made accurately.

Specific goals for using EPIDs should be established before they can be put to clinical use.^25,27 A number of key questions should be addressed; these questions include:

What is the primary purpose of introducing EPIDs into the clinic—as simple film replacement, improved patient positioning, or assessment of localization errors and adequacy or margins?

For which patients and protocols will EPIDs be used—all patients, selected sites, or selected patients?

What will be the imaging frequency and type of acquisition—daily, weekly, single or double exposure, all or some fields?

What type of image evaluation—immediately prior to the treatment fraction or between fractions, visual or computer-based analysis?

What is the patient positioning error threshold for taking corrective action?

Where will image evaluation take place—at the treatment unit, in a central work area, or department-wide on any workstation?

Answers to these questions define the scope of EPID use so that the clinic chooses the appropriate tools and resources to develop clinical procedures.

Other Applications of Electronic Portal Imaging Devices

The use of EPIDs for detecting radio-opaque fiducial markers implanted in or near a tumor, particularly in the prostate, has received increasing attention as a means of correcting errors in target position from internal organ motion.²⁶

Another application is to verify intensity-modulated treatment fields delivered with an MLC. With present algorithms, this is usually accomplished by irradiation of a flat phantom before actual treatment.²⁹ The images of the subfields composing the intensity-modulated field are summed in the computer to obtain a 2-D intensity pattern (Fig. 8-12), which is compared with the intended pattern from the treatment planning system. Other uses of EPIDs include transit dosimetry to measure the accuracy of dose delivery during treatment, and in accelerator quality-assurance procedures.^30,31

FIGURE 8-12 • Electronic portal images of an intensity-modulated treatment field delivered with a multileaf collimator. Eight images are shown out of a total of 25. Each image is converted into an array of dose values, and summed to obtain an integrated dose map.

(From Chang J, Mageras GS, Chui CS, et al: Relative profile and dose verification of intensity-modulated radiation therapy, Int J Radiat Oncol Biol Phys 47:231, 2000.)

Limitations

Portal radiograph quality is limited primarily by the predominance of Compton scattering at megavoltage energies. Because there is no strong dependence on atomic number (Z), there is very little differential absorption, as is the case at diagnostic x-ray energies. Blurring of structures is caused by a relatively large radiation source (focal spot) and by patient movement during radiation exposure.

There appears to be a systematic degradation of portal image quality with increasing accelerator beam energy. This is attributable to a reduction in subject contrast and in resolution, although the relative importance of these factors is uncertain. Although the decrease in contrast with photon energy, changing from the kiloelectron volt (keV) to megaelectron volt (MeV) range, is a result of the reduced probability of photoelectric interaction, the reduction in image quality with sources higher than 1 MeV cannot be similarly explained and may be caused in part by the increased range of Compton electrons generated in the front screen for film, or in the conversion plate and phosphor screen for CR and electronic portal imaging systems.³² At the high end of megavoltage photon energies, the situation is further complicated by increased multiple photon scattering and some increase in pair production. Additional degradation of cobalt-60 beam images is caused by blurring caused by the relatively large cobalt-60 source size.

The PSP plates used in CR systems are sensitive to ambient light, which can erase the latent stored image. Care must be taken to use dimmed ambient light near the CR reader and to minimize the time between removing the plate from its cassette and scanning it in the reader. EPIDs have a more limited field of view relative to film or PSP plates; in addition, care must be taken not to irradiate the readout electronics outside the detector-sensitive area, as radiation damage can shorten detector lifetime.

Portal Radiograph Quality Assurance

A quality-assurance program for portal films should include regular inspection of radiotherapy film cassettes for repair or replacement, periodic image quality testing and evaluation using a standard phantom, and monitoring and preventive maintenance of the automated film processor. Quality assurance of a CR system should include inspection of both PSP plates and cassettes, periodic image quality testing, monitoring and preventive maintenance of the CR reader, monitoring of available disk space on the viewing workstation, and database maintenance. The images should also be checked for spatial distortions by periodically scanning a geometric test pattern with the reader.

Most EPIDs require a calibration process to correct for inherent systematic artifacts (e.g., variations in sensitivity across the detecting device). For amorphous silicon EPIDs, the electronic noise and sensitivity of the photodiode pixels can vary with time. The AAPM Task Group 58 has reported on acceptance testing, commissioning, and periodic quality assurance of EPIDs.²⁷ The recommended quality-assurance tasks include tests of the collision interlock system that prevents collision of the EPID with the couch or patient, accuracy of the EPID positioning relative to the isocenter, image artifacts, image resolution and noise using phantoms designed for this purpose, and geometric localization accuracy.

In-Room Kilovoltage Radiography

The advent of large-area flat-panel x-ray detectors has led to widespread use of kV image guidance systems for patient positioning in the treatment room. At present, several kV imaging developments have evolved into commercial products that use one or more image detectors together in combination with kV x-ray sources in the treatment room.^33–37 The near-diagnostic quality of kV radiographs facilitates automatic image registration and beam alignment just prior to treatment or in real time during treatment, often in combination with implanted fiducial markers (Fig. 8-13). A single radiograph provides 2-D information, whereas two imaging systems enable stereoscopic 3-D imaging. Gantry-mounted systems (Fig. 8-14) provide radiographic as well as cone-beam CT capabilities (described in the next section); such systems are generally mounted at right angles to the treatment beam.

FIGURE 8-13 • Megavoltage portal image (left) and kilovoltage x-ray radiograph of a patient with implanted markers in the prostate. Both types of image detectors use amorphous silicon flat-panel technology.

(From Varian Medical Systems.)

FIGURE 8-14 • Linear accelerator fitted with two amorphous silicon flat-panel detectors, one for imaging with the megavoltage beam, the other mounted 90 degrees to the treatment head and opposite to a kilovoltage x-ray source.

(From Elekta Oncology Systems: Wavelength 6:3, 2002.)

In-Room CT-Based Systems

The principal advantage of CT-based systems in the treatment room over radiographs is that they provide volumetric anatomical information, including soft tissue localization. The use of repeat CT scans during treatment as a means of measuring and reducing target positioning errors has become increasingly widespread; Chapter 12 gives a more extensive description. Different types of systems for imaging the patient at the treatment accelerator are commercially available. “CT-on-rails” systems combine a linear accelerator and conventional CT scanner in the treatment room.³⁸ Helical tomotherapy systems use a CT detector consisting of pressurized xenon gas within a tungsten septa-filled housing with the megavoltage beam (Fig. 8-15). Cone-beam CT systems make use of a 2-D flat-panel imaging sensor to acquire a volumetric image in a single rotation; such systems have been developed for use with megavoltage as well as kV x-ray sources (Fig. 8-16).^39,40 The cone-beam geometry is actually pyramidal in shape, diverging both in longitudinal and lateral directions with its apex at the x-ray source. The source-detector pair moves around the patient to collect a set of 2-D projection images, which are then processed with a cone-beam reconstruction algorithm to obtain a 3-D image set.

FIGURE 8-15 • A, Axial image from a helical megavoltage CT (MV-CT) scan of a prostate patient. B, Axial MV-CT image of a head-and-neck patient.

(From Meeks S, Harmon J, Langen K, et al: Performance characterization of megavoltage computed tomography imaging on a helical tomotherapy unit, Med Phys 32:2673, 2005.)

FIGURE 8-16 • Axial images from a planning CT scan (top) and a kilovoltage cone-beam CT (CBCT, Varian Medical Systems) scan used to correct patient position prior to a single-fraction treatment of a lymph node. Lymph node and bowel are delineated on the plan CT. CBCT image quality is such that node and bowel are readily visible.

Electromagnetic Marker Tracking

An emergent localization technology uses implanted electromagnetic transponders whose position is detectable with nonionizing radiation, providing continuous real-time monitoring. The position of the transponders is detected at a rate of 10 Hz with an electromagnetic source-detector array placed close to the patient.⁴¹ Application of this system for prostate localization and real-time position monitoring has been demonstrated.⁴²

Optical Methods

Optical imaging systems provide positional information on the external anatomy, and commonly use infrared light to determine an object’s location. The object can be active, such as an infrared emitter; passive markers consisting of spheres or disks that reflect the infrared light; or the patient’s surface acting as a reflector. The optical detectors are usually CCD cameras, which are a collection of light-sensitive cells or pixels arranged in a 2-D array, producing a digital image. An optical system consisting of two such CCD cameras provides stereoscopic 3-D information of the detected objects (Fig. 8-17).

FIGURE 8-17 • Schematic of stereoscopic method, in which two 2-D cameras determine the 3-D location of a marker as the intersection of two lines from the marker to the detection point in each camera.

(From Meeks S, Tome W, Willoughby T, et al: Optically guided patient positioning techniques, Semin Radiat Oncol 15:192, 2005.)

The first commercially available optical system for radiotherapy localized patients by means of detecting passive markers attached to a biteplate.⁴³ An early system for extracranial treatment enabled tracking of reflective spheres attached to the patient’s torso.⁴⁴ A respiratory monitoring system, in which infrared light from an illuminator is reflected from a passive reflective block placed on the patient and detected by video camera, is widely used in connection with respiration-correlated CT, breathing-synchronized fluoroscopy, and gated treatment on a linear accelerator.⁷ More recent optical patient positioning systems use stereoscopic imaging of a speckle pattern projected onto the patient,⁴⁵ or imaging of the intersection of a scanned fan-shaped, laser beam with the patient.⁴⁶

Principles of Operation

Most radionuclides emit gamma rays or x-rays and can therefore be imaged by a gamma camera. A gamma camera consists of a detector, usually thallium-activated sodium iodide (NaI : Tl), the purpose of which is to stop the incident photon and convert the energy deposited in the crystal into light quanta and finally an electronic pulse, the amplitude of which is proportional to the energy of the radionuclide emission. To determine the direction of the photon emerging from the body, and thereby produce a sharp, high-resolution image of the radionuclide distribution within the patient, a collimator is required. The resolution of single photon emission images is optimal between 100 and 250 keV. Both lower and higher energies result in lower sensitivity, because of greater self-attenuation in the patient in the first instance, and in the latter instance because of the necessity of thicker collimator septa. More than 90% of conventional nuclear medicine studies use ^99mTc, which emits a single 140 keV gamma ray. Other isotopes in regular use include ⁶⁷Ga, ¹¹¹In, ¹²³I, ¹³¹I, and ²⁰¹Tl. The energy resolution of NaI : Tl is approximately 8%, allowing a multiwindow acquisition of more than one isotope simultaneously (e.g., ^99mTc and ¹³¹I) and the spatial resolution between 7 and 12 mm. Gamma camera imaging is frequently used in whole-body mode, in which the camera heads produce AP projections of the body to produce spot views, or in SPECT mode, in which the heads rotate around the patient to produce a full set of angular projections for tomographic reconstruction. Also, there is a growing market of combination SPECT/CT imaging systems. These systems range from low (2 mA) current devices up to full multislice CT units, and usually provide the flexibility to perform any sequence of SPECT and CT scans, either as single- or multibed acquisitions. However, the long 20 to 40 minute per field of view required for a SPECT scan does not render whole body SPECT/CT currently plausible, as is the case for PET/CT.

Uses and Limitation of Single Photon Emission Gamma Camera Imaging in Radiation Oncology

Nuclear medicine studies are rarely the primary method of choice for the diagnosis of tumors, with the exception of the role of radioiodine in thyroid cancer, for which it also plays an important role in therapy.⁴⁸ Nuclear medicine is a sensitive method for the detection of metastatic spread of disease, the most widely used of which is the bone scan agent technetium-99m methylene diphosphate, for determining the extent of metastatic spread to the bone.⁴⁹ A methodology to quantitatively track the progression of bone metastases has been proposed by Imbriaco and colleagues,⁵⁰ and is called the bone scan index. A broader class of tumor diagnostic (and therapeutic) tracers includes the somatostatin binding peptides, with selective binding to neuroendocrine cancers.⁵¹ Several groups have evaluated the use of ^99mTc-sestamibi for the detection of breast cancer. One example is Lumachi and colleagues⁵² who reported that the sensitivity of sestamibi scintimammography reaches 100% in patients with breast lesions larger than 8 mm, and that the predictive value of mammography, sestamibi scintimammography, and mammography and sestamibi scintimammography together were 63.4%, 95.1%, and 97.6%, respectively. Lymphoscintigraphy is a procedure in which a radiotracer, ^99mTc sulfur colloid, is injected subcutaneously into sites providing lymphatic drainage to the nodes of interest. Spot views are then taken with a gamma camera at two angles for stereoscopic localization of the nodes. The use of lymphoscintigraphy is the precise localization of the internal mammary and axillary lymph nodes, information from which can be used in the design of radiotherapy portals for breast radiotherapy and in guiding sentinel node biopsy. ¹²³I and ¹³¹I-metaiodobenzylguanidine have been widely used for the diagnosis and therapy of neuroblastoma and pheochromocytoma.^53,54 Gallium-67 citrate has been widely used for detection of cancer, in particular lymphoma,⁵⁵ through targeting, internalization, and retention via the over-expressed ferritin receptor. The usefulness and complementarity of ⁶⁷Ga SPECT and CT in the management of patients with lymphoma have been extensively demonstrated.⁵⁶ However, ⁶⁷Ga SPECT has also largely been replaced by FDG PET. Another general cell tumor viability tracer, ²⁰¹Tl, has been used to diagnose tumors as well as measure residual tumor, such as in the brain, postradiation therapy.⁵⁷ Others have used ²⁰¹Tl SPECT imaging in addition to blood-flow images with ^99mTc-HMPAO to assist in the differentiation between viable tumor and necrosis in the target definition for stereotactic radiosurgery.⁵⁸

Radiolabeled monoclonal antibodies are a generic class of tumor targeting agents of greater specificity and selectivity, but have not gained widespread use, in part because of their complexity and expense. Whether antibodies will achieve their potential status as the ideal tumor-specific imaging agents still remains a question that is reviewed by Bischof Delaloye.⁵⁹ The practice of radioimmunotherapy (RIT) for the treatment of non-Hodgkin lymphomas has resulted in considerable success and is being routinely performed within departments of nuclear medicine and radiology, medicine, and radiation oncology around the world.^60,61 There are two approved anti-CD20 antibodies shown to be effective with low grade non-Hodgkin lymphomas. These are ⁹⁰Y-ibritumomab tiuxetan (administered after a predose with ¹¹¹In-ibritumomab tiuxetan) and ¹³¹I-tositumomab, which both induce high response rates, with durable remissions in patients with relapsed or refractory indolent non-Hodgkin lymphomas as well as when used as front-line treatment in patients with indolent non-Hodgkin lymphomas.^62,63 Yet, in spite of more than 30 years of development, RIT has had limited success in the treatment of solid cancers.

Principles Of Operation

PET scans can be limited to the head and neck, thorax, abdomen, and pelvis, but more frequently include the entire torso from the bottom of the ear to the bladder, depending on the type of cancer and the known sites of metastases. In some cases (e.g., melanoma), the entire body from head to feet is scanned. On modern PET/CT scanners, the CT is performed first, in which the CT scout view is used to define a region of the body of interest. Usually the CT is performed over a body length, concordant with an integral number of PET fields of view (typically 12-18 cm depending on the scanner an desired field overlap). This required overlap compensates for longitudinal variation in camera sensitivity, which exhibits a triangular response in 3D because of the diminishing contribution of oblique coincidence events at the end detector rings of the scanner. PET/CT protocols vary widely from center to center. Some centers perform a diagnostic quality CT scan with intravenous (or oral) contrast. Other centers perform low current CT scan (70-80 mA) providing adequate CT image quality for good anatomical structural identification, coregistration, and attenuation correction of the PET emission scan data. The CT component of the exam is rapid (<70 sec) relative to the PET part of the study. On completion of the CT, the CT couch moves into position for the first PET bed position. The patient remains in this position to acquire the desired PET counts, then moves one field of view (minus the designated overlap) to the second position, and so on until PET data for all the bed positions have been acquired. For radiation oncology patients, the curved CT couch top is replaced with a flat couch and the appropriate immobilization equipment deployed. The number of bed positions in a whole-body PET scan can range from a few fields of view up to more than 12 with melanoma patients. The time to acquire each field of view varies from institution to institution, and depends on the amount of activity injected, patient size, whether it is 2-D or 3-D, and detector material. The shortest emission scan times have been 1 minute per bed position, but more typical scan times range from 3 to 6 minutes per bed position. There are certain PET studies, such as in brain or heart imaging, in which only a single field of view is acquired for 20 to 60 minutes. The improved count statistics of this type of image allow lower contrast features to be observed.

PET also has the ability to acquire scans in a dynamic mode of a single field of view, providing temporal resolution of the tracer uptake within a tumor or organ of interest. This type of acquisition is usually employed when investigating new radiotracers (for example to ascertain the optimum imaging time post–tracer injection), or when compartmental analysis is being used to estimate the rate of tracer metabolism or entrapment in a specific intracellular biochemical process. In its simplest form a dynamic acquisition consists of a single field-of-view scan, in which the full duration of the scan is divided into numerous successive time frames (or bins) that allow the kinetics of the tracer within regions of interest (ROIs) to be visualized and plotted to be observed. Dynamic data, gleaned from the images of successive time frames, are frequently used in compartmental models to generate parametric images (i.e., metabolic rate images). One example is dynamic FDG images of the brain, which, using the Sokoloff compartmental model and measurement of the arterial input function, can be converted to a regional glucose metabolic rate.⁷⁷ Dynamic imaging is also useful when establishing the optimal imaging time for a new radiopharmaceutical.

When the emission data are reconstructed, it results in a tomographic image set that has not been corrected for attenuation of the 511 keV photons in the patient. Such images exhibit a characteristic hot periphery relative to the central body organs, which appear cold. To rectify this bias, an attenuation map of the patient is used to correct the emission data. On dedicated PET scanners this map was obtained by a transmission scan of the patient, using either a long-lived (270-day half-life) positron generator (⁶⁸Ge → ⁶⁸Ga) or a ¹³⁷Cs singles source, in which the linear attenuation coefficient of 662 keV is scaled to 511 keV. Because practically all PET scanners in current operation are now PET/CT units, the attenuation map for PET reconstruction is derived from the CT image set in which the attenuation (µ) map is scaled from CT energies to 511 keV.^78–80

FDG Positron Emission Tomography

FDG is an almost ubiquitous cancer-imaging agent for PET. The basis for tumor visualization by PET is the elevation of glucose metabolism by most malignancies. Most PET studies performed today are diagnostic FDG scans. Wahl has edited an excellent book that summarizes FDG PET within different tumors.⁸¹ The increased level of glucose metabolism by cancer cells is frequently reported in clinical practice by the standardized uptake value (SUV). This is a quantitative measure of FDG uptake defined by the ratio of the activity per cm³ in tumor (per gram assuming unit density tissue) to the administered activity per unit patient mass:

An example of an FDG PET/CT scan of a patient with ductal carcinoma with a positive level I lymph node is shown in Fig. 8-19 in which the SUV is indicated for both the primary and metastatic disease sites. Clinical scanners are routinely equipped with an SUV tool, which allows clinicians to draw an ROI around the tumor from which the SUV is automatically calculated. The SUV can be used to differentiate between benign and malignant disease. For lung nodules, an SUV greater than 2.5 is commonly indicative of malignant lung cancer, although other factors, such as size, shape, and location may reinforce the nuclear medicine clinician’s judgment. The magnitude of the SUV is often believed to be correlated with tumor aggressiveness. The variability among tumor types, patients, and even among lesions in the same patient suggests a large variability of glucose metabolism within tumors.⁸² There is now strong evidence to support the role of FDG PET in cancer staging.⁸³ In lung cancer there is a large body of evidence convincingly demonstrating that locoregional lymph node staging by FDG-PET/CT is significantly superior to CT alone.^84,85 In lymphoma and melanoma, PET/CT has been shown to improve the accuracy of staging as well as response assessment over that of conventional anatomic imaging.^86,87

FIGURE 8-19 • FDG PET/CT scan of a patient with ductal carcinoma with a positive level I lymph node.

Clearly, for any SUV analysis to be meaningful, it is essential in comparative studies that the patient be imaged at the same time post-FDG injection, in the same fasted and relaxed state to minimize competitive uptake by the brain, heart, and muscle, and this is of particular importance when using FDG PET for response assessment.

Uses of Positron Emission Tomography

Since the last edition of this book, PET/CT scanners have become widely available within hospitals in the industrialized world. Most of these are located in departments of radiology, but in several cases these instruments have been purchased and are being used as simulators by radiation oncology departments. The ease of this transition occurred because of the pre-existing widespread use of CT simulation, onto which the PET scanner was added using a common patient support system. Digital Imaging and Communications in Medicine (DICOM) transfer protocols readily allowed the porting of PET emission data into the commercial radiation therapy treatment planning systems, leaving radiation oncologists with the decision-making responsibility on how to use PET images in the treatment planning process.

The visualization of FDG (or any other tracer) images requires considerable care, because the viewing windows are less well defined than for CT. Therefore, the PET window in fused PET/CT displays may be set to provide the best match between PET and CT.¹⁰⁶ This practice can result in a misleading complacency. Erdi and colleagues¹⁰⁷ defined the PET window setting based on phantom studies, in which spheres of known size, representing lesions, were imaged within different ¹⁸F background levels. This study concluded that to accurately define the lesion size, the optimum window setting should set the upper window level to the maximum intensity voxel and the lower level set at a value of 42% of the upper level for spherical lesions of size greater than 4 cc. For smaller lesions, in which partial volume effects become significant, the optimum threshold level increased higher than 42% because of the resolution constraints of the camera. Furthermore, the lower-level threshold is not independent of the contrast (lesion–background ratio). Lower contrast necessitated elevating the lower threshold level, if accurate lesion volumetrics were to be obtained. Such empirical methodologies are effective at accurately determining lesion size within a homogeneous background (such as the lung) in which the methodology has been validated, but become problematic when attempting to unambiguously define the boundaries of a lesion within a nonuniform background (e.g., the prostate, because of the close proximity of the bladder). Several recent papers have appeared describing the variability in the appearance of lesion size when different proposed methods are applied to identify lesion boundaries. Schinagl and colleagues¹⁰⁸ compared five methods of tumor delineation from FDG-PET compared with CT-based delineation. Seventy-eight head and neck cancer patients were evaluated, for whom the primary tumor was delineated on CT, and compared with five PET-based gross tumor volume (GTV) determinations: (1) visual interpretation, (2) applying an isocontour of a SUV of 2.5, (3) using a fixed threshold of 40% or (4) 50% of the maximum signal intensity, and applying (5) an adaptive threshold based on the signal-to-background ratio. Whereas there was considerable variability in the results given by the different PET segmentation methods, the authors reported that on average all threshold-based PET-GTVs were smaller than on CT. Nevertheless, PET frequently detected significant tumor extension outside the GTV delineated on CT (15%-34% of PET volume). This study underscores that simple SUV threshold cannot be expected to handle the subleties of FDG uptake within tumors. This can be appreciated from autoradiographic images of FDG uptake that exhibit marked nonuniformity determined by microenvironmental factors.¹⁰⁹ For this reason, more sophisticated methods are being developed. Examples include the Gaussian mixture model segmentation technique proposed by Aristophanous and colleagues¹¹⁰ tested on selected PET tumor regions from NSCLC patients. The Gaussian mixture model assumes that the FDG uptake can be represented as a mixture of Gaussian densities representing different classes such as background (B), uncertain (U), and target (T). Another example is the morphologic filter-based approaches of Geets and colleagues.¹¹¹

Greco and colleagues¹¹² investigated the effect of a PET-derived contour on treatment planning with respect to the CT-only target volume in lung cancer. This study reported that the most prominent changes in the GTV have been reported in cases with atelectasis and following the incorporation of PET-positive nodes in otherwise CT-insignificant nodal areas. Although interobserver variability was still present following target-volume delineation with PET/CT, it is greatly reduced compared with conventional CT-only contouring. One of the difficulties of using PET in thoracic and abdominal tumors is the effect of respiratory motion, which blurs and reduces the tumor volume.

One of the tumor sites that has received the most attention regarding the incorporation and use of FDG PET images for radiation treatment planning is head and neck cancer.^113–116 Here the patient can be rigidly immobilized, allowing nongated PET studies of high quality. Yet how the PET information is used to delineate the tumor boundary or modulate the treatment plan varies between centers.

One of the difficulties of determining the accuracy of PET for defining the tumor volume is the lack of surgical corroboration. Such studies are of course extremely difficult to perform. One of the only such studies to attempt to compare the accuracy of different imaging modalities (CT, MRI, and FDG PET) against surgical data is the study by Daisne and colleagues¹¹⁷ for head and neck cancer. In a comparison of the GTV determined by each modality, it was concluded that the FDG PET volumes were smaller but, surprisingly, for the nine available surgical specimens PET was the most accurate.

There are numerous reviews on the use of FDG PET in radiotherapy planning for lymphoma, esophageal, and other cancers, and the reader is referred to them for more details.^118–122

Dose Painting

An addition to the use of PET to provide metabolic information into the procedure of target-volume delineation is the concept of using PET for biologically based IMRT, so-called dose painting or sculpting.^123,124 This is a large emerging field driven by advances in molecular imaging and its potential ability to determine radiobiologic attributes of the tumor noninvasively. Three of the most promising intratumor targets for dose painting are (1) FDG defining regions of highest glucose metabolism, (2) fluorothymidine (FLT) defining regions of highest cell proliferation, and (3) hypoxia tracers of which the earliest was ¹⁸F-fluoromisonidazole (¹⁸F-FMISO). Several alternative tracers are under investigation in radiotherapy-motivated clinical trials.

The FDG uptake within a voxel can depend on many factors: cell number, fraction of tumor or stroma, cell type, distance from capillary supply, PO₂, mitotic index, inflammation, and so on. Until the dependence on these factors is better understood, caution is advised if FDG is to be used for dose painting. Nonetheless, studies have been performed to assess the feasibility FDG PET–guided IMRT dose escalation in head and neck cancer, in which an upfront simultaneously integrated boost dose was administered to ¹⁸F-FDG PET–delineated tumor subvolumes within the GTV.¹²⁵ Dose-limiting toxicities resulted in the halting of this study. Pugachev and colleagues¹²⁶ investigated the microenvironmental determinants of FDG uptake in a rodent prostate cancer model and used autoradiography to show the highest regions of FDG uptake were correlated with regions of hypoxia. Rajendran and colleagues¹²⁷ showed that FDG was positively correlated with regions of hypoxia defined by ¹⁸F-FMISO imaging, but also found regions of high FDG uptake in regions not high in ¹⁸F-FMISO uptake. In recent work by Thorwarth and colleagues,¹²⁸ the FDG distribution was proposed for a uniform dose escalation of 10%, with a dose painting by numbers approach to the hypoxia map defined from dynamic ¹⁸F-MISO images. Schwartz and colleagues¹²⁹ performed theoretical IMRT plans to explore dose painting to FDG hot spots in which 66 Gy was prescribed in 30 fractions to FDG-avid CT abnormalities and nodal zones directly involved with disease, without prophylactic coverage of uninvolved neck levels. After 66 Gy, FDG-avid disease with 0.5-cm margins was boosted in 220 cGy increments until dose-limiting criteria were reached, but caution is advised on the routine clinical use of FDG-directed treatment planning until the accuracy of FDG PET/CT is fully validated.

Of greater potential relevance for radiation oncology are PET tracers that are associated with cellular proliferation. The most widely used proliferation tracer is ¹⁸F-FLT,^130–133 whose uptake is regulated by cytosolic S-phase-specific thymidine kinase 1 resulting in the tracers’ selective intracellular entrapment in dividing tumor cells. FLT uptake has been shown to correlate with pathology-based proliferation measurements, including the Ki-67 score, in a variety of human cancers.¹³⁴ However, FLT PET scans generally exhibit a lower overall uptake with SUVs less than three being typical, with high levels of activity in liver and bone marrow. In a multicenter study conducted in China by Tian and colleagues,¹³⁵ 16 patients with malignant tumor, 16 with tuberculosis, and 23 with other benign lesions were evaluated by both FDG and ¹⁸F-FLT PET/CT. The sensitivity for malignant tumor was 88% for FDG, significantly higher than for FLT (59%). However, ¹⁸F-FLT exhibited higher specificity at 77% versus 69% for FDG. This study reported that a combination of dual-tracer PET/CT improved the sensitivity and specificity up to 100% and 89.74%. A study investigating the use of FLT in the brain by Ullrich and colleagues¹³⁶ showed limited correlation of the FLT images with viable tumor in high-grade gliomas but that kinetic analysis of ¹⁸F-FLT tracer uptake could provide an in vivo assessment of tumor proliferation. The value of compartmental modeling to enhance the value and potential validity of PET imaging of radiotracers such as FLT is an area os active research.

Some of the radiotracers that have attracted the greatest interest for evaluating tumor response as well as for potential application in dose painting are the hypoxia markers.⁷² This is a consequence of the long-held belief that hypoxia is a major cause of treatment failure in radiotherapy, resulting from the radio-resistance of cells under low partial oxygen pressure. Whether this is true or untrue (because of tumor reoxygenation between treatment fractions), a noninvasive technique to measure tumor hypoxic fraction would allow the role of hypoxia as a therapy outcome prognosticator to be determined. Several hypoxia radiotracers have been proposed. The first to be extensively studied was fluoromisonidazole (¹⁸F-MISO) which was introduced into the clinic by Rasey and colleagues¹³⁷ and has since been continued by Rajendran and colleagues¹³⁸ who have demonstrated the ability to acquire good-quality images of tumor in head and neck, lung, brain, soft tissue sarcomas, and other tumors. It is recognized that such images of hypoxia may be useful for use in IMRT dose painting. One difficulty of ¹⁸F-MISO is associated with the short 110-minute half-life of ¹⁸F, restricting image acquisition to a maximum of about 3 to 4 hours postinjection, and at a time when the hypoxia tissue/blood ratio is only 1.3 to 1.4. As a consequence, ¹⁸F-MISO images are frequently of poor contrast. This results in the difficulty that any threshold level used to discriminate intratumoral hypoxia from nonhypoxia is highly sensitive to the threshold setting. An additional difficulty is the reproducility of the hypoxia imaging under these low contrast conditions. Nehmeh and colleagues¹³⁹ analyzed the intratumor distribution of ¹⁸F-FMISO from 20 head and neck cancer patients who underwent two pretreatment baseline scans to assess the reproducibility of hypoxia imaging for the purpose of dose painting. This study revealed considerable variation in the voxel correlation between scans caused by a combination of count statistics, differences in scan time post injection, and acute changes in the tumor hypoxia. An example of ¹⁸F-FMISO PET scan, showing the complexity of the intratumor distribution displayed alongside FDG and CT of the same patient, is given in Fig. 8-20. Nevertheless, preliminary studies to investigate the feasibility of ¹⁸F-FMISO for treatment planning studies in head and neck cancer patients based on these scans have been described.^140,141

FIGURE 8-20 • Example of multimodality image registration of (top left) computed tomography (CT), (top right) fluorodeoxyglucose (FDG), (bottom left) fluorine-18–labeled fluoromisonidazole (¹⁸F-FMISO), and (bottom right) fused FDG-¹⁸F-FMISO. Three enlarged images (indicated by arrows) of FDG, ¹⁸F-FMISO, and fused FDG-¹⁸F-FMISO also shown.

(From Lee NY, Mechalakos JG, Nehmeh S, et al: Fluorine-18-labeled fluoromisonidazole positron emission and computed tomography-guided intensity-modulated radiotherapy for head and neck cancer: a feasibility study, Int J Radiat Oncol Biol Phys 70(1):2–13, 2008.)

Other ¹⁸F hypoxia tracers are being explored, with higher octanol/water partition coefficient, with the hope that these will provide greater contrast (e.g., ¹⁸F-FETNIM, ¹⁸F-EF5, and ¹⁸F-FAZA).These are also being explored to define targets for hypoxia dose painting.^142–146 The compound EF5 was initially developed as an exogenous immunohistochemical marker of tumor hypoxia using fluorescently tagged antibody. Evans and colleagues^147,148 have shown the prognostic significance of hypoxia in brain tumors as well as in head and neck squamous cell tumors, based on quantitative immunohistochemistry of tumor tissue supporting the development of ¹⁸F-labeled EF5 as a PET imaging agent.

Studies are ongoing to overcome the low contrast associated with ¹⁸F-labeled hypoxia imaging agents and statistical variability and highly time-dependent hypoxia-to-background ratio observed in single-time point imaging with ¹⁸F-FMISO by using compartmental analysis of dynamic ¹⁸F-MISO data.^149,150 One of the models used to determine the rate of hypoxia tracer entrapment is described by Casciari and colleagues.¹⁵¹ Such models use the pharmacokinetics of tracer uptake extracted from dynamic PET emission data of the tumor and the arterial input function data (from arterial samples or ROIs over arterial blood) for the generation of parametric images, in which the voxel intensities reflect the rate constants for radiotracer entrapment. These rate constants are believed to be proportional to the amount of hypoxia within each PET voxel.

Another class of PET hypoxia radiotracer that has been extensively investigated in the context of radiotherapy applications is Cu(II)-diacetyl-bis(N4-methylthiosemicarbazone) (Cu-ATSM), which can be labeled with several positron-emitters: ⁶²Cu (T½: 9.74 min), ⁶⁰Cu (T½: 24.0 min), or ⁶⁴Cu (T½: 12.7 hr). The group at Washington University, St Louis, has shown interesting data suggesting an inverse correlation between the level of ⁶⁰Cu-ATSM uptake in tumor and clinical outcome in NSCL cancer, as well as for cervical carcinoma patients.^152–154 These findings provide strong support for the hypothesis that hypoxia is a negative prognostic indicator of radiotherapy outcome. Prior to these studies, preclinical investigations in a rodent model demonstrated a correlation between oxygen electrode measurements and ⁶⁰Cu-ATSM accumulation. Chao and colleagues¹⁵⁵ conducted a study to demonstrate the feasibility of biologically based IMRT with the selective targeting of dose to regions of high ⁶⁰Cu-ATSM accumulation. The mechanism of hypoxia targeting of Cu-ATSM is still under investigation, but it believed to be an effective tracer for some but not all tumors.^156–158

Evaluation of Treatment Response

PET provides an alternative perspective relative to CT and MRI in measuring treatment response. Several centers perform FDG scans pre- and postchemotherapy and radiation therapy. The ability to discern viable from necrotic tissue has been an important application of PET, and was first used to assess response after stereotactic radiosurgery.¹⁵⁹ However, the difficulty of separating viable tumor posttherapy from inflammation has reduced the reliability of FDG as a quantitative index of response. The potential use of FDG in measuring radiation response has been expressed by Oku and colleagues¹⁶⁰ who reported, for rectal carcinoma, a change in the mean value of SUV from 7.6 to 4.2 before and after radiotherapy. There was a significant difference in SUV reduction between the groups with and without recurrence, suggesting SUV to be a good prognostic indicator for long-term prognosis of rectal cancer patients. Allal and colleagues¹⁶¹ and Greven and colleagues¹⁶² have assessed the value of PET FDG in predicting the outcome of radiation therapy for head and neck cancers, concluding that there is little use in early scans 4 weeks post treatment, but improved predictive capacity at several months post therapy. Erdi and colleagues¹⁶³ investigated the changes in FDG uptake in two lung cancer patients throughout the course of fractioned therapy, imaging patients weekly from before the treatment until the completion of treatment. This study was the first to describe the metabolic changes to the tumor that occurred during the whole course of radiation therapy, and provided a discussion of the observed effects of the radiation on normal tissues in the treatment field. The authors report essentially no changes in normal tissues at doses lower than 9 Gy (four fractions). By 18 Gy, increased FDG uptake was observed in the tissues in the back, suggestive of inflammatory changes in muscle and soft tissues at the treatment portal. At 27 Gy, there was intense FDG uptake associated with clinical esophagitis; by 36 Gy and by 54 Gy, significant reduction of uptake in the normal marrow of the lumbar spine was observed.

The value of PET-FDG in monitoring treatment response to combined chemoradiation therapy is clear in some disease sites. In esophageal cancers there are data that suggest patients with a greater than 55% drop in SUV(max) have a significantly improved progression free survival.^164–166 In colorectal cancers, data suggest patients who respond have a greater than 67% decline between baseline and presurgical assessment.^167,168 PET-FDG is now widely used to monitor treatment response to lymphoma, from therapies with or without radiation. These studies indicate that the persistence of metabolic activity as measured by FDG is a negative prognostic indicator for treatment response and also demonstrate that PET/CT imaging is much more effective than CT alone in predicting response to either chemotherapy or radiation.^169,170 FDG PET scanning for recurrence after radiotherapy is complicated by the inflammatory response in the normal tissues surrounding the tumor. The clinical introduction of FDG PET scanning in brain tumors occurred earlier than for other cancers, because of the availability of dedicated brain PET scanners before the introduction of whole-body PET scanners in the early 1990s.^171–173 The differential diagnosis of brain tumor recurrence versus radiation necrosis is difficult, because on cross-sectional imaging there may be significant mass effect that may increase over time and still be the result of radiation necrosis. For this reason, FDG scanning may be useful in certain situations to detect recurrence versus radiation necrosis, although radiolabeled amino acids may better serve this purpose (see next section). FDG PET is now being widely used to assess the radiotherapy treatment response and recurrence of head and neck cancers as well as for disease staging.^174–178 Posttherapy false-positives have been a problem with FDG PET, but when posttherapy FDG is negative, a high-accuracy negative predictive value of 92% has been shown.¹⁷⁹

Alternative Positron Emission Tomography Tracers

FDG PET has been so successful in diagnostic oncology that it has overshadowed the development of alternative tracers for cancer detection. But not all tumors image well with FDG, because of either low glucose metabolism or proximity to hot structures such as the prostate, unless they are high-grade tumors with metastatic spread.¹⁸⁰ However, which tracers are useful for imaging tumor biology is a big unanswered question and a large area of active research. Some of the PET tracers that have been clinically used for tumor imaging include ¹¹C-methionine, ¹¹C-thymidine, ¹²⁴I-iododeoxyuridine, F-FLT, ¹¹C-choline, ¹¹C-acetate, and ⁶⁸Ga-DOTATOC.^181–188

The short 20-minute half-life of ¹¹C-based tracers limits the use of these tracers to hospitals with in-house cyclotrons and radiochemistry facilities, and requires the radiopharmaceutical to exhibit a rapid (~10-minute) uptake in the tissue of interest. Studies with ¹¹C-methionine in prostate cancer has shown uptake peaking at approximetly 10 minutes postinjection with the tracer remaining at a plateau, thereby facilitating whole-body PET imaging with decay correction for this radiotracer.^181,182 The low metabolic amounts accumulated in the bladder result in prostate images of excellent contrast for ¹¹C-methionine as well as ¹¹C-choline and ¹¹C-acetate relative to FDG (Fig. 8-21). There is an increased interest in ¹¹C-choline, because of evidence of its potential for distinguishing more aggressive versus less aggressive prostate tissue on the basis of differential incorporation of ¹¹C-choline into cell membrane.¹⁸⁹

FIGURE 8-21 • Coronal whole body sections of a prostate cancer patient scanned initially with ¹¹C-methionine (A) 20-minutes postinjection followed 120 minutes later by an FDG (B) PET scan. Note the improved discrimination of the prostate from the bladder in the ¹¹C-methionine image. All sites of bone disease were identified by both ¹¹C-methionine and FDG for this patient.

This approach may offer a nuclear medicine alternative to choline-citrate nuclear magnetic spectroscopy. Studies have also reported potential advantages of ¹¹C-choline over FDG for imaging of the brain and mediastinum.^190,191 However, the short half-life of ¹¹C limits the use of these PET tracers to hospitals with on-campus cyclotrons. For this reason, radiochemists have developed fluorinated amino acid tracers such as FACBC and fluorotyrosine.^192–194 There is also a ¹⁸F-fluorocholine that has been successfully used for clinical imaging of prosate cancer.^195,196 The utility of ¹⁸F-fluorocholine has been compared to sextant localization of malignant prostate tumors from 15 patients who underwent PET prior to radical prostatectomy. Histopathology demonstrated malignant involvement in 61 of 90 prostate sextants, indicating the potential value of ¹⁸F-choline PET at localizing dominant areas of malignancy within the prostate.

The use of thymidine analogs offers the potential to image cell replication directly, rather than a metabolic process inferring viability, which is especially important when measuring posttherapeutic response. Carbon-11–labeled thymidine was compared with FDG in a limited number of patients with lung cancer and sarcoma to evaulate treatment response.¹⁹⁷ There were promising results, but difficulties associated with the short half-life led these investigators to pursue fluorothymidine (¹⁸F-FLT).^129,130 Blasberg and colleagues¹⁸⁴ have used ¹²⁴I-iododeoxyuridine with PET to measure the proliferative capacity of brain tumors, although the low 0.24 positron abundance of ¹²⁴I is a disadvatage for PET imaging with this isotope.

PET tracers also may play a role in assessing the aggressiveness of the disease phenotype. There are several examples of this arsing from multiple PET (or PET and SPECT) tracer studies. For example, patients with metaststic thyroid cancer may exhibit lesions with high radioiodine uptake and low (or negligable) FDG and vice versa, suggestive of a reciprocity of these two tracers in determining cancer differentiation. Good radioiodine uptake and low FDG SUV indicate highly differentiated disease, whereas high FDG and low radioiodine uptake indicate poor differentiation.¹⁹⁸

Another class of nuclear medicine imaging agents, which may provide measures of the level of tumor differentiation, are the hormonal receptor binding tracers such as ¹⁸F-fluoro-17beta-estradiol (¹⁸F-FES) in which uptake has been shown to correlate with expression for estrogen levels in breast cancer^199,200 and ¹⁸F-fluorodihydrotestosterone for androgen receptors in prostate cancer,^201,202 which are now involved in preliminary clinical trials. Alternatively, antibodies with binding specificity for tumor-associated antigens, labeled with a long-lived PET tracer such as ¹²⁴I,^{105 64}Cu,¹⁰⁶ and ⁸⁶Y¹⁰⁷ will provide the ability to perform serial PET imaging for tumor-specific antibodies for several days postadministration, with the benefits of greater quantitative accuracy needed for tumor and normal organ dosimetry during RIT trials. The high specificity of radiolabeled antibodies imaged several days postinjection (after wash-out from the blood compartment) can provide images of tumor-associated antibodies. Late imaging with ¹²⁴I labeled cG250 antibody binds to carbonic anhydrase-IX, which is constitutively expressed in clear-cell but not non–clear-cell renal carcinoma. This allows presurgical identification of the histologic subtype that could have important implications for the choice of treatment for renal cancers.²⁰³

Picture Archiving and Communication System

In modern hospitals and radiology departments, all imaging equipment produces and handles digital images, and data and is interconnected through a comprehensive PACS.²⁰⁴ In a standard PACS design, imaging modalities and equipment such as CR, ultrasound, CT, MRI, SPECT, PET, workstations, archive servers, and film printers are all connected together through a backbone or network. By design and function, a PACS encompasses principally the physical links and communication protocols between the imaging equipment (production, visualization, and printing) and the archival systems. Image production (e.g., individual modalities) and high-level processing and visualization are independent components that communicate through the PACS but are based on different technical principles. A PACS interacts with hospital information systems (HISs) and radiologic information systems (RISs). Whereas the HIS handles the demographics of a patient and his or her global management within the hospital mainly for administrative purposes, the RIS is targeted at radiology’s specific needs and deals with workflow management, including modality booking, clinical reading, and reporting. By natural extension in the age of the Internet, PACSs often include connections to the outside world for teleradiology and telemedicine applications.

Although the PACS concept initially suffered from a relatively closed proprietary approach of most manufacturers, PACSs now benefit from the widespread acceptance and deployment of the DICOM standard from the American College of Radiology–National Electrical Manufacturers Association.²⁰⁵ This standard provides a realistic foundation for the exchange and storage of all medical imaging data across and between compatible imaging and therapy equipment. Through its relative openness, DICOM has also provided a means for independent companies or academic entities to develop archival, imaging, and processing systems that can nicely complement the ones proposed by the major manufacturers.

Radiotherapy PACS

An essential condition for the successful use of electronic portal radiographs in the clinic is that they be a part of a PACS specifically designed for radiotherapy departments. Several such systems are commercially available and provide the various tools that are needed for portal image acquisition and verification, and for integrating this information with other parts of a patient’s treatment (i.e., from treatment simulation, to planning, delivery, and follow-up). Radiotherapy PACS functions include transfer of reference images from the simulator, CT simulation, or treatment planning system; portal image acquisition; convenient selection and display of portal and reference images and related patient data; tools for image processing; clinician’s approval of images through a specific worklist; and means of communicating messages or information related to images among clinicians, therapists, and other radiotherapy staff.

Windowing

Most digital display systems have historically had a depth of 8 bits (also called a byte). This depth produces 256 (2⁸) gray levels or monochrome luminance values from 0 to 255 (= 2⁸−1), which is enough dynamic range for the gray nuances that can be differentiated by the human eye. In medical imaging, for example, because of the initial design of CT and limitations of analog-digital converters (ADCs), images generally have a depth of 12 bits (2¹² > values from 0 to 4095 = 2¹² −1, e.g., for CT) or more. From this discrepancy came early the need to optimally convert a 12-bit data set to an 8-bit representation. For this conversion, a simple proportional scaling ([0,4095] > [0,255]) is the most obvious solution. Because of the dynamic under-sampling that it implies, however (16 successive input values would produce only one output value), such an approach would then make it very difficult to assess the subtle dynamic variations that could actually be contained within the initial 12-bit (or more) data range. Because the full spectrum of 12-bit values is seldom of interest to radiologists, the concept of value windowing was introduced to expand the full contrast of the output/display range over the useful input dynamic range. For CT, absolute density (in Hounsfield units [HU], as described earlier) dynamic windows are either defined by minimum and maximum values or by center and width in absolute units. Most often on clinical systems, preset windows are used to increase the contrast of specific tissue types as the examples in Fig. 8-22A show. Combining several windows on a single representation also allows enhancement of the contrast of several normally disjoint density ranges (see Fig. 8-22A).

FIGURE 8-22 • Contrast enhancement of a chest computed tomographic image through Hounsfield unit windows (A), various look-up table (LUT) shapes (B), and histogram equalization (HE, either global or local) (C). The images in B and C are based on the dual range image in A.

Lookup Tables

In current display systems, the conversion between a digital image and the display screen occurs through an LUT.²⁰⁶ An LUT simply converts an index derived from the input image data to an output value. This output value is then used to drive the luminance input on the display screen to produce a point with brightness proportional to the incoming signal at the corresponding image location (Fig. 8-22B).

Histogram Equalization

The appearance and contrast of images can be enhanced by other (e.g., nonlinear) means than display window or LUT shape. One such method is to redistribute the pixel values so that they optimally cover the complete available display range of Gy levels or intensities. The most common processing of this type relies on the histogram of the initial image. A histogram is a curve that shows, for every possible value in an image, the total number of all pixels at this value. Histogram equalization^207,208 ends up redistributing the initial image values so that the new image’s histogram becomes uniform or linear. This range spreading, which can be confined only to select parts of the image, then produces an optimal contrast in the image (Fig. 8-22C), but, as with changing the shape of the LUT, will then suppress the direct linear relationship between displayed and original values.

Digital Filtering

Space versus Frequency

As with standard photography, digital images can be processed through various filters to modify, suppress, or enhance some of their characteristics. In biomedical imaging, the application of most digital filters is generally empirical and aims at enhancing visual perception to improve the detection or characterization of a particular lesion. Even more so for a clinical use, the user should always assess whether such processing might distort the image presentation in a manner that could be unsafe to the patient.

Any 2-D image can be described both in the spatial domain as a collection of pixels, or in the frequency domain (i.e., as a spectrum of frequency components) after a 2-D Fourier transform (FT).²⁰⁹ Because of the space/frequency duality of images through the FT, frequency-based filters can generally be defined and applied in either of these two domains (Fig. 8-23): (1) on spatial images, a kernel (2-D arrays of values defining a filter) is convoluted with the image: every pixel of the original image 2-D array is mathematically combined with the kernel to produce a new image; (2) on frequency representations, a filter function (frequency window, etc.) is multiplied with the 2-D frequency spectrum representing the FT of the original image. The result is then transformed back through an inverse FT to the original 2-D spatial domain in which the new image can be displayed.

FIGURE 8-23 • Fourier transform and filtering of an abdominal computed tomographic image in the frequency (A) and the spatial domain (B).

In the spatial domain, the main parameter of a filter is the size of the kernel that directly relates to the size of the features to be preserved or affected. Because simple multiplications in the frequency domain take less time than 2-D (or 3-D) matrix operations in the spatial domain—and despite the need for a forward and reverse FT for getting in and out of the frequency domain—there is generally a significant reduction in computing cost between spatial- and frequency-based convolutions. For this reason, under suitable conditions and although space is intuitively simpler to comprehend than frequencies, Fourier filtering is often used to speed up processing times.

Contouring

A volume of interest (VOI) can be segmented from a volume of data with different techniques, depending on the data characteristics and on the requirements of the user.^210,211 Contours can be drawn around a structure of interest on every slice that shows a part of it (Fig. 8-24). At the end of the process, these contours can be displayed on a workstation and interactively manipulated (translations, rotations, zooms) in real time to provide a relatively crude 3-D representation of the corresponding structure (see Visualization).

FIGURE 8-24 • Segmentation of a magnetic resonance brain tumor image through contour, painting, and region growing.

By preprocessing the initial image through edge filters, contouring can become semiautomatic and assist the user in defining more objective—and possibly accurate—outlines for the objects of interest. So-called active contours or snakes^211,212 can be seen as an even more automated version of edge tracking. In active contouring approaches, an initial contour is modified (e.g., inflated) iteratively to maximize contents while preserving edges and minimizing the contour’s energy.

Tagging/Painting, Region Growing

Another manual way of extracting a structure of interest from an image volume is to paint or tag its pixels in every image that contains a part of it (see Fig. 8-24). A semiautomated variation of the painting approach is the thresholding of data that will extract only those voxels with a value—or other specific characteristics—that falls within a user-defined value range.

A variation of tagging is region growing, in which all voxels falling into some kind of user-defined condition (e.g., a combination of value range, texture, continuity, topology, gradient, shape, etc.) are joined together from one or several seeds inside the VOI.^211,213,214 Providing that a good growing criterion can be defined, region growing is a very powerful semiautomated segmentation tool. The seeds can either be any specific location inside the 3-D volume, or the result of a previous segmentation operation. Tagging or region growing produce a 3-D mask over the structure of interest. The resulting mask can then be used to apply further processing on the data that belong to this structure of interest inside the full 3-D data set. Such a mask is generally more comprehensive than contouring because it may include finer and more complex details that are not always easy, or even possible, to extract through contours alone. Of special relevance in a “region-growing” paradigm, classification techniques—that can either rely on single or multicomponent data sets²¹¹—are able to provide statistical masks for the refined characterization of structures that may have a mixed content, either because of their biology or because of a relatively low spatial resolution for imaging.

More Advanced Segmentation

Research in segmentation—an ever-thriving domain in digital imaging—generally aims at providing clinically compliant techniques that are as automated as possible (including in 3-D) to reduce variability while increasing robustness, objectivity, accuracy, and possibly the speed of the process. Many techniques rely on improving and extending to higher dimensions approaches that have already been successfully implemented in 2-D.^211,215,216 Other approaches rely on combining more information into an iterative segmentation process to drive it toward a solution that will satisfy all relevant criteria. For instance, among such approaches and in parallel with recent innovations in clinical imaging, the additional information that is available through multimodality protocols (see later text) can likely be further exploited for improving the characterization and segmentation of structures of interest.^217,218

2.5-D Display of 3-D Data Sets: Slices and Multiplanar Reconstruction

All tomographic modalities produce cross-sectional slices that can be presented as 2-D images and are called source images. Window-based display systems allow the screen space to be partitioned between various images that can either be from the same study or from different ones. Each data set can be displayed with its own parameters (value windowing, color scale, zoom, etc.).

For tomographic modalities, source images can be stacked together to recreate the volume in the field of view at the time of the acquisition. This volume can then be cut along orthogonal planes (standard MPR)²¹⁹ to produce images with other orientations than the source images (Fig. 8-25A). Such a representation can be very useful to display structures that are not “well oriented” in regard to the original acquisition’s main axis. A natural extension of the classical orthogonal MPR approach is the oblique plane extraction and display, which shows a cut of the reconstructed volume along any arbitrary plane (Fig. 8-25B).²²⁰ Another extension of MPR is the curved plane extraction and display, which shows a bidimensional nonplanar cut along a 3-D curved path inside the reconstructed volume (Fig. 8-25C).

FIGURE 8-25 • A, Orthogonal, B, oblique, and C, curved multiplanar reconstruction (MPR) from a magnetic resonance volume of a head.

Because tomographic volumes are generally anisotropic (i.e., the elementary voxels do not have the same absolute size on all three axes of the cartesian space), the initial data most often need to be interpolated to produce adequately reconstructed images in orientations other than that of the acquisition.

3-D Display of 3-D Data Sets

As previously mentioned, tomographic data sets are made up of 2-D slices that are discrete cuts through a 3-D volume considered as static during the whole acquisition. Under this assumption, these data can be used to extract and visualize information or objects in 3-D volumetric renderings. Such 3-D renderings are likely more realistic than 2-D slices alone and can be assessed in direct relation to the patient’s actual morphology.

Contour, Wireframe Display

For a simpler approach, contours can be drawn manually or (semi-) automatically on the source images (see Image Segmentation section). Contours from the same structure can then be grouped together across slices and rendered in 3-D through the spatial parameters (e.g., pixel size and slice spacing) of the source images (Fig. 8-26A). When the contour points of the same structure are joined across slices, the final display makes up a wireframe enclosing the structure of interest (Fig. 8-26B), all of which can be manipulated interactively on a screen (see 3-D Manipulation section).

FIGURE 8-26 • Segmented brain from a computed tomographic (CT) image of a head: A, contours, B, wireframe, and C, surface rendering.

Volume Rendering

Compared with surface rendering, volume rendering (VR)^210,221 does not intrinsically require the segmentation of structures, but uses all the voxels from the volume simultaneously. For VR, voxels are assigned a color and an opacity/transparency that can be linked to their value (e.g., through a LUT scheme) or to any other characteristics (Fig. 8-27A). The volume is then represented to the user or observer through the same ray tracing (or casting) paradigm as that previously described for rendering surfaces. A special (i.e., simplified) mode of VR is maximum intensity projection, in which only the highest value on a ray is projected to the observer (Fig. 8-27B). This technique is mainly used for rendering isolated, high-contrast objects such as bones or vascular structures in CT or MRI angiography. Because there is no binary decision such as when setting a surface threshold, VRs generally look more realistic and convey more textural information than surface renderings. A VR approach also allows arbitrary cuts to be performed through the whole volume and rendered to the user in relation to the 3-D VR.

FIGURE 8-27 • A, Volume rendering (VR) of a computed tomographic (CT) image of a head with variable transparency (full head on the top, and cut head at the bottom); B, maximum intensity projection (MIP); and C, digitally reconstructed radiography (DRR) of the same head.

DRRs, which have been discussed previously in relation to treatment planning, are a special application of VR. They simulate a standard 2-D projection x-ray image (such as on a film or from a portal system) from the original CT volume data. By applying strict models based on the absolute H units of the CT image, on the properties of the x-rays, and on the response of the detector (e.g., film, phosphor, EPID) to be simulated, DRRs can be made to look very realistic. By tuning intensity and transparency, and selecting among various combination techniques (direct integration, weighed integration, etc.) of CT voxels along the rays, DRRs can also look many different ways (Fig. 8-27C). Radiation energies and transport characteristics that mimic standard x-rays, a CT simulator, or portal imaging can be simulated when generating the DRR. An obvious plus when compared with traditional x-ray planar projections (e.g., from film or C-arm) is that the DRR orientation can be totally arbitrary and linked to any given geometric transformation. This feature can be used, for example, when trying to align a planning data set to the patient during treatment. An important characteristic of DRRs that sets them apart from traditional VR is that they can also use a cone beam geometry for the simulation of a point x-ray source. Perspective imaging, which has not been addressed here but can simply be added to the viewing transformation during ray tracing, actually relates to such geometry.

Parametric Imaging

Besides doses calculated by computerized planning systems, relevant parameters for radiotherapy targeting or assessment may be extracted from static or dynamic functional data sets through either PET or MRI. (Note that CT and SPECT may also provide additional functional information.) For MRI or PET, a contrast agent or radiotracer, respectively, is injected and rapid imaging provides a dynamic sequence of the flow and uptake of the material in the body. Image-based dynamic parameters such as time of arrival, maximum uptake slope, or time of maximum contrast can be extracted and mapped onto the anatomic image. Subtle variation in the dynamics of the contrast or tracer may then provide additional information to improve the identification of tumors or other structures.^222,223 Slightly more complex compartmental modeling approaches indirectly estimate metabolic parameters.²²⁴

Multimodality Imaging

Because of different technical principles and responses to specific protocols, data coming from various modalities or protocols—and the parameters that can be extracted by further processing—are most often complementary when it comes to evaluating in vivo morphology or metabolism.^225–227 The goal of multimodality imaging is therefore to merge all information at hand to better synthesize and assess its global content.²²⁶ This approach can be used either for visualizing information of a complementary nature simultaneously, or for enhancing information from one modality with data or parameters extracted from another one.²²⁵

Fusion: Visualizing Multimodality Data Sets

Once registered, two data sets can be represented as two side-by-side 2-D sets of MPR images, with or without a coupled cursor. Another way is to blend both images—with different color scales and a varying blending ratio—into a single one. When dealing with 3-D objects, objects extracted from one data set can be visualized along objects extracted from the other one. Finally, information from one data set (e.g., the “functional” one) can be projected on the surface of objects from the other one (e.g., the “anatomic” one) (Fig. 8-28). Such merging of information²²⁶ can obviously also take place with volume-rendering techniques (Fig. 8-29). A special paradigm of multimodality imaging is the registration and fusion of information that comes from imaging instruments with information that may be entirely parametric or virtual. Such an example of hybrid registration and fusion is the combination of beams and dose estimates from a radiotherapy plan with follow-up PET–CT data sets. Once further established, the goal of such a protocol would be to assist physicians in differentiating what could be a tumor recurrence from what could be a radiotherapy-derived inflammatory response (see Fig. 8-29).

FIGURE 8-28 • Multimodality (MM) visualization of a brain tumor. A, Multiplanar reconstruction (example of an axial slice through the tumor in side-by-side, checkerboard, and colorwash modes) of computed tomography (CT), magnetic resonance (MR), and positron emission tomography (PET) for the same patient after spatial registration of the three modalities. B, Surface rendering of the skull and brain (solid brain on top, wireframe at bottom) derived from CT, tumor derived from MR (bottom), and color derived from PET. MM volume rendering can follow similar paradigms.

FIGURE 8-29 • Multimodality (MM) paradigm extended to parametric simulation data combined with anatomy and function. Left, Anatomy from simulation CT combined with radiation therapy planning and the estimation of beam geometry and dose. Right, Posttherapy PET/CT that demonstrates various uptakes in the thoracic area. Center, Nonrigid registration of the posttherapy data with the simulation CT and planning permits a critical assessment of whether any leftover and new uptakes might be true recurrences or possibly caused by posttherapy changes (e.g., pneumonitis) inside the beams.