One of the most important pioneers in “conformation therapy” was Shinji Takahashi, who described many of the important concepts of conformal therapy delivery and 3D treatment planning in a 1965 monograph.¹ Takahashi’s innovations included early multileaf collimators, automated (mechanical) conformal beam shaping, dynamic conformal treatments, orthogonal light beams to identify the machine isocenter, and 3D tumor models based on early tomography.¹ Other notable work in this area was performed by Harold Perry and colleagues² in Detroit and by Proimos, Wright, and Trump at the Massachusetts Institute of Technology (MIT)–Lahey Clinic.³^–⁷ Another early approach to conformal therapy, known as the Tracking Cobalt Project,^8,⁹ was led by Green, Jennings, and others at the Royal Northern and Royal Free Hospitals in England. First reported in the late 1950s,¹⁰ and summarized by Jennings,⁸ a series of mechanical, electronic, and, finally, computer-controlled treatment machines were developed to track disease spread, particularly along lymph node chains. By 1980, the computer-controlled version of the tracking system was in clinical use,¹¹ although Brace summarized the major limitations to the delivery technique: “The major obstacle to the routine use of conformation therapy is treatment planning.”¹¹ Finally, workers at the Joint Center for Radiation Therapy (JCRT) in Boston added computer control to a modern linear accelerator, so that the treatment table, gantry, collimator, collimator jaws, dose rate, and other parameters could be controlled dynamically while the beam was in use. The JCRT achieved the delivery of what is now called “dynamic conformal therapy,”¹²^–¹⁴ ¹⁵ a modern basis for computer-controlled conformal therapy.

As already described in the quote (above) by Brace, treatment planning was one of the main limitations for the conformal delivery techniques that were developed during this time period. The introduction of computed tomography (CT) in the early 1970s was a key to the development of the modern 3D planning that is crucial to conformal therapy, because it made available a complete 3D description of the anatomy of each patient that could be the basis for planning. Early evaluation of the use of CT in treatment ¹⁶^–¹⁸ quickly led to widespread use of CT-based planning (see Ling¹⁹) as well as new interest in the use of inhomogeneity-corrected dose calculations, because CT provided the necessary electron-density maps of the patient.²⁰ Other imaging data, including magnetic resonance imaging (MRI) and positron emission tomography (PET), also became available and began to be used for planning in the mid-1980s.²¹

With the widespread implementation of CT-based planning, it became possible to make use of continuing improvements in computer technology and new software developments to create fully 3D treatment planning systems that incorporated 3D graphics and the “beam’s-eye view” (BEV), a 3D graphic reconstruction of the patient anatomy projected into the divergent geometry used by the x-rays in the radiation beam ²²^–²⁶ (Fig. 15-2). Using BEV displays to design field shaping²⁷ and evaluate coverage of tumor and sparing of normal tissues is perhaps one of the most effective concepts in the entire 3D planning paradigm. Routine clinical use of 3D radiation treatment planning (RTP) began in 1986,²⁸ and many academic centers began development and then use of 3D planning systems in their clinics.²⁹^–³²

Figure 15-2 Beam’s-eye view of a brain tumor showing a multileaf collimator shaped to the violet target volume. Normal tissues include brainstem (white), chiasm (green), and optic nerves (red).

The development of 3D treatment planning systems helped drive the need for and design of more sophisticated treatment delivery systems because the 3D planning systems demonstrated treatment improvements that would be possible to use clinically if more sophisticated machinery were available to make efficient delivery possible. The first treatment machine designed specifically to perform computer-controlled conformal radiotherapy (CCRT), the Scanditronix MM50 Racetrack Microtron, was developed during this same time period.^33,^34,³⁵ Among other unique features, this machine included a fully computerized control system and a computer-controlled multileaf collimator (MLC)³⁶ consisting of two sets of thin tungsten leaves that are used to shape the radiation field. Virtually all other radiation therapy machines have since that time also implemented computer control systems and MLC systems.³⁷^–³⁹

The capabilities of computer control and MLC systems have made possible the delivery of very complicated plans, including those that make use of modulated intensities (a beam with different intensities in different parts of the field). Intensity modulation created using multiple segments ⁴⁰^–⁴³ ^,44 ^,45 or dynamic MLC motions^33,^46,⁴⁷ and computer plan optimization (inverse planning)^48,⁴⁹^–⁵¹ ^,52 have been integrated into IMRT.⁵³ The basic concepts of IMRT were described in 1987 by Brahme^33,^48,⁵⁴ and a practical implementation was described by Bortfeld soon after.⁵⁵ The combination of the flexibility of computer-controlled IMRT delivery with sophisticated plan optimization techniques has made IMRT an extremely powerful tool that can be used to perform conformal therapy.

The initial commercial IMRT implementation by NOMOS in 1992 ⁵⁶ was a form of IMRT now called serial tomotherapy, in which patients were treated slice by slice (as with early CT scanners) by the machine rotating around the patient and a special multileaf collimator (MIMIC) that performed the intensity modulation. Within a few years, all major vendors had implemented MLC systems with leaf widths varying from 1 cm to a few millimeters that could perform IMRT using either dynamic motions of the MLC leaves (DMLC) or a number of static segments (shapes), now called SMLC.⁵³ A more sophisticated implementation of tomotherapy based on helical delivery of IMRT, helical tomotherapy,⁵⁷ also became widely disseminated. In the last several years, a number of vendors have now developed a rotational MLC-based IMRT technique (IMAT), which was originally described by Yu,⁵⁸ and is now called VMAT (volumetric modulated arc therapy).^59,⁶⁰ Inverse planning has also developed substantially during this time. Though much of this optimization makes use of quadratic weighted sum cost functions and simple gradient-based search algorithms, there have also been developments of sophisticated cost functions⁶¹ and the use of more biologically related costs such as normal tissue control probability (NTCP) models and equivalent uniform dose (EUD). Most systems use the weighted sum cost functions, but there has been development of sophisticated multicriteria methods^62,⁶³ that more directly take into account the numerous optimization goals involved in a typical clinical radiotherapy treatment plan.

One of the developments that made the conformal therapy revolution possible was the development of amorphous silicon flat panel imagers,⁶⁴ which allowed effective electronic portal imaging verification of the accuracy of these newly conformal fields. This technology then was further developed, first for kilovoltage (diagnostic quality) imaging and then to provide cone beam CT (CBCT) capability using kilovoltage imaging systems mounted directly on the treatment machine.⁶⁵ The availability of these high-quality CBCT or kilovoltage imaging modalities directly on the treatment machine led to the development of image-guided radiation therapy (IGRT), in which diagnostic imaging was used to correct patient setup and positioning for treatment every day. IGRT processes have greatly increased the delivery accuracy possible and have led to the possibility of much smaller margins for setup errors, as well as enough confidence in targeting accuracy that stereotactic body radiation therapy (SBRT) is now used routinely to give very high doses (as high as 20 Gy/fraction) to well-localized targets in the liver, lung, and other sites. The use of IMRT and the proper handling of patient motion, respiration, and other “4D” issues is a major thread of much research and development at the current time.

Issues For Clinical Use

A number of important clinical features are crucial for planning and delivery of high-quality conformal therapy, and these issues should be carefully considered throughout the conformal therapy planning and delivery process.

• Conformal therapy attempts to carefully conform the dose to the target, so target delineation and the desired dose distribution must be carefully and accurately defined.

• Patient immobilization and localization (setup accuracy and motion) must be considered throughout the process to minimize normal tissue that will be irradiated because large margins will significantly decrease the advantages provided by conformal treatment.

• Improvement of the clinical results achieved by conformal therapy, compared with standard techniques, depends on choosing the correct tradeoffs between target coverage and normal tissue sparing, so these choices must be made carefully and appropriately.

• The dose distribution achieved is important. The techniques used for planning and delivery (e.g., DMLC IMRT, VMAT) are simply the means to achieve the desired dose distribution.

• The quality of the delivered dose distribution can be destroyed by motion, respiration, setup error, or involuntary motions during treatment, so use of motion control (e.g., active breathing control) or respiratory gating, 4DCT for planning, and IGRT for setup and motion analysis are all important for the quality of the treatment.

Planning For Conformal Therapy

Treatment planning is one of the most critical parts of the conformal therapy process. In this description, we include all preparatory aspects of the planning process, including many activities that occur outside the radiation therapy planning (RTP) system. Many treatment delivery issues (e.g., setup accuracy, patient motion, portal and localization imaging) are briefly mentioned here and are more completely described later. Figure 15-3 shows a schematic of the basic components of the planning process for both forward (interactive) and inverse planning (i.e., IMRT optimization).

Figure 15-3 Schematic chart of the basic components of the planning process—shown for both forward and inverse planning.

Positioning and Immobilization

One of the basic ideas of conformal therapy is to minimize the dose to normal tissues while conforming the dose to the target, so it is of course crucial to accurately position and immobilize the patient for each procedure in the planning and delivery process. One of the first clinical decisions to be made for each patient includes what position to use for the patient’s treatment and whether any positioning and immobilization devices or aids will be used.

Basic patient positioning, including location of the arms and legs and positioning of the patient (supine, prone, or in some other more unusual position), depends mainly on two issues: (1) patient comfort and stability and (2) the beam directions that will be used. In most cases, conformal therapy plans make use of three or more beams that cross fire on the target from a number of different positions arranged around the patient, so the patient is typically positioned with both arms up (if the target is somewhere in the torso) or arms down (head and neck and brain targets). For superficial targets, the target is typically positioned facing up, but for most deep tumors, the cross-firing beam directions can be achieved with the patient in standard supine or prone positions, whichever is most stable and accurately set up. There have been studies of the benefits of various positioning decisions (e.g., prone versus supine) for patients with prostate tumors ⁶⁶; there is some debate about the relative merits of the possible anatomic changes that occur for the prone versus the supine position, relative to other advantages and disadvantages for planning, daily setup, and respiratory motion-related stability.

The use of various types of so-called immobilization devices to help with patient positioning and immobilization for conformal therapy has run the entire gamut of possibilities, from the use of stereotactic head frames and other such devices that are physically attached to the patient’s skull to other techniques that do not use any immobilization device. Early conformal therapy (1980s to 1990s) often incorporated a foam cradle device to help position the patient ⁶⁷; it is currently thought that more precision can be achieved without use of the cradle devices. In the end, each clinic should document the setup accuracy that is achieved with their chosen methods, for each clinical site, so that the planning and delivery process can take proper account of the expected systematic and random setup uncertainties. The use of in-room imaging systems (e.g., diagnostic and megavoltage CBCT) has provided more detailed information about setup accuracy, and makes it possible to improve setup accuracy and minimize margins using IGRT setup.

Computed Tomography, Magnetic Resonance Imaging, and Other Forms of Imaging

The development of x-ray CT in the 1970s and its application to radiotherapy planning ¹⁸ were absolutely crucial milestones in the development of conformal therapy techniques. Without the cross-sectional anatomic imaging provided by CT (and MRI), there was not enough anatomic knowledge about the tumor or normal anatomy to consider the use of highly conformal dose distributions. Certainly, once the detailed anatomic information provided by CT became available, it was clear that radiotherapy planning and treatment should make use of this new and detailed description of the patient to better spare normal tissues and more accurately deliver dose to the tumor. Conformal therapy is a logical response to the detailed information provided by CT.

Modern conformal therapy is always based on a 3D anatomic model of the patient, which is typically based on a CT scan of the involved region. Usually, a specific type of CT scan, the treatment planning scan, is obtained for use as the basis for treatment planning. Features of the treatment planning scan are listed in Box 15-1.

Box 15-1 Treatment Planning CT Scan Features

• Use a flat tabletop to mimic the treatment machine couch. Make use of couch coordinates or other documentation of exact positioning if possible.

• Use patient immobilization devices to position the patient in the same way as for treatment, including registration of the device with the couch top (if available).

• Scan extent should include completely any organs to be considered as organs at risk because planning typically requires delineation of entire organs to make use of biologic effect data. For example, treatment anywhere in the thorax would typically require scanning from the apex of the lung to below the diaphragm, so that the entire lung could be delineated.

• Modern CT scanners are efficient enough that 1- to 3-mm-thick slices can be obtained throughout the entire region of the scan, though use of thicker slices outside the target region is possible. This can amount to several hundred CT slices per study.

• Careful consideration must be given to the advantages and disadvantages of using contrast during the scans. Contrast can help identify organs but can also make accurate dose calculations difficult (because it distorts the apparent electron density of the tissue, affecting dose calculations).

• The treatment planning CT scan study is used for (1) target definition and delineation, (2) normal tissue delineation, (3) the overall anatomic model of the patient, (4) planning, (5) dose calculations based on electron density obtained from CT numbers, and (6) creation of digital reconstructed radiographs (DRRs) used to set up and verify patient position. Each of these uses may require some individualization of the CT scan protocol or use.

• The CT scan protocol should be defined for each clinical site, including slice thicknesses, scan extent (top and bottom), reconstruction window size, use of contrast, patient positioning and immobilization, patient breathing control or instructions, and so on.

Developments in CT and treatment delivery technology have made the consideration of motion during CT scanning (and radiotherapy treatment) an important research topic. 4DCT describes various techniques for obtaining CT data correlated with patient respiratory phase information so that the changes associated with respiration (or other motion) are displayed. For certain clinical sites (e.g., the lung, the breast), it is clear that consideration of respiratory (and, perhaps, cardiac) motion will be an important aspect of the initial imaging of the patient, so that an appropriate model (perhaps, a time-varying or 4D model) of the patient can be used for further planning and analysis. 4DCT,⁶⁸ respiratory gating,⁶⁹ active breathing control (ABC),^70,⁷¹ or other methods are often used for many treatment sites, though which combinations of techniques and methods are most efficient and appropriate is not yet clear.

CT provides anatomic and electron density information that is critical for most treatment planning, and it also provides a geometrically accurate base for planning. However, it provides only anatomic information, not the physiologic and functional information that should be very helpful for planning, and it provides only a limited amount of soft tissue contrast. MRI can provide complementary kinds of data, including excellent soft tissue contrast and different kinds of physiologic information. In addition, functional magnetic resonance imaging (fMRI) studies can provide some of the functional information that to this point has been unavailable. Other kinds of imaging also contain complementary or new information. PET and single-photon emission computed tomography (SPECT) provide functional and physiologic information and can be quite important in helping define target volumes and regions that should be included or excluded from the radiation fields. Which modalities, scans, tracers, and analysis methods should be used for specific features is well beyond the scope of this work. However, in order to quantitatively make use of any additional imaging modality for treatment planning, one should incorporate a number of important procedures into the imaging process, as listed in Box 15-2.

Box 15-2 Imaging for Treatment Planning (Not Including CT Imaging)

• Use a flat tabletop to mimic the radiotherapy Linac couch.

• Use patient immobilization devices to position the patient as you would for treatment.

• Ensure that the imaging protocol provides enough anatomic information to allow geometric registration of the new scan study with the base CT information.

• Imaging protocol parameters should be optimized to provide the anatomic, physiologic, or function data that are desired.

• The scan protocol should be defined for each clinical site and protocol, including slice thicknesses, scan extent (top and bottom), scan orientation (nonaxial slices are possible in most non-CT methods), reconstruction window size, use of contrast, patient positioning and immobilization, patient breathing control or instructions, and so on. For MRI, many additional scan technique parameters must be defined.

• CT scans should be performed in a manner that will allow accurate registration of other scan information (e.g., MRI, PET) with the CT scans.

• Motion management (consideration of how motion will be managed for CT scanning and for treatment) is an important part of the preparation for planning. It is appropriate to (1) control respiratory motion or (2) perform a 4D CT scan ⁶⁸ for patients who demonstrate any significant motion in the target region. It may be necessary to perform a quick scan to evaluate motion before deciding on the final planning scan protocol and methodology.

To use more than one imaging dataset for planning, the additional datasets will have to be registered geometrically to the original (base) CT dataset (as described later). It is important during the imaging process (1) to position and align the patient similarly for each of the imaging studies, as this makes the registration process more straightforward, and (2) to obtain all the information necessary so that the dataset registration and fusion process can be performed quickly and accurately.

Anatomy for Treatment Planning

Treatment planning is a computer simulation of the process of radiotherapy treatment, and it is based on creating a model of the patient inside the planning software, simulating radiation beams and the dose that those beams deliver to the patient. The definition of the virtual model of the patient, based on CT, MRI, and other types of imaging data, and how that anatomic model is used are crucial parts of the radiotherapy treatment planning process.

Abbreviation	Name	Description
GTV	Gross tumor volume	Volume of macroscopic tumor that is visualized on imaging studies
CTV	Clinical target volume	Volume that should be treated to a high dose, typically incorporating both the GTV and volumes that are assumed to be at risk due to microscopic spread of the disease
PTV	Planning target volume	Volume that should be treated in order to ensure that the CTV is always treated, including considerations of systematic and random daily setup errors and intertreatment and intratreatment motion

The gross target volume (GTV) is typically delineated by drawing the imaged tumor on each of the imaging studies that are available. CT is used often, but for many sites, MR and PET can be very useful. When multiple imaging studies are available, the GTV can be drawn on each study, and then, using dataset registration to geometrically align the different datasets (see Multiple Imaging Modalities: Dataset Registration and Fusion), one can transfer the different GTV contours onto a single dataset. How to combine the various GTVs defined is the subject of ongoing research; however, typically, one will combine or take the union of all the defined GTVs in order to make sure that no gross tumor is missed within the defined GTV.

The definition of the clinical target volume (CTV) is probably the most important thing that the physician does in the conformal therapy process, because the CTV defines the region that is supposed to be treated with the prescribed dose. The CTV typically combines the GTVs plus any volumes that may contain microscopic disease that has not been imaged. The CTV depends on knowledge of the patterns of disease spread and incorporates any other clinical knowledge of the disease or the specific risks for spread that apply to the individual patient. The CTV is usually created by combining two kinds of information: (1) often, an expansion of the GTV by some margin (0.5 to 1 cm, typically) is used to account for microscopic invasion, and (2) additional anatomic areas may be included in the CTV based on standard directions of spread for the particular tumor type. In the end, the goal is to outline all the areas that should receive the intended dose.

Whereas GTV and CTV definition are the job of the physician, definition of the planning target volume (PTV) is the responsibility of the physicist and treatment planner, as the goal of the PTV is to make sure that the CTV is adequately treated in the face of setup error, intertreatment and intratreatment motion, delineation errors, and other errors in the planning and delivery process. The definition of the PTV should be done with as much information as possible because the region between the CTV and PTV contours is all “normal tissue” and increasing the PTV margin will just cause more normal tissue to be irradiated.

Often, the PTV is designed by simply defining an isotropic margin (e.g., 1 cm), and the CTV is expanded by this margin to create the PTV (Fig. 15-4). This expansion should be performed in three dimensions because expansion of contours only in the axial plane will lead to PTVs that are not correct in the third dimension. If the uncertainties are not isotropic but are larger in one direction than in the others (e.g., due to respiration), then the margin to be applied should be anisotropic.

Figure 15-4 Noncoplanar IMRT plan for treating a brain tumor, with the ICRU-50 target volumes shown: gross tumor volume (white), clinical target volume (pink), and planning target volume (yellow).

There has been a great deal of work studying patient positioning, motion, and target volume delineation errors, and analysis of these issues has led to specific recommendations for the size of the margin (between the CTV and the PTV) that should be used for the PTV. As described in Consideration of Setup Error and Patient Motion, one reasonable method for deciding the PTV-CTV margin has been determined to be 2.5 × Σ + 0.7 × σ, where Σ is the standard deviation of the systematic error and σ is the standard deviation of the random errors for the population of patients treated in that particular site.⁷³ To apply this formula, it is important to have measured, for your institution and each clinical site, the two standard deviations. As can be seen from the formula, the systematic errors in the process, such as incorrect contouring or use of a nonrepresentative CT scan for target delineation, are much more important issues than random day-to-day setup errors.

Further discussion of other types of target volumes, including the internal target volume (ITV), is included later in the section on motion.

For an individual patient, there can be multiple sets of GTVs, CTVs, and PTVs because there is often a CTV and PTV that correspond to each individual GTV. In the head and neck, where often a number of different nodal CTVs need to be treated, it can be very important to develop an organized and clear naming convention for the various CTVs and PTVs.

Normal Tissues

Definition of normal tissues is also a critical task for conformal therapy because identifying the critical tissues will allow the treatment planner to avoid or at least minimize dose delivery to those normal tissues. The planning tools used to avoid these structures can be simple graphic tools such as the BEV display, which allows the planner to shape the radiation fields to avoid important structures, or it may involve detailed dosimetric and DVH analysis, as is often the case for IMRT planning.

In order to perform dose-volume histogram (DVH) or other dosimetric analysis, it is important that each organ to be analyzed be contoured completely because most current DVH data are characterized with respect to the whole organ’s volume (either absolute volume or as a percentage of the whole organ). This has several implications:

1 The CT scans or other studies for a given site must image all the relevant normal tissues completely as well as the tumor. For example, a lung tumor CT scan should always include images from the neck down to below the bottom of the diaphragm, so that the complete volume of the relevant normal tissues can be identified (e.g., lung, heart, esophagus, spinal cord).

2 Contouring of normal structures must be done consistently if the DVH information is to be useful. Whole organs should always be contoured. For tube-like structures (e.g., spinal cord, rectum, esophagus) that can often extend out of the region of the tumor, it is important to have a defined protocol for how the structure will be defined, so that the superior and inferior extents of the contours for the structure are specified (e.g., the rectum during prostate planning).

Multiple Imaging Modalities: Dataset Registration and Fusion

Although CT scans are the primary imaging modality used for radiotherapy planning, information from other types of imaging, particularly MRI and PET, can be very useful for identifying disease or better identifying functional or anatomic areas that should be spared. Target volumes and normal structures can be identified on these additional imaging datasets, and that information can be incorporated into the treatment plan along with the contours and data from the CT scans. As described before, a CT scan set is taken to be the geometric basis for the treatment planning because the CT data are of high resolution, are geometrically accurate, describe the electron density information needed for inhomogeneity corrections, and are quickly obtained.

Several issues need to be solved to make quantitative use of the additional imaging information.

• For each imaging procedure, the patient should be positioned within the imaging device using the same position and immobilization and localization devices as for treatment. However, there are often positioning differences that must be taken into account.

• Even if the patient positioning is perfect, the coordinate system used in the new imaging device is usually different than the coordinate system used by the original CT scan. Therefore a geometric registration of the new imaging information must be performed, to align the new images with the base imaging information. Otherwise, with different coordinates, one does not know how to map pixels in the new image dataset to the coordinates of the original image set.

• For many imaging modalities, for example, MRI, various kinds of distortion are possible. If the new image dataset has geometric distortions, then these must be corrected (or at least accounted for) before the imaging information can be transferred into the base coordinate system for use in planning.

• The resolution, slice thickness, and slice orientations of different imaging modalities can often be different than those of the original CT dataset, so any quantitative use of the new imaging data must also handle these kinds of differences.

To address these issues, the process of dataset (or image) registration is used to align the coordinates for the various imaging datasets so that information can be passed from one image set into a coordinate system to be used for treatment planning. The registration process finds the geometric transform between the new dataset’s coordinate system and the base coordinate system (typically, the CT scan). If one considers only rigid body registration, the transform can consist of x, y, and z translations, or both translations and rotations, and it can include scaling as well (although typically, the scale of each dataset is known accurately and should not be modified).

Handling of distorted image registration is an important current research topic. These efforts work to develop methods for mapping distortions from one system to another, so distortions due to imaging or to patient motion (e.g., respiration) can be taken into account. Many different mathematical methods have been employed, including thin plate splines, B-splines, demons, and others. However, the main current issue is that distortion is well handled for things that are imaged well, but there is no way to determine how to do the distortion mapping for tissues that are not well imaged. Most algorithms do not take into account anatomic constraints and sliding organs (e.g., the diaphragm and lungs during respiration). Much work remains here.

To determine the best registration transform, an optimization algorithm is applied to the problem. The optimization process consists of choosing the metric to be optimized (some metric that describes the quality of the registration) and choosing an optimization search algorithm that will perform the search over possible transforms so that the optimal one can be found. The metric can be something as simple as the sum of the squares of the distances between predicted and actual point locations, if it is possible to define point-based landmarks on both imaging studies, or it can be image-based metrics such as the correlation between gray scale values of two CT scans, or mutual information, that can be used to register different image studies (e.g., CT and MRI or PET). This is a rapidly developing area of research.

No matter what kind of registration algorithm is used, it is necessary to verify the registration and then to use the data from the various imaging studies. Verification typically consists of image-based or structure-based comparisons between the two datasets, with the goal of confirming that known structures from the two imaging studies accurately line up (Fig. 15-5). The quality of the registration depends on what parts of the images are most important clinically and must be reviewed by the planner/physician because at this point, no quantitative measure accurately takes into account all the clinical knowledge of the case. Once the registration is verified, then contours or 3D structure definitions from one dataset can be transferred into the base coordinate system for planning. This combination of data from multiple imaging sources is sometimes called image fusion.

Figure 15-5 Split-screen comparisons of CT (top) and MR (bottom) image data to confirm the accuracy of the geometric registration of CT and MR scans of the liver.

Motion, Setup, and Four-Dimensional Anatomy

So far, there has been little consideration of the facts that real patients breathe, move, are different from day to day, and change over time. Until recently, it was difficult to take such motions and localization differences into account within treatment planning, with the exception of defining appropriate PTV margins for the tumor. Fast helical CT scans, fast MRI, 4D-CT, and 4DCBCT imaging using the treatment machine have begun to provide detailed anatomic data as a function of time. These data have clearly demonstrated that a static anatomic description of the patient is not always appropriate and that treatment delivery schemes must also consider setup and motion effects if we are to achieve the optimal delivery of dose to the patient.

Several methods to handle motion and setup effects are in use or being investigated:

• As described in the section on Target Volume Definition and Margins, the standard way to handle motion and setup error is to determine the appropriate margin and then to expand the CTV by that margin to make a PTV that is the target for planning. If done correctly, the PTV ensures that the high-dose region always encompasses the CTV, even as the CTV moves around due to motion or setup error. The price, however, is that the larger the margin is, the more normal tissue is irradiated. Therefore, if the motion and setup error are taken more directly into account, this margin may be decreased, reducing the amount of normal tissue irradiated.

• Stereotactic treatments and treatments using daily setup correction (IGRT) try to significantly reduce the daily setup variations by carefully reproducing the position of the patient each day.

• With imaging with 4DCT, respiratory-correlated CBCT, and other such “4D” methods, it is now possible to visualize the motions of the tumor (and normal tissue) with the CT data, and to create structure outlines that better approximate the clinical targets and structures to be avoided. However, this technology is currently undergoing very rapid developments.

• With use of active breathing control (ABC), it is possible to control the breathing of the patient with the goal of turning on the treatment beam only when the patient is in the correct breathing state. This method can help minimize respiration motion and allow smaller PTV margins.

• It is also possible to gate the radiation beam, turning it on only when the target structure is in the correct location, based either on external markers or, potentially, on internal markers.

• Various radiographic and radiofrequency transponder markers are also used to help identify the position of the target. Transponders that identify their position 10 times per second allow real time tracking of the target and can help to dramatically reduce the PTV margin needed for treatment.

Over the next several years, a great deal of technical progress can be expected in this area.

Use of Functional Imaging Modalities

Standard CT and MRI techniques provide basically anatomic information, although some simple physiologic information is obtained from MR. However, PET, SPECT, and various functional MRI (fMRI) imaging sequences can all lead to much more detailed knowledge of the physiologic and functional state of various tissues or organs. The details of these many different types of studies are well beyond the scope of this chapter. However, several features of any such use are similar: the patient must be positioned as close to the treatment position as possible, the imaging study must be carefully registered geometrically with the rest of the treatment planning imaging information, and the actual meaning of the imaging information should be known. Although many different imaging scans show some response in the image, it takes detailed clinical knowledge and trials to clearly define that information, which is crucial before the information is used clinically.

Plan and Beam Definition

After the anatomic model of the patient has been established, the next major step in the planning process is to use the planning system to create a set of beams to be used for planning. This collection of beams, usually known as a “plan,” can be created using standard protocols (“treat all prostates with a four-field box of conformal fields”) or designed based on the specific anatomy of the case. Basic decisions on beam technique are typically made very early in the planning process, using experience and/or site-specific protocols. Typically, these decisions include picking the energy and number of beams, the basic orientations for the beams, and the type of beam shaping or intensity modulation to be used.

Beam Technique (Energy, Direction, and Type)

The first things to be decided as a treatment plan is generated are the number of beams to be used, their energy (and modality), and their direction. These choices are all interrelated, so typically this decision is based on standard experience or protocols. Although it is hard to summarize all of the useful ways to make this decision, there are a few standard rules that apply to most conformal planning.

• Avoid opposed beams. Simple nonconformal planning often used opposed beams (AP-PA and opposed laterals or obliques) to create a relatively uniform region of high dosage extending through the patient. For conformal planning, one typically wants to avoid pairs of opposed beams, because one wants to conform the dose to the target. For IMRT plans, this is even more important because opposed beams can cause degenerate solutions for the optimization search, leading to the optimization bouncing between two fairly equivalent solutions for opposing beamlets.

• Typically, lower energies are used for relatively superficial targets or for beams that traverse less normal tissue on the way to the target, whereas beams that must penetrate a greater volume of tissue are typically of higher energy. Changing energies for individual beams often makes a relatively small difference in the dose distribution achieved, and this possible improvement should be balanced against the additional complexity of the plan if multiple photon energies are used.

• For sites in which one is attempting to make sure the dose in the buildup region is relatively high, lower energies are usually used because there is less skin sparing for these beams.

• The most conformal plans can be achieved by ensuring that there are beams arranged around the target in all three dimensions, so that the plan includes some nonaxial beams.

• There is a great deal of debate (still) over the best energy beams to use for plans that involve large inhomogeneities such as the lung.

• As one creates more complex IMRT plans that have many degrees of freedom available to be optimized, it is possible to achieve very conformal plans. Typically, these plans include an odd number of fields, although the goal is just to make sure there are no opposing fields. There has been much discussion saying that IMRT plans can all be accomplished with 6-MV photons and that higher energies are not necessary, but there is little hard evidence to demonstrate that such a conclusion is correct. In addition, it is often said that IMRT plans can be achieved with only axial beams, but the use of noncoplanar beam arrangements will still result in at least slightly better plans (typically reducing somewhat the normal tissue dose for the same target conformality).

Shaping with Blocks and Multileaf Collimator

Although beam directions are important, shaping the radiation field to conform to the shape of the target volume is one of the crucial and defining concepts for “conformal therapy.” The shaping can be accomplished equally well by focused blocks or with an MLC, as illustrated in Figure 15-6. The conformal shaping of focused blocks is in fact “more conformal” than the jagged shape created by an MLC, although the MLC has a number of other advantages that have led to its popularity.

Figure 15-6 Shaped blocks and multileaf collimators can be used to shape the beam conformally.

The routine use of conformally shaped fields designed during treatment planning depends in large part on the availability of the BEV display in the planning system because this view of the target shows the projection of the shape of the targets from the point of view of the radiation beam, which is what is needed to design field shaping. The simplest method used to conformally shape the fields (with either blocks or MLC) is to create a uniform geometric margin around the projection of the targets in the BEV and to set the shape to that margin, as shown in Figure 15-7. This method, the basis of the simplest type of conformal therapy, is sometimes called geometric conformation, or beam’s-eye view, targeting. Shaping a block to a given contour is easy, but with an MLC it is more complicated³⁸: The most commonly used method is the so-called equal area method (Fig. 15-7).

Figure 15-7 When using geometric conformal shaping for an MLC, the most common method is to create a contour at the desired margin (red contour) and then push the MLC leaves up to that contour so the areas overlapped and underlapped by each leaf are equal.

Using a uniform geometric margin for the field shaping does not lead to the most conformal dose distribution. To truly conform the dose distribution to the target, one must optimize the shaping of each of the beams so that the dose distribution is conformal. Figure 15-8 demonstrates the types of differences that occur when beam shapes are designed with a uniform margin and when the shapes are optimized to conform the dose to the target. Beam directions, the penumbra, and how the beams cross fire on the target affect the margins required for individual beam shapes.

Figure 15-8 Beam’s-eye view of conformal fields: uniform geometric margin (left) and conformal dose distribution (right). Using a uniform geometric margin typically does not achieve a dose distribution that is as conformal as possible. For this three-field pancreas treatment plan, the beam on the right had its shape optimized to make the dose distribution more conformal to the target.

Collimator angle is one more thing that can directly affect the conformality of a plan, mainly when an MLC is used. The leaves from an MLC move in only one direction, so to minimize the “stair-step” or jagged edges caused by MLC leaves when shaped to an angled contour, one may use collimator rotation to cause the MLC leaves to best fit the shape of the target or normal tissue. By minimizing the jaggedness of the MLC edges, one can decrease the amount of penumbra in that particular region of the beam, thereby allowing the beam to do a better job making a sharp penumbra between a target and normal tissue. For example, for beams trying to make a sharp dose gradient between the prostate and the rectum, rotating the collimator to parallel the edge of the rectum can help make the edge sharper.

It is possible to create intensity-modulated beams using a limited number of MLC-shaped “segments” all from the same gantry angle. This “segmental” IMRT can be created using the normal interactive planning paradigm (“forward planning”)⁷⁴ or the limited number of segments can be created by an inverse planning paradigm (see, e.g., “direct aperture optimization”).⁷⁵ The use of a few segments to improve target homogeneity (e.g., in the treatment of breast cancer with tangential fields) is a logical extension of the concept of wedged tangents when 3D planning is available.

Other Beam Technique Decisions

There are numerous other decisions to be made when creating the plan.

• One must choose the intensity of each beam, typically called the beam weight. How this decision is made depends on the specific beam normalization methods used within the planning system. Because the beam weight or intensity is directly related to the number of monitor units (MUs) that will be delivered from each beam, it is essential that translation of beam weights into MUs be understood and carefully handled, because errors in this step deliver the wrong dose to the entire field.

• For certain beam combinations, use of wedges in the field can help achieve a more uniform dose in the target volume. Standard two-dimensional (2D) planning makes routine use of pairs of fields known as “wedged pairs,” where the wedges help compensate for the fields entering the patient at 90 degrees (or some other angle) to each other. For conformal planning in three dimensions, there is a broader range of combinations of fields for which wedges are useful. For example, a 3D cone has some number of fields placed around a cone, and each of the fields usually requires a wedge to keep the target dose uniformity. This beam distribution is the 3D analog of the wedged pair.

• When the target is near the surface of the patient, it is often difficult to achieve the desired dose coverage of the target due to low doses in the entrance region of megavoltage photon beams. In this case, some amount of bolus can be added to the skin of the patient, for one or more beams, so that the photon beams can transit more material, therefore decreasing the skin sparing. This bolus description can be designed using the planning system, as long as care is taken in creating the physical bolus so that it actually matches the description that was put into the planning system.

• Most conformal treatment plans are isocentric, with the isocenter of the machine placed within the center of the target, and the beams rotating around the isocenter in the target. Other arrangements are possible, depending on the geometry of the normal anatomy and targets. For some situations, it is also possible to create pseudo-isocentric beams, which use 3D table motions to move the patient away from the head of the machine along the beam central axis, so that the plan is apparently isocentric but with a larger source-to-isocenter distance than the machine has. This technique can be used to create a larger range of table and gantry angles that can be used for treatment without the machine’s head getting too close to the patient.⁷⁶

Intensity-Modulated Radiation Therapy

Soon after conformal therapy began to be used clinically in the late 1980s, Brahme,⁵⁴ Bortfeld,⁵⁵ and others introduced the idea of modulating the intensity across each radiation beam, assisted by computer-based optimization algorithms to help determine the intensities required of the different parts of the beam. IMRT is now commonly used to create highly conformal treatment plans.

Intensity-modulated fields can be achieved in a number of different ways. There is a continuum of situations ranging from a flat field to multiple shaped segments to a beamlet-type description created by either a series of SMLC segments or a dynamic DMLC sequence. For plan optimization strategies, there is a similar range from simple (forward) iterative planning to optimized (inverse) planning that is driven completely by a mathematical cost function. Typically, the most complex intensity distribution (IMRT) is generated with inverse planning, but it is also possible to perform optimization for flat field conformal therapy.⁷⁷

Beam design for IMRT depends on the type of planning to be used (forward or inverse) and the photon energies and beam arrangement expected and should consider the tradeoff between the planning goals chosen and the complexity of the plan that will be allowed. If simple multiple segments are planned, then the individual segments may be created by the planner, whereas for more complex IMRT planned with “beamlets,” the beamlet size (1 × 1 cm, 0.5 × 0.5 cm, etc.) to be used for planning will be chosen by the planner.

Dose Calculations

Once the initial treatment plan is designed, the next step is typically to perform a dose calculation, so that the planner and physician can evaluate the dose distribution expected from the plan. Currently, because biologic effects are not well documented and understood in general, the physical dose distribution is the main parameter that is used (1) to choose between plans, (2) to choose what dose to deliver to the patient, and (3) to evaluate the quality of various plans proposed by the planner.

Three-Dimensional Dose Calculations

Treatment planning dose calculations have been performed on one (or more) 2D slices (or contours) of the patient since the 1940s. Often, these calculations were performed by hand, from table or chart look-ups, using a single traced contour of the external shape of the patient on a single axial slice of the patient at the center of the treatment fields. Even if performed on a number of slices, these dose distributions were in principle 2D and did not give a complete description of the dose to be delivered to the patient.

Continuing increases in computer capabilities and improved treatment planning and dose calculation algorithm developments resulted in availability of 3D dose calculations in the 3D planning systems that became available in the late 1980s. Since that time, most planning systems have implemented 3D calculation algorithms that (1) calculate the dose throughout a 3D volume of points, (2) calculate the dose with algorithms that take 3D scatter into account, (3) have algorithms that take the 3D effects of inhomogeneities into account, (4) use a 3D description of the anatomy, (5) fully take into account 3D beam divergence, and so on. Because the calculation of dose with high resolution using an accurate and realistic dose calculation algorithm still takes a significant amount of time, every dose calculation algorithm and implementation makes tradeoff choices between accuracy, speed, computer resources needed, resolution, features and effects that are correctly modeled, and other factors. Determination of the appropriate mix of approximation, compromise, and robustness needed for particular kinds of clinical planning dose calculations is an important responsibility of the radiation oncology physicist.

To accurately perform planning for conformal therapy, accurate and 3D dose calculations must be performed. An accurate 3D anatomic model is required; the dose must be determined throughout the volume encompassing the targets and the critical normal tissues; and the calculations should be done at high resolution if one wants to know how conformal the plan actually is. It is critical that all limitations in the calculations be understood and the effects of those limitations should be considered in any clinical decisions that are based on the results of the calculations.

Algorithms

Many different types of calculation algorithms have been developed for photon and electron beams, and new improvements or implementations are continually becoming available. It is beyond the scope of this text to describe any algorithms in detail. Algorithms ranging from simple table look-ups based on measured data to Monte Carlo simulations that require many hours of central processing unit (CPU) time of the fastest computers all have their place and are the appropriate choice of algorithm for one particular situation or another. Table 15-2 summarizes some of the advantages and disadvantages of each class of photon algorithm. If there is a choice of algorithms for conformal planning, the choice should be made with careful consideration of the potential limitations of the chosen algorithm, and the radiation oncology physicist should carefully commission the algorithm for clinical use by comparison with appropriately measured data for the local machines in order to demonstrate the adequacy and limitations of the algorithm for clinical use.⁸⁸

TABLE 15-2 Advantages and Disadvantages of Each Class of Dose Calculation Algorithms for Photon Therapy

Other Dose Calculation Issues: Grids, Resolution, Inhomogeneities, and Other Issues

Aside from the inherent accuracy of the dose calculation algorithm that is used, many other user-controlled factors affect the final accuracy of the doses predicted by the calculations that are performed for a plan. Unfortunately, it is not possible to address all these issues with a few simple guidelines.

Every step of the treatment planning, dose calculation, and plan evaluation process involves decisions about how much time, effort, and precision to spend defining or reviewing each aspect of the planning, and the accuracy of the final product depends on all those individual decisions. If one chooses to calculate the dose on a grid of spacing of 0.5 × 0.5 × 0.5 cm, then it will be hard to evaluate the dose 2 mm from the target with a significant degree of confidence, given that dose gradients at the edge of a beam can be approximately 10% per millimeter. Highly conformal therapy will require that dose calculations be performed with high-resolution grids (perhaps, 2 to 3 mm), leading to very large numbers of points that need to be calculated and longer calculation times. Similarly, if truly conformal therapy is going to be performed in lung tumors, where inhomogeneities are significant, an advanced calculation algorithm that can accurately predict the dose in inhomogeneities should be used if one wants to understand the real dosimetric situation in the patient. For IMRT plans, there are further dose calculation-related issues (discussed later).

Plan Evaluation Tools

After the dose distribution from the plan has been calculated, the next step in the process is to evaluate the plan and the predicted dose distribution. Typically, the dose distribution is evaluated by looking at isotope curves on individual cuts through the plan, looking at isotope surfaces (3D displays of the isosurfaces), and then using DVH analysis of the dose delivered to individual organs. If the capability and appropriate data are available, it is also possible to use biologic modeling results such as normal tissue complication probability (NTCP), tumor control probability (TCP), and the equivalent uniform dose (EUD) to help with the plan evaluation.

Dose Display

The most commonly used type of display of the dose distribution for a plan consists of displaying contour lines of constant dose, or “isodose lines,” on top of the anatomic information that was used for the plan. These kinds of displays have been used for many years—first with only the contours of the patient (obtained by hand measurement using solder wire or other techniques) and then later displaying the isodose lines on top of the CT scan. For conformal planning, not only axial cuts should be used for the isodose lines but also coronal and sagittal reformatted CT images, because display of isodose lines on multiple orthogonal cuts can give the planner a more 3D sense of the coverage of the target volume. 3D graphics techniques can be used to put the images and dose lines in 3D perspective, as shown in Figure 15-9. A variation of the isodose line display is the colorwash display, in which the dose level calculated for each pixel of the image is used to assign a color value (see Fig. 15-9C).

Figure 15-9 Dose display for a conformal brain plan. A, Axial CT with isodose lines. B, Axial MRI with isodose lines. C, Three-dimensional axonometric view with dose colorwash. D, Target volume (red contour) and isodose surface (yellow contour).

Any image that is part of the anatomic model of the patient should, in principle, be usable for dose calculation and dose display (though this is not always permitted by some TPS systems). If MRI, PET, or other image datasets have been registered with the basic anatomy, then the PET or MR images can also be the backdrops for the dose display. By comparing the location of the isodose lines with the contours of the targets and critical normal tissues, the planner and physician can evaluate the quality of the dose distribution that is obtained for this particular beam arrangement and decide whether the plan is adequate or whether further modification of the plan is necessary.

The goal of conformal therapy is typically to conform the shape of the high-dose region to the target in three dimensions, so display of the 3D dose distribution shape may be a help when evaluating the conformality of the plan. The 3D analog of an isodose line is called an isodose surface. Also, isodose surfaces, sometimes called “dose clouds,” are typically displayed in a 3D perspective graphic image (see Fig. 15-9D).

The dose displayed in isodose curves or surfaces, or any other dose display, can be shown in a number of ways. The most common mode used for conformal therapy is the relative dose distribution, where all the dose display is normalized to the dose value at the isocenter or center of the target. With this kind of relative dose normalization, typical conformal therapy plans would have the 95% isodose surface surrounding the target, with the shape of the 95% surface (or isodose lines) conforming to the target. One could, however, equally well display the dose in absolute terms so that the high-dose region demonstrated the desired total dose for the plan (e.g., 60 Gy) or the desired dose per fraction (e.g., 2 Gy/fraction). It is of course essential that the output of the plan that is used for treatment preparation should be carefully understood and documented, so that there is no confusion between the different ways the dose can be displayed.

Dose-Volume Histograms

Review and evaluation of the dose throughout the patient in three dimensions can be complex and time-consuming processes, and it is also difficult to give specific guidelines for normal tissue or tumor responses with respect to that complex data. It has become standard to evaluate the dose received by the target volumes and normal tissues using DVHs. To form a DVH for any 3D object, one looks at the dose value for each voxel in the object and forms a histogram, counting the number of voxels that receive each different dose level. Because the volume of each voxel is known, the volume of the organ receiving each dose level is known. See the review by Kessler ⁸⁹ for more detail.

Both the volume (vertical) and dose (horizontal) axes can be displayed in absolute terms (as cubic centimeters [cc] or Gray [Gy]) or in relative terms (% volume or % dose), depending on how the planner wants to analyze the results. DVHs are displayed in three different forms: direct, cumulative, and differential.

• The direct histogram is the generic form of any histogram, displaying the volume of the organ that receives dose within each dose bin (1% or 0.5 to 1 Gy is a typical dose bin width). The direct DVH is most useful for display of the dose-to-target volumes, because one can easily visualize the minimum dose, the maximum dose, and the dose most representative of the dose to the entire target volume (Fig. 15-10B).

• The DVH display most commonly used in radiotherapy is the cumulative DVH, in that the volumes receiving at least a given dose value are plotted. The cumulative DVH integrates the direct histogram, so it always begins at 100% (100% of the organ receives at least 0 dose), and ends at the maximum dose (see Fig. 15-10A). Figure 15-10A illustrates a desirable cumulative DVH for a target (PTV, which has a uniform target dose with no underdosing or overdosing of the target). Figure 15-10A also shows two normal tissue DVHs: normal brain, which has some volume of the organ receiving a high dose, and chiasm, with a smaller volume of organ receiving high dose, even though the mean dose of the two DVHs is about the same. Whether a normal tissue DVH shaped like the example normal brain DVH is better than one shaped like the example chiasm DVH is only known once reliable clinical data are obtained for the two situations.

• The third type of DVH used for radiotherapy analysis is the differential DVH. This type is necessary in order to appropriately compare DVHs formed with different dose bin sizes, because the volume contained in any dose bin changes as the dose bin size changes. The differential DVH plots (1/D_bin) _* DV/DD, so different differential DVHs can be compared even if their dose bin sizes are different.

Figure 15-10 DVH displays for a brain treatment plan. A, Cumulative DVHs. B, Direct DVHs. LON, left optic nerve; RON, right optic nerve.

DVH analysis is a very important part of conformal therapy planning, but DVHs of target and normal structures do not tell the entire story. The DVH of an organ summarizes the dose to that organ, but it does not give any information about the geometric distribution of the different doses within the organ. If the DVH for a PTV shows cold and hot spots, it is not possible to tell if either the hot spots or the cold spots are in the center of the PTV or along the edges. Whenever the location of some dosimetric feature may make a difference, it is important to review the dose distribution using dose display tools.

Minimum, Maximum, and Other Quantitative Metrics

DVHs are only one kind of dosimetric analysis that can be used for conformal planning, and many other kinds of analysis are possible. For example, simple dosimetric metrics are often used to help make planning decisions. The maximum dose allowed within a target volume, or inside critical organs, is often specified in conformal therapy protocols. Particularly for IMRT conformal plans, the maximum dose may be a fairly difficult metric to use clinically because it may depend strongly on the dose calculation grid spacing used and on details of the dose calculation algorithm implementation. A volume-based maximum dose criterion (e.g., the maximum dose to 1 cc of the object) is typically less sensitive to algorithmic details. Similarly, the minimum dose to the PTV is subject to the same algorithm-related limitations and may be of limited value, particularly for IMRT planning. Another common dose metric is the mean dose to a structure, often used to describe target doses and for certain volume-effect organs such as the lung and liver, where the normal tissue complication probability (NTCP) generally scales with the mean dose to the normal tissue.⁹⁰

Use of Equivalent Uniform Dose (EUD) and “Biologic Models”

To this point, all the treatment plan analysis has been based directly on the dose distribution. In reality, of course, it is the biologic effects that are important, especially the probability of controlling the tumor and of causing acute or late complications in normal tissues. Analyzing the dose distribution is in many ways a surrogate for the analysis that is really needed, the biologic effect analysis. It is expected that there will be major and continuing improvements in biologic effect knowledge and modeling over the coming years, and as these improvements happen, they should be integrated into use for treatment planning. At the moment, however, the only biologically related parameters that are commonly used in treatment planning are the NTCP, the TCP, and the EUD. This section discusses each briefly.

The most commonly used “biologically related” parameter used in treatment planning is the NTCP. The term NTCP actually has several meanings: (1) The NTCP is a probability that can be determined clinically for an organ; (2) there are NTCP models that attempt to model how the probability of a particular complication changes as a function of dose, volume of the organ irradiated, and potentially other factors; and (3) the NTCP is used as the value of the complication probability that has been determined by an NTCP model for a particular situation. It is important to make sure that it is clear how to differentiate between the real clinical NTCP, an NTCP model, or the particular expected value of the NTCP for a situation.

There are many kinds of models that have been developed for the NTCP, and these models have been applied to specific complications for many different organs. Discussion of these different models is beyond the scope of this work, and here we briefly describe only the most well known model, the Lyman NTCP model. John Lyman developed the three-parameter “Lyman” model early in the 1980s.^91,⁹² This power law model is a phenomenologic model that characterizes complications using three parameters: TD₅₀ for uniform irradiation, the slope of the dose sensitivity (n), and the volume parameter (m):

where

To use this model for a real clinical DVH, one typically uses the Kutcher-Burman DVH reduction method ⁹³ to convert the clinical DVH curve into a DVH with a single dose and volume, which are then used inside Equation 1. Together, these techniques are typically called the LKB (Lyman-Kutcher-Burman) model.

When developing this model, Lyman made no claim that it was a real biologic model; he simply developed the simplest model that agreed with some of the most basic behavior of NTCPs, at least as characterized at that time. The model has been used extensively, either as a way to characterize complication data or for plan evaluation and comparison. A very important starting point for study of NTCPs of various organs was published by Emami,⁹⁴ who summarized a tabulation of then current knowledge of clinical complication expectations (based on a physician working group) using Lyman NTCP model parameters. Recently, complication data and parameterizations for many clinical sites have been reviewed by site-specific groups of experts in order to provide updated complication modeling information.⁹⁵ The Lyman model has proven to be a useful way to parameterize clinical NTCP data and has been used as part of dose escalation studies based on treating patients with a specific isocomplication level (for liver⁹⁶ and lung⁹⁷).

One of the most important things that can be done to improve conformal therapy planning is to perform clinical studies that (1) document, for each normal organ, the distribution of doses and volumes of various organs that are irradiated, (2) include careful patient follow-up, and (3) are analyzed to find the dose-volume-complication relationship for each organ. The dose-volume-complication relationship for each organ (and each complication) is unique and must be determined clinically. Once these results are known for all normal tissues, we will know better how to optimize treatment plans.

Just as the dose-volume-effect relationship is important to know for normal tissues, it is also very important to know for the tumor. The tumor control probability (TCP) is the subject of clinical studies and modeling and is a way of comparing expected tumor responses with planned dose distributions. A number of different models exist, including the Niemierko-Goitein ⁹⁸ and Nahum⁹⁹ TCP models. Most of these models use various basic assumptions about tumor cell density and distribution, the statistical interactions between dose and tumor cell survival, and the incorporation of population-based statistics for tumor heterogeneity, and may also consider effects that depend on tumor stage, hypoxia, and other issues. Tumor cell biology and predicting local tumor control are very complicated subjects that are well beyond the abilities of current models. TCP modeling is used in a reasonably limited way because it is known that the predictions are of limited accuracy.

The equivalent uniform dose (EUD) is often regarded (at least partially) as a biologically related parameter that is often used for treatment plan evaluation and optimization. The EUD was originally described by Niemierko ¹⁰⁰ by using a biologically and statistically influenced method to derive a dose parameter that was related to the effects one would expect on the tumor. The EUD has been generalized further^101,¹⁰² to be useful for normal tissues and targets. Because the EUD is in principle a generalized mean of the dose distribution according to a specific weighting method, one may use the parameters to represent the sensitivity of the tumor to underdosing in a small region or the effect of hot spots in the tumor. Because the concept of the EUD is relatively new, numerous publications and analyses continue to explore and expand the characterization of tumor and normal tissue responses using various EUD representations.

Forward Planning

Since computerized treatment planning began in the late 1950s until the last decade, nearly all planning has made use of the basic paradigm that is currently called “forward” planning, using the following process:

1 Define the anatomy. Use contours on multiple axial cuts through the patient (e.g., CT scans or manual contours) to define the external surface, targets, and important normal structures.

2 Create the plan. Create a beam arrangement, choosing beam energies, locations, directions and shapes, weights, and addition modifiers if necessary (e.g., wedges, compensators).

3 Calculate the dose. Calculate the dose distribution that results from that beam arrangement, on 2D slices or through the full 3D volume.

4 Evaluate. Evaluate the dose distribution, usually by looking at isodose line displays relative to the anatomy and (in recent years) by calculating DVHs for the normal tissues and targets.

5 Optimize the plan iteratively. If the plan is not adequate, or if the planner wishes to try to improve the plan, then the planner will iteratively modify the plan technique in an attempt to improve the plan result. Each iteration involves modifying the plan, performing a new dose calculation, and reevaluating the plan.

6 Complete the plan. Once the plan is judged acceptable (or complete) by the planner and physician, the iterative optimization stops, and the plan is prepared for use. This involves calculation of MU for treatment, calculation of digital reconstructed radiographs (DRRs) to be used for patient localization, and transfer of the plan to the treatment delivery system.

Iterative forward planning is driven by the treatment planner and usually relies a great deal on the visual dosimetric evaluation performed by the planner or physician.

Setting Planning Goals

Creation of a high-quality conformal plan using forward planning depends crucially on the planning goals set by the physician. A typical set of conformal therapy planning goals may include the following:

• Usually, the plan is normalized to the dose at the isocenter of the plan.

• The plan will typically make the 95% isodose line (for example) conform to the PTV (target volume) shape, so the minimum dose to the PTV will be 95% of the plan normalization point dose.

• The dose uniformity inside the PTV should be specified (e.g., ±5%), which will determine the maximum dose allowed in the PTV (in this example, 105%).

• Keep the maximum dose to the spinal cord less than some limit (typically, 45 to 50 Gy).

• Minimize the dose to all normal tissues.

• Prescribe the target dose—either prescribe the dose to the plan normalization point and accept ±5% uniformity throughout the target, or prescribe the minimum dose to the target volume, and accept that the dose in the target may be as much as 10% (for example) higher.

• If normal tissue responses are known, specify reasonable normal tissue limits for the planner. For example, one can specify that the mean dose to the normal lung stay below some given value, because it is known that mean dose can be used to predict the chances that pneumonitis will develop.

The physician should also specify any other constraints or evaluation expectations that will be used for final plan evaluation. One of the primary causes of additional planning iterations and frustrated planners is a physician who chooses to use some plan evaluation criterion or rule that was not described to the planner as a goal or expectation.

Beam Arrangements

Beam arrangements useful for conformal therapy planning are site specific and depend on the geometry of the target volumes and normal tissues. Virtually all conformal plans have at least three different beams, and seven, nine, or many more beam directions are sometimes used (e.g., conformal stereotactic radiosurgery). Plans with more than seven beam directions are relatively uncommon when forward conformal planning is used, due simply to the complexity involved in designing and determining the correct beam orientations, shapes, and weights interactively when there are so many different beams to consider. Nonopposing beam directions are very useful, as are beams that separate the target from the most critical normal tissues (increasing the distance between the target and normal tissues when viewed in BEV display). It is useful to have beams not in the axial plane, as they can help improve the conformality of the plan in all three dimensions.

Iterative Planning

Shaping beams is a very important part of the iterative planning process. Typically, a BEV margin of 6 to 7 mm is a good starting point for conformal beam shaping, but the shape of each field will then need to be modified to improve the conformality of the calculated dose distribution, as discussed earlier. This is the part of the iterative forward planning process that is most time-consuming, potentially, because in principle one needs to optimize the location of each leaf of the MLC for each beam, and there often are more than 100 involved MLC leaves in each field. The changes in shape that are needed are somewhat predictable, though, because larger MLC margins are needed in regions where there are no other beams (e.g., the superior and inferior aspects of the PTV for a plan with axial-only fields), whereas the margins can be much smaller if there are other beams that are transiting from an orthogonal direction. In the end, however, many iterations of the shapes may be necessary to get reasonably close to the optimal shaping. This is the part of forward planning that is often time-limited by clinical needs, and it could potentially be significantly improved by the application of computerized optimization methods to perform this shaping automatically.⁷⁷ Beam weights, use of wedges, and collimator angle rotation followed by revised beam shaping can also be part of the iterative planning process.

Plan Evaluation for Forward Planning

Evaluations of plans resulting from forward planning are typically based on isodose line display and DVH plots for target (PTV) and normal tissues. Typically, the physician evaluates the plan using these tools and decides if the plan is adequate for clinical use, as well as whether the plan has the potential to be improved through plan technique changes, or by the physician modifying the dosimetric planning goals that were specified initially for the planning. In forward planning, most clinical tradeoff decisions are made in a qualitative way by inspection of the displayed dose distribution and the DVHs.

Inverse Planning

Rather than trying plans and seeing what kind of dose distribution can be achieved, as in forward planning, the basic concept of inverse planning is to decide up front what the dose distribution should look like and to then “invert” the problem to solve for the beams (and beam intensities) that will give the desired doses. This inverse problem, which is the inverse of the CT back-projection process, would be wonderful except for two small problems: (1) Because there are no negative radiation intensities, it is not possible to invert the problem, and (2) planners probably do not know what the best “achievable” dose distribution would be to use as a goal. Planners do know that the goal of full dose to the target and zero dose to the normal tissues is not possible.

To get around these problems, inverse planning makes use of computerized optimization techniques to search for the best plan among all possible candidate plans, using an objective or cost function to drive the optimization toward the plan that is “optimal.” This kind of inverse planning or plan optimization process could in principle be performed for any kind of radiotherapy plan. In practice, however, the plan needs to have many adjustable (optimizable) parameters so there is enough flexibility in the plan to allow the optimization search to find a good solution to the planning problem. Therefore, up to this point, most inverse planning and optimization efforts have been applied to IMRT plans created with many separate beamlets or intensity “bixels” per beam, so there are a large number of degrees of freedom in the plan that make it possible for the optimization process to work effectively.

A typical inverse planning process can be summarized as follows:

1 Define the anatomy. The planner and physician must define (contour) all the target and normal tissue structures of interest. Whereas in forward planning one is not “forced” to delineate all important structures, in inverse planning one must define all anatomic objects that are to be considered by the optimization. One must also define how calculation points are to be distributed throughout the involved objects.

2 Create the plan. Define the beam energies and directions, determine the largest extent of the beam shape, and divide the beams into beamlets (intensity bixels).

3 Calculate the dose per beamlet. For each relevant dose point, calculate the dose per unit fluence from each beamlet. This “D_ij,” or “influence,” matrix tells how much dose each point receives as the intensity in an individual beamlet is changed by the optimization search algorithm.

4 Prepare the optimization algorithm parameters. One must define the cost function (and possibly the optimization search algorithm if more than one is available) to be used for the inverse plan. The cost (or objective) function definition is crucial, as it will be used to direct the optimization search and to drive the solution toward the desired dose distribution. The cost function is defined individually, at least for each planning protocol, and perhaps for each patient, and is likely to be the most crucial decision involved in determining what kind of plan will result from the optimization process.

5 Iterative plan optimization. Inverse planning uses the optimization search algorithm, driven by the evaluation of the chosen cost function, to perform iterative plan optimization, with the search algorithm determining how the plans are modified, and the cost function serving as the judge of plan quality. In most current inverse planning, only the beamlet intensities are varied, so to recalculate the change in the plan dose distribution and cost function, one only needs to re-sum the dose contributions from the changed beamlets, a calculation that can be performed quickly. The optimization search will continue until predefined stopping criteria are achieved.

6 Plan evaluation and reoptimization. Once the optimization process is complete, the physician and planner must evaluate the plan, dose distribution, cost function results and metrics, and any other relevant information, to decide if the plan is acceptable. If it is not, then the plan optimization process must be rerun; however, it is necessary to make changes in the cost function to drive the plan toward a different result, because running the optimization again using the same cost function should result in approximately the same plan result as the first time.

7 Plan completion. Once the plan is judged acceptable (or complete) by the planner and physician, the plan is prepared for use. For a beamlet-optimized IMRT plan, this completion step includes leaf sequencing, which is the preparation of the MLC leaf trajectories (DMLC delivery) or MLC segment shapes (SMLC delivery) that will create the desired beamlet distribution. Leaf sequencing issues are described in more detail later. Other preparation steps include calculation of MU for treatment (usually included in leaf sequencing for IMRT), calculation of DRRs to be used for patient localization, and transfer of the plan to the treatment delivery system.

Comparison of this process for inverse planning with the previously described forward planning process (see Forward Planning) shows that only a couple of steps are really different; most of the planning process is approximately the same. However, for inverse planning, most interactions aimed at improving the plan must be made by modifying the cost function used for the optimization, a much more indirect type of control on the plan parameters than one has when modifying a beam shape while doing forward planning.

Goals for Inverse Planning

As with forward planning, it is critical for the physician and planner to decide the overall goals for the inverse plan and to prioritize the various clinical issues. Goals for inverse planned IMRT may include the following:

• The dose to the target volumes should be specified: typically, this is the minimum dose, the mean dose, or the dose to a specific point inside the targets.

• Often, the dose distribution is defined in absolute terms (i.e., a plan will specify either the total dose [for the course of treatment, e.g., 60 Gy] or the dose/fraction [e.g., 2 Gy/fraction]). Relative dose can also be used, but given the use of IMRT, the dose to any defined reference point can be variable, so the relative dose mode of treatment planning is somewhat less useful than it is for flat field conformal therapy. In addition, if one wants to use the biologic model or other such evaluation or analysis tools, then the absolute dose mode of planning is much more useful than the relative dose mode.

• The dose uniformity inside the PTVs should be specified (e.g., ±5%). Many current inverse planning systems may have a hard time achieving the ±5% uniformity in the target volume that is commonly used for conformal treatments, but this is not caused by using IMRT per se but rather by the kinds of cost functions used within those planning systems.¹⁰³

• Keep the absolute maximum dose to the spinal cord to less than 50 Gy. Note that for many (but not all; see earlier) cost functions, there are tails on the DVH of many structures that may allow higher doses to a small volume of the organ.

• Minimize the dose to all normal tissues.

• If normal tissue responses are known, specify reasonable normal tissue limits for the planner. For example, one can specify that the mean dose to the normal lung stay below some specified value, because it is known that mean dose can be used to predict pneumonitis.

The physician should also define and specify any other constraints or evaluation expectations that will be used for the cost function, because there is no other way to specify to the optimization method what it should do or should not do.

Beam Arrangements for Inverse Planned Intensity-Modulated Radiation Therapy

The rules usually applied to beam arrangements for forward planned conformal therapy also apply to inverse planned IMRT beams in general. The beam arrangements that are most useful still depend on the site to be treated and on the geometry of the target volumes and normal tissues. Virtually all IMRT plans have at least three different beams, and many beam directions are sometimes used. Another type of IMRT called volumetric modulated arc therapy (VMAT, discussed later) uses rotational beams. In much of the early IMRT literature, it was common to find plans with seven or nine axial beams, with beam directions evenly distributed about the patient. This number of beams allowed enough flexibility for the optimization to achieve an adequate plan, even if the beam directions were not tuned with respect to the anatomy. However, as IMRT planning has progressed, the use of nonaxial and geometrically optimized beam directions has become more popular. Contrary to some early IMRT literature, it is still useful to use beams from outside the axial plane because they help to improve conformality. However, with IMRT, it is often only obvious that the nonaxial beams are useful when considering low-priority normal tissues because the IMRT optimization can typically achieve all of the higher-priority goals without use of nonaxial beams. One feature of most IMRT plans still remains the same: Typically, one avoids the use of opposed beams because opposing beams tend to lead to degenerate solutions that can cause difficulties for some optimization search algorithms, preventing them from reaching the optimal solution.

Optimization Method

The most important issue affecting the quality of the final IMRT plan is typically the type of cost function used and, in particular, how the different parts of the cost function are defined for each relevant organ or other anatomic object. In some systems, there is very limited flexibility in the type of cost function available because the cost function is limited to a simple method (often, a quadratic function of dose) in order to allow the gradient descent-based optimization search method to easily calculate the derivative of the cost function for each variable. However, other cost function methods can be quite general, allowing the use of dose, dose-volume, biologic model, or other types of costlets that can be combined into an overall cost function.⁶¹ In any event, it is the relationship of the different costlets (individual pieces of the cost function) that determines how important the various parts of the plan evaluation are, so by changing one or more parameters in a costlet or costlets, one can drive the solution of the plan toward a different type of solution. Learning how to modify the cost function to modify the kind of solution achieved for the plan is one of the most important aspects of inverse planning and may take a significant amount of effort to master.

Regardless of the type of inverse planning system or cost function used for inverse planning, one of the easiest ways to choose the appropriate cost function parameters for the plan is to first make sure that the clinical goals for the plan have been prioritized. Then, one constructs the cost function using the highest weights or power or importance for the high-priority issues and decreasing weights (or whatever) for the lower-priority issues. How this works in detail is specific for each type of inverse planning system, search method, and clinical site, but the general concept holds true for most inverse planning methods.

A number of different types of search algorithms are used for IMRT optimization, and these different methods can have some specific characteristics that are useful to know. Many current inverse planning systems, however, make use of only one search method, so many planners will not be able to choose different methods for specific patients or plans. Many inverse planning systems use gradient descent-based search methods because they are fast, but this forces their cost functions to be limited to easily differentiable functions, typically quadratic dose penalties with a form such as w_i(D − D_i)², where w_i is the weight of the penalty and D_i is the dosimetric goal for the ith object. This type of cost function tends to lead to tails on the DVHs because a small volume of the object can go higher than the desired dose without causing too much of a penalty. Another common concern about gradient-descent algorithms is the possibility of local minima in the cost function. For simple cost functions this is not a problem, but use of many costlets or complex functions can potentially lead to local minima that may trap the optimization search at a solution that is not the “global” minimum (the optimal solution). To avoid the possibility of being trapped in such local minima, stochastic search algorithms such as simulated annealing⁵² and others can be used. These algorithms can perform a global search that is more certain to achieve the global minimum of the cost function, but they often are quite slow and can also be sensitive to the search parameters used (as in fact are all of the search algorithms).

Plan Evaluation for Inverse Planning

If we knew exactly how to define the cost function so that it correctly summarized all of the physician’s goals and desires for the treatment plan, then if the optimization method worked correctly and achieved the global minimum, we would know that the best plan had been found, and that would complete the plan evaluation. However, this is not the case at the current time. Most clinical inverse planning is still limited by many things, so many IMRT plans still involve some “forward” iterative plan optimization. Where an inverse plan is performed, the plan is evaluated by the physician or planner, and the cost function is modified, and then the plan is reoptimized as the planner or physician attempts to push the plan toward some goal that he or she believes is a better clinical result than what the inverse plan gave on the first attempt. As cost functions and search methods become more sophisticated, we can expect that this iterative forward planning use of inverse planning technology will become less necessary.

This problem demonstrates that there are some important issues for plan evaluation. The greatest problem to resolve is that the cost function determines what the search method will choose as the optimal plan, but the physician evaluation of the plan may not be consistent with what the cost function is telling the optimization to do. Whenever this happens, the physician will find the “optimized” plan to be less beneficial than desired and will often attempt to make modifications in the cost function and then reoptimize the plan, perhaps many times, in order to achieve what was desired. A second complication is the fact that the metrics that physicians use to evaluate plans are often not the same kinds of metrics used by the cost functions, causing another mismatch between physician and cost function. Methods that use clinically relevant metrics as part of the cost function may help decrease this problem somewhat. A third problem with IMRT plan evaluation is that most planners and physicians still have a difficult time knowing what is actually achievable with a given plan, beam arrangement, and anatomic situation, and so, even when an acceptable plan is achieved, they are not sure whether they might be able to achieve a “better” plan by changing something. New methods are needed to help give the planner and physician tools that can show them the kinds of changes that may be possible, so they can better know when the “optimal” plan is achieved.

One final plan evaluation issue involves the many compromises and approximations that are involved in current inverse planning and IMRT processes. Due to the huge number of calculations involved in a single iteration of the optimization method, and the large number of iterations that are necessary to optimize a given plan for some search methods, it is essential to make compromises in the dose calculation method, the evaluation of the cost function, and other parts of the process, and to limit the resolution of dose calculation points used within the calculation. In addition, as described in more detail in the next section, various compromises in the plan must be made when converting the ideal intensities of the beamlets into a deliverable set of MLC segments (SMLC) or trajectories (DMLC), and these compromises will generally degrade the quality of the plan that is delivered to the patient. As any plan is evaluated, one must also keep in mind the further degradation that will occur. The limited resolution used for dose calculations can often lead to changes in apparent conformality or target coverage if a new dose calculation (with different resolution or calculation grid placement) is used. More discussion about plan degradation caused by MLC sequencing is included in conformal therapy delivery.

Plan Preparation

After a plan has been approved by the physician for clinical use, it must be prepared for treatment delivery and transfer to the delivery system. These preparations include the following:

• Documentation of plan approval. Documentation of the physician’s approval of the plan for treatment should be clearly performed (e.g., signature on the treatment plan output, or electronic signature in the electronic chart). Other approvals (e.g., quality assurance [QA] approval for IMRT treatments) may also be required.

• Preparation of treatment instructions. Creating a treatment plan for the patient usually involves determining the treatment beams to be used, but often does not consider other parts of the treatment delivery process, including patient setup and localization imaging, the treatment delivery methods to be used (e.g., automated delivery of all the beams, or manual delivery). For complex or automated treatments, creation of a complete treatment delivery script may be required. A graphic treatment delivery simulator can be quite useful.¹⁰⁴

• Calculation of MUs may be required for non-IMRT fields, although some treatment planning systems also provide the MUs for the plan. In any event, an independent check of the MUs required for treatment should always be part of the process.

• Quality assurance (QA) checks for IMRT and other complex plans. Before delivery of a complex plan to the patient, it is often appropriate to perform various QA checks on the treatment plan delivery, to assure that the treatment will work correctly and accurately. These kinds of QA checks are routine for IMRT treatments to confirm the correct dosimetric delivery of the complex beam intensity patterns that IMRT uses, and for stereotactic radiosurgery treatments to verify that the geometric accuracy of the treatment delivery is adequate, because large single doses are often delivered for these treatments.

• Hard copy plan output. Once a plan is finalized, it is important to create a hard copy (or electronic) output describing the treatment plan, to be used in the patient’s chart.

• Calculation of DRRs. The creation of DRRs is an important part of the plan preparation process because these DRRs will be used for localization checks of the patient position while one is performing the setup of the patient.

• Plan transfer. Transfer of the plan information into the paper or electronic treatment chart and the treatment delivery system is a very important part of the process. For manual transfer methods, one of the biggest potential problems is the large probability of transcription errors as the various parameters for each field are transferred from the planning system to the delivery system or chart. For electronic transfers of the plan information to a computer-based delivery system, the errors that happen are typically not random transcription errors, but systematic errors are more likely. The difference between the manual and electronic methods and their most likely errors means that the QA procedures for the two methods should be carefully defined to be appropriate for the method in use.

• If IMRT beams are used, then rather than calculation of MUs, “leaf sequencing” is performed for each beam to convert the planned IMRT intensity pattern into either multiple fixed MLC segments (SMLC) or dynamic trajectories of each leaf pair if the DMLC method is used.

MLC Leaf Sequencing

To convert an IMRT intensity pattern into a delivery prescription that will create such a pattern of intensities, a leaf sequencing algorithm is used. Many different leaf sequencing algorithms exist, and all have advantages and disadvantages. Any leaf sequencing algorithm is really an optimization procedure that attempts to find the best way to create the desired intensity pattern while still adhering to all the constraints or limitations of the MLC and machine that are to deliver the IMRT intensity pattern. There are always compromises in the final result achieved by any of these algorithms, so one always knows that the plan has been changed (usually degraded) by those compromises. It is thus important to verify that the final “deliverable” plan is still acceptable. Many IMRT systems use a second dose calculation, based on the actual delivery sequence, to check that the plan is still within acceptable limits even after the compromises.

Several different methods of IMRT delivery make use of MLC systems (Fig. 15-11). SMLC, or segmental MLC, uses a set of fixed MLC segment shapes to deliver the intensity pattern, with the beam off as the MLC moves from one shape to the next. Many algorithms have been developed for this kind of MLC sequencing.^45,¹⁰⁵ DMLC (dynamic MLC) uses MLC leaves that move on a trajectory that is defined so that the desired intensity pattern is created. Typically, a “sliding window” algorithm is used to derive the trajectories required.⁴⁷ Volumetric modulated arc therapy (VMAT) is a relatively new IMRT method that combines rotational (or arc) delivery and MLC-based IMRT.^59,⁶⁰ Development of more efficient or effective algorithms or methods and characterization of the details of the differences between the various methods already in use are areas of active research. It appears that no one method or system is clearly better than the others, although there can be specific advantages or disadvantages for particular treatment types that are associated with a particular sequencing or delivery method.

Figure 15-11 Segmental multileaf collimator (SMLC) and dynamic multileaf collimator (DMLC) leaf sequencing.

Conformal Therapy Delivery

Careful, efficient, and accurate treatment delivery is just as important as sophisticated treatment planning if the patient is to receive the benefits typically ascribed to conformal therapy.

Patient Setup and Localization

One of the most crucial aspects of patient treatment involves the setup and localization of the patient for each treatment fraction. For many years, this process has typically involved lining the patient up to laser lines using skin marks defining lateral and anteroposterior (AP) projections of the plan isocenter. To verify correct patient positioning, lateral and AP localization films were obtained (typically, once per week) to confirm the correct placement of the isocenter inside the patient. The “orthogonal pair” films were compared by eye with the expected location (shown by DRR or BEV display from treatment planning) to document the accuracy of the setup. Use of a calibrated gradicule in the accelerator head when making the megavoltage images ¹⁰⁶ was very helpful in making the comparison somewhat quantitative.

In the last several years, many new developments have begun to change the setup and localization process to allow much more quantitative and automated setup procedures. The development and implementation of electronic portal imaging devices (EPIDs)¹⁰⁷ for megavoltage imaging, diagnostic x-ray imagers,¹⁰⁸ and megavoltage or diagnostic CBCT scanning⁶⁵ using accelerator-based systems have revolutionized the setup and localization process. If the new digital imaging capabilities are integrated with the computer-control system of the treatment accelerator, it is possible to perform relatively automated patient setup with accuracy much improved over the old manual method.¹⁰⁹ Over the next years, it is clear that integration of cone beam imaging into the treatment localization and setup process will lead to significantly improved accuracy for routine patient setup.

Manual and Computer-Controlled Treatment

The treatment delivery process used for modern conformal therapy has also changed. The usual method involved two or more therapists carrying the paper treatment chart into the treatment room, setting up the patient (as described above using lasers and skin marks), and then positioning the accelerator manually for each treatment field using the accelerator hand pendant controls. The therapists would exit the room in order to allow the treatment of each field, which was controlled by the treatment machine console in the control room. Although some machines were equipped with computer-based “record and verify” systems that would confirm the parameters to be used for each field, many of these systems were of limited sophistication. Most parameters used for patient treatment were set by hand and susceptible to various random errors.

Since the 1990s, integrated computer-controlled systems have entered widespread use for control of the treatment process, usually tied to a treatment plan or information system database that contains all the treatment plan information. To varying degrees, the progression through the setup process and treatment delivery is often automated, controlled via the machine’s system. However, much remains to be done to optimize the efficiency, safety, and accuracy of the computer-controlled treatment delivery process. There has been some published work on the treatment delivery process,^104,¹¹⁰^–¹¹² but too often a treatment process taken directly from the old manual techniques is implemented with the computer-controlled system, without attention to changes needed for accuracy, safety, or efficiency. As the patient setup process, along with CBCT or localization imaging using integrated electronic imaging systems, is integrated into the computer-controlled treatment process, additional effort to optimize the process will be required. Much work is needed in this area.

The use of IMRT treatment delivery has also caused a dramatic change in aspects of the delivery process. With simple nonconformal fields, it was common to outline the shape of each treatment field on the patient’s skin and to confirm at treatment that the shape and placement of each radiation field agreed with the shape drawn on the skin. This limited very large errors but certainly was not a highly accurate positioning check. With modern IMRT treatment fields, understanding the fluence pattern that will be delivered requires computer-generated images, and there is no intuitive way to check the field shape, position, or intensity pattern directly on the patient, so more sophisticated QA checks or procedures are necessary. With modern EPID systems, so-called portal dose measurements (intensity or “dose” measured with the EPID) can be compared with that which is expected, and used as a QA check that can potentially identify both geometric and dosimetric differences between the desired and delivered dose distributions.^113,¹¹⁴ It is also possible to reconstruct the delivered intensity distribution by analysis of MLC trajectory information.¹¹⁵ The further development of quantitative online delivery QA checks along with their integration into the treatment delivery process is an area of significant ongoing experience that is expected to lead to much more sophisticated delivery systems.

Patient Treatment Chart

As computer systems have increased the sophistication of planning, treatment delivery, and treatment verification and QA, the need for more sophisticated treatment documentation has also grown. Computer-controlled treatment plans and IMRT require the use of electronic patient treatment charts because there is just too much technical information to allow sole use of a paper treatment chart. The trend toward the use of electronic patient charts is certainly mirrored throughout the health care system as all major medical systems convert their paper-based record system to an electronically based system.

It is incumbent on all users of such new electronic chart systems to carefully implement and use the new technology. It is not appropriate to just convert all the old paper-based documentation practices into electronic forms, as some standard paper-based methods just do not work well in the electronic world, such as writing notes in a chart in pencil to show what should happen in some number of future treatments. Some analysis of the needs for electronic treatment charts has been published,¹¹² but the ongoing transition from paper to electronic charts should be evaluated and performed with care. Complex IMRT-based prescriptions and plans probably require new methods that were not supportable in a paper-based chart, and users of highly sophisticated conformal therapy will have to develop their own migration from paper to electronic prescriptions and treatment charts.

Consideration of Setup Error and Patient Motion

As the field of radiation therapy has made the transition from four-field boxes and AP-PA fields to sophisticated conformal therapy based on 3D treatment planning and IMRT, it has become more obvious that considerations of daily setup uncertainties and errors and patient and organ motion during treatment are important considerations that should affect both treatment planning and treatment delivery methods. During multiple fraction treatment, all of the following effects should be considered:

• Daily setup variations

• Changes in patient anatomy and target shape over time

• Patient movement and organ motion during treatment

For many years, the ICRU-50 concept of the PTV ⁷² has been the sole response to these issues. However, the desire to perform precise conformal therapy has led many institutions to study these issues in more detail and to measure the uncertainties associated with these different issues.

The importance of correcting the systematic setup uncertainty, if possible, has also led many institutions to convert their positioning verification procedure from the use of weekly port films, with position correction if a large-enough error is seen, to a more sophisticated offline or online repositioning scheme, and to more sophisticated setup, localization, and imaging strategies that are now called image-guided radiation therapy (IGRT). Use of daily IGRT setup correction based on imaging (either megavoltage imaging, kilovoltage imaging using an integrated kilovoltage tube and imager, or kilovoltage or megavoltage CBCT) is the most accurate method for daily patient positioning, although these IGRT techniques require more effort. Even relatively straightforward daily imaging techniques using EPID imaging and automated repositioning of the treatment table demonstrate large improvements in setup accuracy (e.g., improvement in Σ from >8 mm to 2.3 mm for liver patients).¹¹⁶ It is also possible to apply “decision rule” protocols^117,¹¹⁸ in an offline manner: Verification imaging is performed several days in a row, and only after the systematic error in the setup is determined is a correction made. Daily IGRT also provides detailed information about the stability of patients and their positioning, and can make possible adaptive treatment changes that can allow individualization of the margins and setup techniques used for individual patients. This is a rapidly developing area, and promises improvements in patient treatment precision, as well as in improving our ability to image¹¹⁹ and respond to changes in the patient, tumor, or normal tissue behavior through the course of therapy.

Clinical Considerations

To this point, the technical process of conformal therapy planning and delivery has been described. This final section describes some of the clinical considerations involved in the application of conformal therapy techniques and provides a very brief summary of some clinical results of IMRT treatments.

Immobilization, Setup Uncertainties, and Patient and Organ Motion

Patient positioning and immobilization should be comfortable but minimize motion and setup uncertainties. The clinically measured range of motion and setup uncertainties for each clinical site should be known, as they will determine the margins required for the PTVs (and planning organ at risk volume [PRVs], if used). Additional margin information can be obtained from the literature, as margins for a sample of patient populations using commercial, widely available immobilization systems have been determined and published; for example, the setup error standard deviation when using a commercial thermoplastic mask for head and neck immobilization has been determined to be 3 to 4 mm if patient-specific information is obtained and daily patient repositioning protocols using portal imaging are enacted.¹²⁰ Although weekly verification is considered the standard method for conventional radiation therapy and 3D radiation therapy, daily imaging of the skeletal anatomy or implanted fiducial markers using an EPID has reduced systematic and random variations.

The systematic setup error can be established from imaging during the first three to five treatments ¹²¹ and can then be corrected, leaving only the random deviation to be accounted for by adequate PTV margins. Daily patient imaging and correction of setup deviations can reduce it even further, making it possible to reduce the PTV margin further, thereby increasing the sparing of adjacent noninvolved tissue. Individual measurements of motion and setup uncertainties, and their corrections, are especially important in sites where breathing or internal motion is significant. Techniques for dealing with this motion include active breath hold or gating for respiratory motion,^69,¹²² beam tracking of moving targets,¹²³ or fiducial markers for correction of prostate movement due to rectal filling–related changes in target position.¹²⁴

Considerations of patient setup and immobilization approaches are complex. For example, in prostate cancer, prone patient position improves the separation between the prostate and the rectum compared with supine position,¹²⁵ thus potentially improving sparing of the rectum. However, prone positioning also increases breathing-related motion of the prostate.¹²⁶ In head and neck cancer, immobilization of the shoulders is essential if targets in the low neck are included in the conformal plan but is not required if the low neck is treated with an anterior field matched with the IMRT plan for the high neck.

Determining the Targets

The dosimetric advantages of IMRT need to be balanced with potential pitfalls related to the production of tight dose distributions around targets outlined on the CT scan. The most important issues are the reliability and reproducibility of outlining the targets that typically rely on a single planning scan performed prior to the start of treatment. The GTV is outlined on the treatment planning CT scan using clinical and radiologic information. In many cases, contrast-enhanced CT combined with clinical information derived from physical examination is the main modality used for the delineation of the targets. For some sites, CT is not the best imaging modality for the definition of the extent of the macroscopic tumor, and ancillary studies such as MRI or fluorodeoxyglucose (FDG)-PET may add important information. MRI can be limited by its sensitivity to artifacts, difficulty in interpretation, long examination time, and cost. FDG-PET is often limited by a lack of specificity regarding tumor versus inflammation, and by uncertainties in interpretation, which may be subjective and differ depending on the observer’s experience. PET-defined target volumes can also depend substantially on the standardized uptake value (SUV) chosen to contour the PET scan. The best added information from these studies is derived following their registration either with diagnostic CT or with the planning CT scans. Rigid body geometric registration of multiple scan types can now be performed in many cases using commercially available software, whereas registration of nonrigid structures such as the rectum or bladder is the subject of ongoing research.¹²⁷

Future improvements in the anatomic and metabolic imaging of tumors are expected to decrease the uncertainties in outlining the GTV to determine the volume to receive high radiation dose. However, the definition and outlining of the tissue volumes at risk of harboring subclinical disease (CTVs) depend on clinical judgment alone. It is not surprising that large interobserver differences have been noted in outlining these volumes.^128,¹²⁹ The uncertainties in outlining the target volumes raise concerns about the potential of highly conformal radiotherapy to miss disease while striving to spare organs adjacent to the targets. Efforts to define the volumes at risk for subclinical disease for each tumor site have been made for head and neck cancer,^130,¹³¹ breast cancer,¹³² and cancer at other sites.¹³³ They constitute the initial steps in this direction. Clinical validation, based on data regarding the sites of locoregional tumor recurrence and their relationships to the targets, is necessary. To date, relevant data are scant. They suggest that when target outlining is performed by experienced investigators, the large majority of the locoregional tumor recurrences are in-field, and only few marginal failures are observed.^134,^135,¹³⁶ Until more data are available, target outlining should be done in a conservative manner. Target volume definition requires thorough knowledge and understanding by the radiation oncologist of the anatomy and the locoregional tumor spread pattern.

An outline of imaging studies required for each tumor site and other considerations for determining the targets are described next.

Head and Neck Cancer

Gross Tumor Volume

Most head and neck target volumes are defined using CT information. However, MRI is a necessary adjunct to CT for tumors close to the base of skull (i.e., nasopharyngeal and paranasal sinus cancer), where it provides better details of tumor extension and better details of the parapharyngeal and retropharyngeal spaces, compared with CT.¹³⁷ MRI is essential for delineating the targets in these cases. FDG-PET, in most cases, defines smaller primary targets compared with CT, and the volumes depicted by each modality do not completely overlap.¹³⁸ Relying on FDG-PET may result in changes in the outlining of the GTVs compared with outlining based on CT, at least in some patients.¹³⁹ Whether FDG-PET is more accurate than CT in delineating the primary tumor GTV has not yet been established, due to the paucity of data correlating the imaged extent of the gross tumor and its size in pathologic specimens following resection. Current data from series involving imaging followed with surgical validation suggest slightly more accurate definition of the gross tumor by FDG-PET compared with CT. In the delineation of neck lymph node metastases, FDG-PET has a higher sensitivity than CT.^139,¹⁴⁰ Until more data are available, the most prudent practice seems to be outlining the GTV as the composite of the lesions observed on CT and PET. It should be emphasized that FDG-PET has poor sensitivity for occult disease (25% in the clinically negative neck¹⁴¹) and cannot be relied on to determine the CTV. In cases of recurrent cancer, where extensive scars from previous surgeries confound the CT-based delineation of the tumor, FDG-PET has a clear advantage¹⁴² and should be used as the primary modality for the delineation of the GTV.

Clinical Target Volume

The CTVs for head and neck cancer include subclinical disease in the vicinity of the primary tumor and lymph node groups at risk of subclinical disease. Outlining the CTVs requires knowledge of the natural history and pattern of failure of each tumor site. Suggestions for the selection and delineation of the targets have been published,^130,^131,¹⁴³^–¹⁴⁶ and atlases detail the recommended delineation of various lymph nodal groups. Studies detailing tumor failure patterns and how these patterns can affect modification of target outlining rules are emerging.^134,¹³⁶

Lung Cancer

Gross Tumor Volume

In lung cancer, the primary lesion should be contoured on the CT scan using window/level settings for pulmonary visualization (i.e., a pulmonary window). FDG-PET adds significantly to the staging information gained from CT regarding the tumor extent and is an essential tool for the delineation of the GTV, with significantly better accuracy gained by fused PET-CT compared with each modality performed separately.¹⁴⁷ This is particularly helpful when an FDG-PET avid but anatomically indistinct tumor is present. PET can differentiate atelectasis from tumor, preventing unnecessary inclusion of noninvolved lung tissue in the GTV.¹⁴⁸ In the free-breathing patient, FDG-PET depicts the tumor extent in both expiration and inspiration, providing information about breathing-related tumor motion and position changes that is not gained from the fast CT. FDG-PET is expected to improve significantly the outlining of the GTV for radiation therapy purposes.¹⁴⁹

Clinical Tumor Volume

The CTV for the primary tumor requires 6- and 9-mm expansions for squamous cell carcinoma and adenocarcinoma, respectively, according to histopathologic studies.^150,¹⁵¹ Outlining mediastinal lymph nodes at risk of harboring subclinical disease may be performed using the surgical definition of the mediastinal nodes and an atlas describing their positions for radiotherapy planning purposes.¹⁵² Some centers (University of Michigan¹⁵³ and Memorial Sloan-Kettering Cancer Center¹⁵⁴) do not define mediastinal nodal CTVs and do not intend to irradiate lymph nodes at risk without evidence of involvement, due to the pattern of locoregional failures in these patients that is predominantly within the GTVs.

Because respiratory motion can significantly affect the dose distribution, lung target volumes often require use of an internal target volume-like margin (from ICRU-62 ¹⁵⁵) for respiratory motion of the target, depending on whether measures to reduce the motion are used (e.g., ABC or gating). In the absence of such measures, large variability in tumor motion is observed among patients, on average 1 cm in different directions. However, the variability among patients is extensive and requires individual measurements.¹⁵⁶ Advanced image-guided systems, including CBCT and tomotherapy, can visualize the tumor and organs at risk in three dimensions prior to treatment that allows for online correction. This strategy is especially useful in treating early-stage lung cancer patients with stereotactic body radiation.

Brain Cancer

Gross Tumor Volume

The GTV is defined using MRI with and without contrast and then registered with the planning CT. MR T1-weighted images with contrast show contrast-enhancing tumors such as meningioma and glioblastoma. T2-weighted images show edema that is likely to contain microscopic tumor in cases of high-grade glioma. T1-weighted fluid-attenuated inversion recovery (FLAIR) images can differentiate brain infiltrated by tumor from edema caused by a mass effect and may enhance the definition of nonenhancing tumors.¹⁵⁷

Clinical Tumor Volume

Lacking well-defined anatomic compartments, the CTV for a brain tumor typically consists of a uniform expansion of the GTV by 1 to 2 cm depending on tumor grade. At the University of Michigan, dose-escalation studies have aimed at increasing the dose to the GTV alone, whereas the CTV dose remained constant in order to limit brain volume receiving a high dose.¹⁵⁸

Gastrointestinal Cancer

Many clinical trials in conformal radiotherapy in gastrointestinal cancer have been performed in pancreatic cancer, in which tumor control rates are poor and 3DCRT and IMRT may reduce the volume of small bowel and duodenum irradiated, potentially allowing higher radiation doses.¹⁵⁹ Outlining the targets must take into account significant respiratory-related organ motion, ranging up to 25 mm in various directions.¹⁶⁰ Without methods to reduce respiratory movement, the large PTV margins required to accommodate this motion are expected to increase markedly the volumes of the small bowel and duodenum receiving target doses and limit the ability to escalate tumor dose. Highly conformal radiation techniques also show significant acute toxicity benefit in the treatment of anal cancer. To assist with target delineation, CTV guidelines have been proposed by the RTOG consensus panel for anal cancer treatment.¹⁶¹

Prostate Cancer

The gross disease within the prostate gland cannot accurately be defined using conventional imaging, and routine practice is to outline the CTVs. Research in magnetic resonance spectroscopy (MRS) suggests that it may be used to identify tumor cell foci within the prostate using the choline-to-citrate ratio, which is higher in tumor cells compared with prostatic cells.¹⁶² If confirmed, MRS imaging may be used to define the GTV, or the boost volume, within the prostate.¹⁶³ The prostate’s superior margins at the base of the bladder and its inferior margin (apex) are better defined on MRI compared with CT. MRI is also better at defining the rectal and bladder boundaries and the penile bulb. Prostatic motion is on average 5 mm in both anterior-posterior and superior-inferior directions. If adjustments for this motion are not made, 1-cm margins are required to take into account motion and setup uncertainties. The PTV is typically defined using these margins around the prostate alone in patients with a promising prognosis; around the prostate and seminal vesicles in intermediate-risk and high-risk patients; and including the pelvic nodes in high-risk patients. Some centers limit the margin posteriorly at the interface with the rectum to 0.6 cm.¹⁶⁴ High-risk patients and select adjuvant/salvage basis are considered for postoperative bed and/or pelvic IMRT to decrease gastrointestinal and genitourinary complications, In cases of pelvic IMRT, a nodal CTV is generated using a 7-mm margin around the iliac vessels with sparing of the bowel, bladder, and bony structures.

Gynecologic Cancer

Accurate target delineation is vital in the conformal radiation treatment of cervical cancer. MRI defines gross abdominal disease better than CT and is strongly recommended in defining the tumor volume, which may be further aided by FDG-PET.¹⁶⁵ Fusion of the T2-weighted MRI images to the simulation planning CT scan is advised when possible. To assist in delineating targets, the RTOG consensus panel has proposed guidelines.¹⁶⁶ In brief, the CTV should include the GTV, remaining cervix, uterus, parametria, ovaries, and vaginal tissues. The nodal CTV must include involved nodes and the draining nodal groups, including the common iliac, external iliac, and internal iliac nodes as well as the obturator and presacral nodes. Inclusion of the para-aortic lymph nodes is based on the extent of disease and results of staging investigations. In defining the nodal volumes, most clinical studies outlined along blood vessels with margins of 0.5 to 1.5 cm.¹⁶⁷ Due to substantial organ motion, tumor regression, and deformation associated with cervical cancer patients, margins of 1.5 to 2 cm around the CTV are recommended for the primary PTV with a PTV margin of 7 mm around the nodal CTV. Whether IMRT can be useful in the treatment of an intact cervix, replacing brachytherapy or, more likely, serving to augment GTV doses after brachytherapy, is not yet known. If IMRT is considered for the intact cervix, daily soft tissue image-guided verification is required to avoid the risk of geographic miss.

Breast Cancer

Some work on breast cancer treatment has used targets obtained by outlining the whole breast or chest wall while trying to reproduce the tissue volume that would have been irradiated with standard tangential fields, although this target is not really based on anatomy but on the older type of treatment. In order to achieve such a reproduction, marking the borders of clinical tangential fields by radiopaque wires such that they are apparent on the planning CT ensures inclusion of all tissue irradiated traditionally. There have been some efforts to define anatomic target volumes.^168,¹⁶⁹ For boost planning, the lumpectomy scar is outlined as the CTV and a 1-cm margin is assigned as the PTV, for breathing motion and setup uncertainties. When regional lymph nodes require irradiation, treatment complexity is higher, and heart volumes in left-sided breast cancer therapy may be substantial. For the outlining of the lymphatic CTVs, particularly the supraclavicular lymphatics, some anatomic guidelines have been published.^132,¹⁷⁰

Accelerated partial breast irradiation (APBI) is increasingly being used to treat early-stage breast cancer patients. The optimal margin of tissue requiring radiation after lumpectomy in patients treated with APBI remains somewhat controversial. However, contemporary radiographic and pathologic data suggest that a margin of 10 mm around the tumor bed appears adequate to cover any disease remaining in the breast after lumpectomy in most (>90%) of patients, provided the final margins are negative.¹⁷¹ The CTV is then limited to 5 mm from the skin surface and lung chest-wall interface. The CTV is further expanded by 1 cm to create the PTV.¹⁷²

Noninvolved Normal Tissue

Forward planned 3DCRT allows intuitive decisions by the planner regarding beam apertures and directions, but such decisions cannot be made in inverse planned IMRT. This requires that all targets and organs at risk be determined and outlined in the planning CT for use in the inverse planning. In addition, inverse planning typically requires the definition of “nonspecified tissue.” This information is usually obtained by subtracting all targets and specified organs at risk from the external contour of the patient in the CT dataset. It allows constraints on the maximum doses delivered outside the targets and the organs at risk, which is necessary to reduce unexpected damage to soft tissue, nerves, blood vessels, and so on.

Treatment Goals and Rationale for Highly Conformal Radiotherapy

Head and Neck Cancer

Delivery of the prescribed dose to the targets in advanced head and neck cancer is often limited by the dose to the spinal cord and brainstem, especially in advanced nasopharyngeal cancer, posterior pharyngeal wall cancer, and thyroid cancer. Also, irradiation of gross disease in the posterior neck may be suboptimal due to dosimetric deficiencies due to the off-cord photon and posterior neck electron beams. By making possible concave dose distributions, IMRT can overcome these deficiencies and potentially improve tumor control rates. Dose escalation to the GTVs ¹⁷³ or to hypoxic subvolumes within the GTV¹⁷⁴ using IMRT has been proposed.

Sparing of the parotid glands, in an effort to reduce xerostomia, has been a major goal of earlier 3DCRT ¹⁷⁵ and subsequent IMRT studies.^103,¹⁷⁶^–¹⁷⁸ Randomized studies of IMRT versus 2D radiation therapy in nasopharyngeal cancers demonstrated improvement in salivary production^179,^180,¹⁸¹ and in observer- and patient-reported xerostomia.¹⁸⁰ Sparing the contralateral submandibular gland in cases where contralateral level IB is not a target has been proposed, suggesting a mean dose of less than 39 Gy as a goal.¹⁸² However, IMRT cannot substantially preserve the function of the submandibular glands in some cases because these glands lie anterior to the subdigastric nodes that are important targets in both sides of the neck, and this poses a limitation on reducing xerostomia by IMRT. Similarly, reducing the doses to the noninvolved oral cavity, striving to spare the minor salivary glands¹⁸³ and to reduce mucositis, is an important planning goal. Additional objectives include reduced doses to the optic pathways and inner ears in patients treated for advanced nasopharyngeal and paranasal sinus cancer,^178,^184,¹⁸⁵ to the skin,¹⁸⁶ to the carotid arteries in cases of stage T1 to T2 glottic laryngeal cancers in patients for whom treatment of regional lymph nodes is not planned,¹⁸⁷ and to the swallowing structures damage to which following intensive chemoirradiation may cause dysphagia and aspiration.¹⁸⁸

Brain Cancer

Dose escalation trials attempting to increase the currently poor local control of high-grade gliomas have been conducted at the University of Michigan.¹⁵⁸ These trials used 3DCRT or IMRT delivering a high dose to the GTV while keeping a constant dose to the CTV and limiting the doses to the optic pathways and noninvolved brain. IMRT may be superior to 3DCRT in some cases in tumors that are close to the optic nerves or chiasm.

Lung Cancer

Facing low locoregional tumor control, the goals of conformal irradiation are to improve target irradiation while reducing lung volumes receiving a high dose. Using 3DCRT, trials of dose escalation have been conducted using various constraints on uninvolved lung doses, such as the mean dose, the effective dose (V_eff), or the lung volume receiving more than 20 Gy (V₂₀).¹⁸⁹ These constraints require that only grossly involved lymph nodes be included in the targets because inclusion of nodes at risk but without evidence of involvement increases significantly the lung volumes treated to high doses.¹⁵⁴ Using similar dose constraints, IMRT has been used to treat lung cancer.¹⁹⁰ However, substantial geometric uncertainties about beamlet doses in lung tissue may influence the accuracy of imaging, treatment planning, and delivery. To improve the optimization of IMRT beam direction, the role of ventilation perfusion scans is being explored so that well-ventilated lung may be more appropriately spared.¹⁹¹ To this point, the clinical use of IMRT for lung cancer remains relatively restricted by respiratory motion and by uncertainties about beamlet doses in lung tissue.¹⁹²

Breast Cancer

IMRT can produce more homogeneous doses to the intact breast compared with lateral tangential beams, potentially improving cosmetic results. In addition, IMRT may help reduce dose to lung and heart during therapy for left-sided breast cancer, especially in patients requiring comprehensive breast and nodal irradiation.¹⁹³ However, it has been argued that similar improvements in breast dosimetry can also be achieved with advanced forward planned 3DCRT.

The use of partial breast irradiation is a recent strategy for the treatment of early-stage breast cancer. The main advantages of this technique include shortened overall treatment time, better access and compliance, as well as potential reductions in normal tissue complications due to reductions in treatment volumes. This accelerated hypofractionated irradiation regimen to the lumpectomy site takes into account the pattern of breast relapse that predominates in the original tumor quadrant and the assumption that good cosmetic outcome may be achieved if high fraction doses are delivered to limited breast volumes using IMRT. 3DCRT APBI has been delivered using three to five noncoplanar photon beams with a planning goal to treat 90% of the planning target with at least 90% of the prescribed dose while ensuring that less than half the breast gets 50% of the prescription dose.¹⁹⁴

Prostate Cancer

Dose escalation to the prostate ^125,^195,^196,¹⁹⁷ and concomitant delivery of a high dose to the intraprostatic gross tumor defined by various imaging modalities described above, while limiting the doses to the rectum and bladder to reduce the main toxicities of therapy, have been the major goals of 3DCRT and IMRT. Efforts have been reported to reduce the doses to the penile bulb¹⁹⁸ or pudendal arteries¹⁹⁹ to reduce the rates of treatment-related erectile dysfunction. In high-risk prostate cancer patients, when IMRT is used to treat the pelvis along with the prostate and seminal vesicles, the main purpose is to decrease the dose to the small bowel while maintaining lower doses to the bladder and rectum as above.

Gastrointestinal Cancer

Pancreatic cancer has high rates of local and distant failure despite aggressive surgical resection. In an attempt to improve this dismal prognosis, the main goal of highly conformal therapy is to treat with smaller radiation fields to allow dose escalation and more aggressive chemotherapy.²⁰⁰ In contrast, anal cancer patients treated with definitive chemoradiotherapy have much higher cure rates. The aim of IMRT in this setting is primarily to avoid treatment breaks by reducing acute dermatologic, hematologic, and gastrointestinal toxicity.

Gynecologic Cancer

Most current studies of IMRT for gynecologic cancer aim at reducing the volume of small bowel irradiated during postoperative treatment.²⁰¹ Escalating the dose to the GTV using IMRT as a replacement for brachytherapy has been proposed,²⁰² but it is unlikely that escalated external beam doses can match the extremely high doses delivered safely by implants to tumor in the vicinity of the implant sources. To improve hematologic toxicity, an interesting concept of bone marrow sparing using IMRT in conjunction with bone marrow imaging has been proposed by Mundt.^203,²⁰⁴

Pediatric Tumors

The rationale for highly conformal radiotherapy in pediatric tumors is to limit the high-dose volumes in growing organs, especially bones. Although IMRT may achieve this goal better than 3D radiation therapy in some cases,²⁰⁵ it increases the volumes receiving low-dose radiation and the potential for future radiation-related cancers. This should be an important consideration. Only when an obvious benefit is expected should IMRT rather than simpler techniques be used in children. For example, in medulloblastoma, IMRT can reduce the dose delivered to the cochlea when a posterior fossa boost is planned with IMRT compared with parallel-opposed beams, but similar benefit may be gained by the use of 3DCRT.²⁰⁶

Target and Organ Prescription: Dose Constraints

Standard 2D radiotherapy has typically been delivered in two or three phases. For a typical head and neck example, in the first phase radiation is delivered to all targets, including nodal and high-risk volumes. After a dose that is likely to eradicate subclinical disease is delivered (typically, 46 to 50 Gy), additional irradiation is delivered only to the high-risk targets, to a total of approximately 70 Gy. In this scheme, all of the targets receive the same daily dose of 1.8 to 2 Gy. In contrast, with IMRT, the most commonly used plan is a single treatment plan that improves dose conformity compared with sequentially optimized plans.²⁰⁷ In this single integrated boost (SIB) technique, the high-risk targets receive both a higher total dose and a higher daily dose compared with the lower-risk targets, and compared with critical normal structures whose total maximal dose is constrained at total doses that are lower than the prescribed target doses. Smaller daily doses reduce the biologic effect of the doses delivered to the critical organs (normalized total dose [NTD]), so the SIB technique creates the situation in which the maximum critical organ doses usually allowed in standard radiotherapy become much more conservative when used within the SIB IMRT technique. In addition, the maximum doses specified in IMRT or 3DCRT are delivered to smaller (or much smaller) organ volumes compared with conventional radiotherapy. IMRT or highly conformal 3DCRT treatments may be safer than corresponding standard radiotherapy treatments even if nominal maximum critical organ doses are similar. On the other hand, higher-than-standard total target doses, delivered inadvertently due to nonuniform dose distributions typical of many IMRT plans, or due to intentional GTV dose escalation in an effort to increase tumor control rates,^183,²⁰⁸ are associated with increased daily doses, causing a further increase of the NTD. This has the potential to increase toxicity related to tissue embedded within the target. Such toxicity may be apparent long after therapy and its prevalence is not yet known. Dose escalation relying on the ability of IMRT to restrict the high-dose volume to the GTVs should be conducted only within careful clinical trials.

When Should the Use of Highly Conformal Radiotherapy/Intensity-Modulated Radiation Therapy Be Considered?

Planning, delivery, and QA of 3DCRT/IMRT are more complex, costly, and work-intensive compared with these factors for previous technologies. The use of 3DCRT/IMRT is justified if it offers apparent clinical advantages. The advantage in the dose distributions achieved by IMRT compared with 3D radiation therapy is mainly in the ability to form concave, horseshoe-like dose distributions. Such distributions are desirable in cases in which the target partly encircles a critical involved structure whose tolerance is less than the desired target dose. This includes head and neck cancer cases in which the targets are arranged anterior and lateral to the spinal cord and are bounded laterally by the major salivary glands; in prostate cancer, in which the rectum invaginates into the prostate target; in lung cancer, in which the target (usually, the mediastinal lymph nodes) may lay close to the esophagus; in esophageal cancer, in which sparing the lungs from high doses is an objective; in gynecologic cancer, in which the lymph node targets are arranged lateral to and posterior to the small bowel; in left-sided breast cancer, in which the target is concave anterior to part of the lung and heart; in brain tumors near the optic pathways; in medulloblastoma, in which the posterior fossa partly surrounds the inner ear; and in others. IMRT may also be indicated in cases in which minimizing the extent of the tissues receiving a high dose (at the expense of higher volumes receiving low doses) is likely to be beneficial, such as re-treatment of recurrent cancer.²⁰⁹ On the other hand, it is less likely that a dosimetric benefit will be gained from IMRT in cases in which tumors are remote from sensitive tissues or are adjacent to a sensitive tissue but do not (partly) surround it, compared with simpler conformal techniques. An example of a case where the dosimetric differences between 3DCRT and IMRT are small is prostate cancer, where the anterior wall of the rectum invaginates somewhat into the posterior prostatic target. Even these small differences may be translated into a clinically meaningful benefit in reducing rectal complications by IMRT compared with 3DCRT.¹⁶⁴

The high flexibility in creating desired dose distributions by IMRT provides the ability to deliver high doses to part of tumors judged to be at higher risk than other parts. Clinical accomplishment of this concept, termed “dose sculpting,”²¹² depends on the verification of the utility of innovative imaging of tumor physiology and early tumor response prediction.²¹³

IMRT treatment plans are often characterized by nonhomogeneous dose distributions in the targets that produce “hot spots,” such as target volumes receiving substantially more than the prescribed dose. This characteristic has been credited with a high rate of tumor control in nasopharyngeal cancer.¹⁷⁸ The potential for increased local toxicity due to “hot spots” is not yet clear and may depend on the site irradiated: Very high doses delivered to the nasopharynx may be well tolerated, as attested by the common use of intracavitary boost with radioactive sources for nasopharyngeal cancer, but may not be well tolerated by the mucosa in other sites in the head and neck. In any case, heterogeneous dose distributions by IMRT are not a necessity, because relatively homogeneous doses can be obtained. The decision of whether to deliver homogeneous dose belongs to the planner.¹⁰³

Potentially Negative Aspects of Intensity-Modulated Radiation Therapy

Several potential negative aspects of IMRT exist, for which there is as yet no clinical validation. Although IMRT reduces the tissue volumes receiving high doses, larger tissue volumes receive low doses compared with standard radiation therapy or 3DCRT. This is due primarily to use of many beams (often), many MUs, and leakage through the MLC leaves. This characteristic may increase the risk of radiation therapy–related malignant disease because the risk of radiation therapy–related mutations and carcinogenesis increases at intermediate rather than at high doses.²¹⁴ This risk is especially relevant for young patients. As the risk of radiation therapy–related cancers increases over time, usually past 5 to 10 years after therapy, clinical data are not available at this time.

Another theoretical concern is the loss of biologic effect of radiation therapy when treatment delivery time is prolonged.²¹⁵ Prolonged treatment delivery time is characteristic of some IMRT delivery techniques. There are differences in IMRT delivery modes that make a difference in this respect. For example, tomotherapy delivers sequential treatment throughout the target volumes, so that the exposure time of each tumor cell to daily radiation is short. In contrast, other systems irradiate all the targets simultaneously over a relatively prolonged daily treatment time. Whether the prolonged fraction delivery time translates into a clinical difference is not known.

Intensity-Modulated Radiation Therapy Clinical Results

Clinical results of the use of IMRT are still quite limited but have begun to emerge in recent years, mainly in head and neck and prostate cancer.

Head and Neck Cancer

Interpretation of the tumor control results of many of the clinical IMRT series reported to date is limited because they are very heterogeneous regarding tumor sites and stages, have relatively small patient numbers or patient selection factors, and are characterized by relatively short follow-up periods.^134,^135,^136,¹⁷⁸ These factors prevent meaningful direct comparisons of tumor control rates with similar patient series treated with standard radiation therapy. These series reported locoregional tumor control rates ranging between 81% and 97%. These rates seem to be better than those of most series of standard radiation therapy for similar tumors, suggesting that there may be no compromise in tumor control rates following IMRT. However, the recent emergence of human papillomavirus (HPV)-associated oropharyngeal cancer, associated with a better prognosis compared with smoking-related oropharyngeal cancers, may contribute to the high rates of tumor control in the IMRT series, which have been conducted at about the same time that HPV-associated oropharyngeal cancers have emerged. Multi-institutional prospective noncontrolled phase II studies of IMRT in nasopharyngeal²¹⁶ and early oropharyngeal²¹⁷ cancers conducted by the Radiation Therapy Oncology Group (RTOG) demonstrated high rates of locoregional tumor control and suggested that centralized QA can be successfully employed in the community, assuring a high compliance with strict IMRT specifications. Few small randomized studies of IMRT versus 2D radiation therapy or 3D radiation therapy for nasopharyngeal^180,²¹⁸ and other head and neck cancers¹⁸¹ demonstrated no significant differences in locoregional tumor control rates.

Most head and neck IMRT studies use fractionation schemes that strive to mimic those used in standard radiation therapy: total GTV doses of 70 to 73 Gy at 2 Gy/fraction and high-risk nodal volume (CTV) doses of 64 to 56 Gy at 1.8 to 1.6 Gy/fraction, all in 35 fractions.^176,¹⁷⁸ For advanced tumors, these schemes should be delivered concurrently with chemotherapy. For early oropharyngeal cancer, the Radiation Therapy Oncology Group (RTOG) conducted a phase II study in which the GTV, high-risk CTV (subclinical disease surrounding the GTV and first echelon nodes), and low-risk CTV (other lymph nodes at risk) were prescribed 66, 60, and 54 Gy, respectively, in 30 fractions (2.2, 2, and 1.8 Gy, respectively). This scheme represents a moderately accelerated course (biologically equivalent GTV dose of 70 Gy at 2 Gy/fraction, delivered over 6 weeks). More aggressive schemes include the simultaneous modulated accelerated radiation therapy (SMART) scheme developed at Baylor University, consisting of a GTV dose of 60 Gy at 2.4 Gy/fraction and a CTV dose of 50 Gy over 5 weeks for advanced cancer.²¹⁹ This group reported that delivering this scheme concurrently with chemotherapy resulted in intolerable acute toxicity.²²⁰ Using a similar scheme (SIB), the Medical College of Virginia group performed a phase I dose-escalation study in which they determined that 70.8 Gy in 30 fractions at 2.36 Gy/fraction delivered to the GTV was the maximally tolerated dose.²²¹ It is likely that different sizes and locations of GTVs affect the potential toxicity of GTV dose escalation regimens, and they should not yet be done outside of a clinical study. Also, the inability to deliver concurrent chemotherapy safely while increasing GTV doses is expected to limit the acceptance of similar regimens by other institutions.

Partial sparing of the parotid glands has been reported to result in partial preservation of salivary flows and in improved patient-reported ¹⁸³ and observer-rated^222,²²³ xerostomia, which improves even further over time.^183,²²³ Several recently published randomized studies of IMRT versus 2D radiation therapy or 3D radiation therapy support the improvement suggested in the previous noncontrolled studies. Pow and colleagues¹⁸⁰ reported lower rates of xerostomia and improved general quality of life measures in nasopharyngeal cancer patients receiving IMRT compared with patients receiving 2D radiation therapy. Similarly, Nutting and associates¹⁸¹ reported lower rates of observer-rated xerostomia in head and neck cancer patients receiving IMRT compared with those receiving 2D radiation therapy. In comparison, Kam and coworkers¹⁸² reported that while salivary output and observer-rated xerostomia were significantly better in nasopharyngeal cancer patients receiving IMRT compared with 2D radiation therapy, there was no difference in patient-reported xerostomia in the two arms. The reason for this discrepancy is not clear. It is possible that effort to spare additional salivary glands, such as the submandibular glands¹⁸² and minor glands dispersed in the oral cavity,¹⁸³ will achieve an additional gain in xerostomia. The improvement in xerostomia seems to be translated into improvements in broad aspects of quality of life.^224,²²⁵ It is possible, however, that the sparing of additional tissue may play a role in improving the quality of life. For example, IMRT may reduce the irradiated volumes of tissues whose damage or malfunction causes late dysphagia and aspiration after intensive chemoradiation therapy.¹⁸⁸ A clinical study assessing functional results of sparing by IMRT the structures whose malfunction after chemoirradiation may cause dysphagia demonstrated a high tumor control rate and very mild dysphagia after therapy compared with pretherapy, suggesting an improvement compared with previous studies.²²⁶ Dose-response relationships in the parotid glands have been reported by several investigators. The mean dose has been established as the most important dosimetric factor predicting salivary output after irradiation, and the relationships between the mean dose and the salivary output have been characterized as threshold relationships,²²⁷ exponential²²⁸ or linear.²²⁹ The mean doses below which substantial sparing of the salivary output was achieved were reported to be in the range of 20 to 39 Gy.²³⁰ The reasons for these discrepancies lie in different methods of determining salivary gland function, the different models used to assess response relationships, and the neglect (in most studies) of clinical factors such as certain medications and dehydration that have been found to contribute to the dose-related reduced salivary production.²²⁷

Excess toxicity associated with IMRT (compared with conventional radiation therapy) includes higher acute skin toxicity, which may be addressed by including the sparing of the skin in the optimization cost function,¹⁸⁶ by treating the low neck using an anterior beam matched to the upper neck IMRT fields (in cases of N0 neck), or by cutting the mask so that a bolus effect in the low lateral neck is avoided. Excess mucositis may be the result of inhomogeneous GTV doses in cases of oral cavity or oropharyngeal cancer or excess doses in the oral cavity outside the targets that can be reduced by sparing the noninvolved oral cavity.²³¹

Prostate Cancer

Several prospective and retrospective series suggest that prostate doses of greater than 75.6 Gy, delivered via conformal radiotherapy, increase freedom from biochemical failure rates compared with lower doses, especially in intermediate-risk patients.^195,^196,¹⁹⁷ A randomized dose escalation study at the M.D. Anderson Cancer Center compared 70 Gy with 78 Gy using a four-field box boost technique in the low-dose patients and a conformal six-field boost technique in the high-dose patients. Biochemical freedom from relapse in intermediate-risk to high-risk patients (prostate-specific antigen [PSA] levels >10 ng/mL) was significantly higher in the high-dose group compared with the low-dose group, while no significant difference was observed in low-risk patients (PSA levels <10 ng/mL).¹⁹⁷

Reducing rectal toxicity, a major dose-limiting factor in the therapy of prostate cancer, may allow dose escalation and a potential for improved cure rates. Partial sparing of the rectal wall seems to be the major advantage of IMRT, which may be essential in securing low rates of rectal toxicity while higher-than-standard doses are delivered to the prostate. The largest experience in this regard has been accumulated at the Memorial Sloan-Kettering Hospital in more than 700 patients.¹²⁵ This group reported that when doses of 81 Gy were delivered, IMRT resulted in significantly less acute¹²⁵ and late¹⁹⁵ rectal toxicity compared with previous techniques. They found that an average of 98% of the CTV could receive the prescribed dose of 81 Gy by IMRT compared with 95% with 3DCRT, and smaller volumes of the rectal wall (9% vs. 13%, respectively) and bladder wall (29% vs. 32%, respectively) received less than 75 Gy. Although they are statistically significant, these relatively small differences are usually not reported to be apparent clinically. However, the authors reported that a nonrandomized comparison of toxicity between patients receiving CTV doses of 81 Gy using IMRT or 3DCRT showed a significant reduction in acute rectal side effects and late rectal bleeding, suggesting that these high target doses should only be delivered with IMRT.¹⁶⁴

The relationships between the rectal wall volumes, dose, and rectal toxicity have been explored by several groups. The results vary depending on whether the percentage of the rectum versus the absolute volume is used.²³²^–²³⁴ The adoption of a certain set of volumes and doses, chosen from one of these publications, is necessary for the optimization of IMRT plans for prostate cancer. An example of one of the possibilities is provided by Zelefsky and colleagues¹⁶⁴ following their experience with dose escalation using IMRT. Another example can be derived from the M.D. Anderson Cancer Center randomized trial, in which an increase in the rectal toxicity grade of higher than 2 was observed when more than 25% of the rectal volume received more than 70 Gy.¹⁹⁷ To reduce the rectal dose, several investigators reported partial blockage of the overlap region between the PTV and the rectum such that the maximal dose to the rectum was limited to 72 Gy, notwithstanding the blockage of part of the target in the posterior prostate.¹⁶⁴ A hypofractionated schedule delivering 70 Gy in 2.5 Gy/fraction using IMRT and resulting in a low rectal toxicity rate was reported from the Cleveland Clinic.²³⁵ In addition to the rectum, constraints regarding bladder, femoral head, and penile bulb doses are required for IMRT planning. However, dose-volume-complication data for these organs are scarce, or relationships are weak, and published constraints for these organs are quite arbitrary to date. Although the use of IMRT with image guidance to treat definitive prostate cancer has become routine, IMRT in the postoperative setting has recently been investigated. Initial data reveal that when high-dose IMRT is used without daily image guidance, the acute toxicity is much greater (grade 2 or higher acute gastrointestinal and genitourinary toxicity of 15% and 31% in comparison with 8% and 14%, respectively).²³⁶ Similarly, in high-risk patients, pelvic IMRT with image guidance has demonstrated low genitourinary and gastrointestinal toxicity.²³⁷ Efficacy results of such treatment are still pending.

Lung Cancer

Efforts to increase locoregional control in the lungs include increasing the radiation dose using conformal radiotherapy. The results of several studies suggest that increased dose using 3DCRT tends to increase local tumor control.^238,^239,²⁴¹ These studies aim to treat gross disease alone (primary tumor and lymph nodes avid for FDG-PET) without adjuvant irradiation of subclinical mediastinal disease. At the same time that radiation dose escalation studies have suggested that local control can be improved, randomized studies have demonstrated that sequential chemotherapy combined with standard-dose irradiation is superior to standard-dose irradiation alone and that concurrent chemoirradiation is superior to sequential therapy. The next logical step in improving both local and systemic control of unresectable non–small cell lung cancer would be to escalate the radiation dose, using conformal therapy, with concurrent systemic chemotherapy. Because the time periods required for dose escalation regimens using daily 1.8- to 2-Gy fraction doses are exceedingly long, recent trials use a strategy of escalating the dose per fraction, keeping the overall time constant. When combined with concurrent chemotherapy, such regimens may be associated with prohibitive esophageal and lung toxicity, so a strategy that allows safe delivery of higher-than-standard fraction doses and chemotherapy using highly conformal radiotherapy is currently being sought in several institutions.

Data about dose/volume/pneumonitis risk following 3DCRT of lung cancer have been accumulated in recent years and should aid in evaluating treatment plans and protocols. Partial lung volumes receiving specified maximal doses, such as V₂₀ (the partial volume receiving >20 Gy), and others, have been used as metrics related to significant prognostic factors for pneumonitis risk.¹⁸⁹ These partial volumes are highly correlated with measures that have been found to correlate with the risk of pneumonitis, such as the mean lung dose¹⁸⁹ or the effective volume (V_eff), which denotes the lung volume receiving a homogeneous dose causing the same complication probability as the prescribed nonhomogeneous dose distribution.²⁴⁰ It should be noted that clinical factors, in addition to dosimetric ones, are likely to play a role in the risk of pneumonitis, such as the location of the tumor, preexisting lung function abnormalities, added chemotherapy, and others.¹⁸⁹

Acute esophagitis is another common toxicity of high-dose irradiation of lung cancer. Data suggest that this toxicity has a relationship with the dose and length of esophagus receiving a high dose.²⁴² The addition of concurrent chemotherapy markedly increases the risk of this complication.

Although dosimetric studies show advantages of IMRT over 3DCRT in some lung tumors, especially large tumors and those close to the esophagus, clinical data remain sparse for such treatment, as issues regarding lung motion, dose calculation accuracy, and treatment dose need to be addressed before large-scale clinical trials are conducted.²⁴³

Breast Cancer

Clinical experience with breast-only IMRT has been reported by several groups using simple dosimetric requirements as the IMRT technique goals.^244,²⁴⁵ Excellent cosmetic results, assumed to relate to the high degree of dose uniformity achieved with this technique, have been reported by Vicini.²⁴⁴ Additionally, a prospective randomized study comparing IMRT with standard noncoplanar tangential beam arrangements has shown that with IMRT, the rate of acute radiation dermatitis is reduced, which may lead to better long-term cosmetic results.²⁴⁶ Using IMRT for comprehensive treatment in patients requiring breast and regional nodal irradiation may have dosimetric benefits in reducing the lung and heart dose,¹⁶⁸ compared with 3DCRT, but it is limited by respiratory motion. Clinical studies of therapy using various techniques to accommodate motion are ongoing. Regarding APBI using the 3DCRT techniques, preliminary results of the phase II RTOG 0319 study demonstrated acceptable skin toxicity.¹⁷² However, a study from the University of Michigan using IMRT and active breathing control (ABC) revealed worse cosmetic outcomes.²⁴⁷ Results of the phase III NSABP B39 study randomizing whole breast irradiation to APBI will shed more light on this controversial topic.

Gynecologic Cancer

Tumor control rates following postoperative IMRT of endometrial cancer are high, suggesting that irradiation of carefully selected targets, rather than the whole pelvis, does not compromise tumor control rates.²⁴⁸ When extended fields are needed in gynecologic malignant disease, IMRT has been found to be particularly beneficial.²⁴⁹ These studies also suggest that acute gastrointestinal toxicity (and, to a lesser extent, genitourinary toxicity) is reduced compared with historic control patients who had received standard four-field radiation therapy.²⁵⁰ Late gastrointestinal toxicity was also found to be reduced following IMRT compared with standard radiation therapy in a retrospective analysis.²⁵¹ Another normal tissue that is relatively spared by IMRT compared with standard radiation therapy is the bone marrow, where sparing is especially relevant for patients receiving combined chemoirradiation therapy. Relative bone marrow sparing and improved blood counts were found following “standard” IMRT compared with conventional radiation therapy, and they may improve even further when the sparing of the bone marrow is included in the optimization cost function following bone marrow imaging.²⁵²