Clinical Rationale for Image-Guided Adaptive Radiotherapy

In a pioneering work on the clinical implementation of an offline adaptive radiotherapy process, Martinez and collegues¹¹ described a 150-patient prostate cancer protocol in which repeat CT scans and portal images from the first week of treatment were used to define a patient-specific PTV,¹² which was used to re-plan the patient’s subsequent treatments. The patient-specific PTVs were decreased by 24% with a range of 5% to 43%. They found that the level of dose escalation achievable on a per-patient basis was variable. For intensity-modulated radiation therapy the average dose increase with adaptive radiotherapy was 7.5%, with a 2.5% to 15% range.

The offline approach does not reduce the random component of error. The effectiveness of these techniques depends on the relative magnitude of random error, which contributes to systematic error correction uncertainty and potential clinical target volume (CTV) underdoses resulting from large daily fluctuations in target location. Online image-guided implementation uses each day’s estimated anatomy to correct for both random and systematic organ motion. Ghilezan and colleagues¹³ demonstrated that “idealized” online adaptive radiotherapy for prostate cancer (exactly reproducing plan optimization on each day’s anatomy), could escalate doses by 13% (relative to a fixed level of severe rectal complications) to more than 50% for relative radiosensitive structures (bladder). However, such conformality is achievable only if the uncertainty of online anatomy estimates is small relative to the daily random error.¹⁴

Motion and tumor shrinkage, as described in the introduction, make head and neck, thoracic, and abdominal tumor sites good candidates for image-guided adaptive radiotherapy. In describing a feasible online adaptive radiotherapy strategy of deforming intensities, Mohan and colleagues¹⁵ found a substantial decrease of coverage of the CTV and lymph nodes without adaptation resulting from the changing anatomy; however, with adaptation, target coverage was maintained. Similarly, Lee and colleagues¹⁶ found a 15% daily variation in the delivered versus planned mean parotid dose of head and neck cancer patients resulting from parotid migration into the high-dose region progressively throughout treatment. For stage I non–small cell lung cancer, Underberg and colleagues¹⁷ found significant (>5 mm) shifts of the average tumor position with respect to the bony anatomy in 26% to 43% of cases. Tumor shifting toward other normal tissues (e.g., the esophagus and spinal cord) may result in unintentional dose increases to these structures, which could be ameliorated by adaptive strategies. Ramsey and colleagues¹⁸ conducted a planning study on lung patients imaged with daily megavoltage CT. The observed tumor shrinkage (60%-80% through the course of treatment) was taken into account in their study, and they demonstrated a significant (17%-23%) reduction in the volume of dose, receiving more than 20 Gy in the ipsilateral lung. A caveat to focusing on the shrinking gross tumor alone is that there is concern about the lack of coverage of the subclinical disease in the CTV.¹⁹

An important consideration when performing image-guided adaptive radiotherapy with x-ray–producing devices is the dose received by the patient. The benefit of the use of the x-ray imaging devices must be outweighed by the benefit to the patient of more accurate treatment to obtain the optimum trade off for imaging frequency. A discussion of the dose from imaging procedures and guidelines for use is summarized in the American Association of Physicists in Medicine (AAPM) Task Group 75.²⁰

To date, there are no known prospective randomized trials comparing adaptive radiotherapy and standard-of-care, and the ethics and viability of such a protocol are an issue of much debate.²¹ It is the responsibility of the scientific community to develop carefully designed studies to quantify the clinical effect of adaptive radiotherapy and make a compelling argument for its use.

Timescales of Image-Guided Adaptive Radiotherapy

There are three timescales for image-guided adaptive radiotherapy: offline, online, and real-time as depicted schematically in Fig. 12-1. As the time between the intervention is reduced from offline through to real-time adaptation, the reliance on automated processes is increased. The increased potential for anatomic conformality offered by adaptation on a shorter timescale also increases the error risk of adaptive processes and careful human operator oversight is necessary.

FIGURE 12-1 • The different timescales—and associated reliance on automation—for image-guided adaptive radiotherapy.

A key tool for image-guided adaptive radiotherapy is deformable image registration that allows dose and contour information to be transferred between images, substantially reducing workload. As the timescales for adaptation decrease, the computation time of deformable image registration must correspondingly decrease. Development of an effective image registration technique has been one of the most important areas of research in medical applications of imaging technology.^22–28 Briefly, registration means to align two imaging data sets in a common coordinate space by transforming one image set while keeping the other one fixed. Deformable registration is complicated because it entails modeling of voxel-dependent distortion. Clinically, the need for robust deformable image registration algorithms to fuse images representing the same structures imaged in different conditions or on different modalities is ever increasing because of the extensive use of multimodality imaging and the emergence of new imaging techniques and methods. Examples of such applications in adaptive radiotherapy include (1) better tumor target definition,²³ (2) image enhancement,^29,30 (3) propagation of organ contours from one image set to another,^31,32 (4) calculation of accumulated dose in organs experiencing inter- or intrafraction organ deformation for 4-D or adaptive therapy planning,^24,33–38 and (5) 4-D image reconstruction.^39–42

According to the nature of image matching, registration can be broadly divided into manual, landmark-based, surface-based, intensity-based, and mutual information–based registrations. There are numerous algorithms developed for different applications in medical sciences and other fields. Depending on the mechanism or method used to model the deformation, registration can usually be categorized into elastic model,^43–45 viscous fluid model,⁴⁶ optical flow model,^47–49 finite element model,^44,50 radial-basis function model such as basis spline model,^51–53 and thin-plate spline model.^54–57 Each of these approaches has its advantages and disadvantages. Deformable image registration is a rapidly evolving area and intense efforts are being made from both academic institutions and industrial companies to make deformable registration a clinical reality.

A significant promise of adaptive radiotherapy is the compensation of uncertainties, including organ deformation and interfraction organ motion as well as dosimetric errors incurred in previous fractions.^15,58–61 To realize adaptive radiotherapy clinically, and maximally exploit the potential of this new form of IGRT, a robust planning strategy must be in place to support the decision-making process. In addition, there are a number of other software-related issues, such as visualization of doses, statistical process control, and so on, that need to be resolved before adaptive radiotherapy becomes widely accepted in the clinic.

Offline Image-Guided Adaptive Radiotherapy

Offline adaptation means that the imaging session and treatment session are separate. In Fig. 12-1, the time separation is shown to be on the order of a day or more. Head and neck cancer is particularly amenable to offline adaption because tumor shrinkage is significant. The rationale for offline image-guided adaptive radiotherapy is shown in Fig. 12-2, in which substantial shrinkage of the tumor throughout the course of therapy is observed between the anatomy at the time of the original CT scan and the cone beam CT scans acquired during treatment. A further example of tumor changes during head and neck cancer radiotherapy is given in Fig. 12-3, in which repeat CT scans (the second acquired for re-planning purposes) were acquired. The GTV contours from the original CT scan are shown overlaid on the mid-treatment scan, as well as the new GTV volumes deformably mapped from the original scan to the mid-treatment scan using Mimvista.

FIGURE 12-2 • Rationale for offline adaptive radiotherapy. Observed anatomic changes in a head and neck cancer patient throughout the course of radiotherapy. A, Original treatment planning CT. B and C, Cone beam computed tomography after 37.4 Gy and 44 Gy respectively. Note the marked decrease in the gross tumor volume with the corresponding increase in the air cavity during treatment progression.

FIGURE 12-3 • A, The contours drawn on the original treatment plan computed tomography (CT) scan (target: red; brainstem: light green). B, The original contours copied to the mid-treatment scan CT. C, The contours from the original scan deformably mapped to the mid-treatment scan (target: pink; brainstem: light green).

The CT scans from Fig. 12-3 were used for treatment planning, with the 95% and 100% isodose curves shown in Fig. 12-4. Notable are the large differences between the treatment volume from the initial scan and the mid-treatment scan. The corresponding dose volume histograms are shown in Fig. 12-5, demonstrating, as expected, that smaller treatment volumes result in significantly reduced normal tissue doses, and conversely, the larger normal tissue doses in the absence of adaptation. Care must be taken, as with all radiotherapy processes, with adaptation. Images are only that, and the interpretation of the images has a large interobserver variability.^62–65 Furthermore, questions arise for how to adapt the CTVs, in which by definition there is suspected but no image-visible disease. Further study as to the optimal clinical method or methods for adaptation, which are likely to be tumor-site–specific, is needed.

FIGURE 12-4 • A, The dose distribution from the original treatment plan computed tomography (CT) scan; target shown in red. B, The original dose distribution recalculated onto the mid-treatment CT scan indicating the dose in the absence of modification. C, The dose distributions of the adapted treatment plan; new target shown in pink.

FIGURE 12-5 • Dose volume histograms of spinal cord (A), brain stem (B), and targets of the treatment plans referred to in Figure 12-4 (C). Original treatment plan on the original treatment planning computed tomography (CT) scan (dotted blue); original treatment plan recalculated onto the mid-treatment CT scan indicating the dose in the absence of modification (solid blue); adapted treatment plan (dashed red) (C). A substantial dose reduction to normal tissue is possible in some cases with adaptive radiotherapy in the presence of shrinking tumors.

The imaging system used for offline adaptive radiotherapy is typically either a conventional CT scanner or an in-room CT scanner. The adaptation can be triggered by a protocol (for example after 20 and 40 Gy delivered dose), by clinical observations of masks not fitting, weight loss, or other surface changes, or by changes observed on an in-room imaging system. Because of an increased scatter contribution found in cone beam rather than fan beam images, cone beam CT scans, particularly kilovoltage cone beam CT, can result in cupping artifacts and erroneous electron density values.^33,38,66 Improved reconstruction algorithms and other methods to solve or reduce the effects of this problem are actively being developed.

Online Image-Guided Adaptive Radiotherapy

Online adaptation means that the imaging session and treatment session occur in the treatment room within a given treatment fraction. In Fig. 12-1 the time separation is shown to be on the order of minutes or less. Head and neck cancer is particularly amenable to online adaptation, because variable neck flexion can be significant and change the position of the lower neck and supraclavicular node targets with respect to the primary tumor and upper-neck nodal targets from that planned. Involved field prostate radiotherapy is also amenable to online adaptation because the prostate can move independently of the treated pelvic nodes, and therefore the two targets are in different geometric locations from each other to that planned. The rationale for online image-guided adaptive radiotherapy is shown in Fig. 12-6 (head and neck) and Fig. 12-7 (prostate).

FIGURE 12-6 • Rationale for online adaptive radiotherapy. Example of differential flexion between the computed tomography (CT) scan used for initial treatment plan (gray scale) and a cone beam CT scan (blue-gold blending). The anatomy matches in the lower cranium; however, the cervical spine is much straighter for the on-treatment scan.

FIGURE 12-7 • Rationale for online adaptive radiotherapy. Differential motion between the prostate (as detected by implanted markers) and the pelvic bony anatomy assumed to be fixed with respect to the location of the pelvic nodes for an involved-field intensity-modulated radiation therapy treatment in which both the prostate and pelvic nodes are simultaneously treated. The digitally reconstructed radiograph from the planning computed tomography scan is on the left; the electronic portal image is shown in the right.

The imaging system for online image-guided adaptive radio-therapy is typically an in-room CT scanner, which can be a fan beam or cone beam kilovoltage or megavoltage imaging system (all four combinations are available). To avoid introducing additional geometric errors, it is important that the time between the imaging procedures and treatment completion be minimized.