Molecular Imaging to Plan Radiotherapy and Evaluate Its Efficacy

Image Acquisition

It is important that patients undergo similar preparations before imaging and before receipt of each fraction of radiation therapy. For example, for abdominal tumors, fasting before imaging and treatment increases the likelihood that the stomach and bowels are the same size during treatment planning as they will be for each treatment session. For ¹⁸F-FDG PET imaging, fasting also helps to keep blood glucose levels within acceptable ranges for imaging. Blood glucose levels should be checked before ¹⁸F-FDG injection, and the scan should be rescheduled if levels are outside a predetermined range (2).

Molecular imaging for radiation therapy planning requires equipment and procedures additional to those typically used for diagnostic imaging. Unlike diagnostic images, images for radiation therapy planning will eventually need to be coregistered with treatment-planning CT images. Reliable target delineation is contingent on the accuracy of registration between the functional image and the planning CT image. As such, molecular imaging acquisitions for radiation therapy planning need to follow principles similar to those recommended for planning CT (or simulation CT) acquisitions (6).

The patient should be positioned and immobilized during image acquisition in the same manner as during receipt of each fraction of therapy. Because flat-top couches are used for radiation therapy delivery to ensure repeatable positioning, a flat-top table should be placed on top of the patient table. Any immobilization devices, padding, or bolus material should be in place during the scan. For nuclear medicine, placement of positioning equipment outside the transaxial imaging field of view but within the scanner bore should be avoided, as it can lead to attenuation artifacts (7). Scanners with large bores are preferred to accommodate the extra equipment. Positioning for functional MRI can be more challenging, depending on the tumor site (8). For example, for head and neck cancer, personalized masks are often used to ensure reproducible positioning and immobilization, but these masks often do not fit inside standard head coils. Furthermore, MRI bores are often smaller than CT bores, potentially limiting the use of some positioning devices. In these cases, creative solutions may be necessary (9). In addition to the internal positioning lasers included with imaging systems, it is preferable to have mobile external lasers for patient marking/tattooing and positioning, similar to those used during CT simulation. The positioning of the lasers should be precise and maintained at a known spatial relationship to the image center, so that patients can be marked in a way that will ensure repeatable positioning at subsequent treatment sessions. The alignment of the external laser should be tested for accuracy; such quality assurance tests have been described by the American Association of Physicists in Medicine (AAPM) task group 66 (6).

For PET/CT imaging, there are different options for acquiring the CT component of the scan. Attenuation and scatter correction for PET image reconstruction can typically be performed using either low-dose CT or high-quality diagnostic CT. The preference is to acquire a single high-quality CT scan together with the PET scan, which can then be used for treatment planning. In this case, no additional image registration is needed to align the PET image and the planning CT image. Treatment-planning CT, however, is sometimes performed with intravenous contrast material. Unless corrected for, contrast-enhanced CT can cause the PET voxel values in regions of high contrast density to be increased when the scan is used for attenuation correction (10). In these cases (or when the planning CT scan has been acquired separately), an additional low-dose CT scan should be acquired together with the PET scan. This low-dose CT scan can be used for attenuation and scatter correction but is not suitable for treatment planning. If contrast-enhanced CT is to be performed during the PET/CT examination, it should occur after both the attenuation CT and the PET acquisitions. A promising alternative to PET/CT imaging, which will become more prevalent in the future, is PET/MRI, in which pseudo CT images are created and used for PET attenuation correction (11) and potentially also for radiotherapy planning (12). The challenge of pseudo CT generation is that the MRI intensity value of a single voxel cannot be uniquely mapped into the γ-ray attenuation coefficient. Although several methods to improve the accuracy of pseudo CT mapping (e.g., registered atlas CT) have been developed (11), these have yet to be thoroughly validated for use in radiation therapy planning.

Registration

When the planning CT image is not already registered to the functional images, the two image coordinate systems need to be aligned. Rigid registration algorithms rotate and translate two images to maximize the spatial similarity of their intensity values. Most modern image-analysis software packages include methods for performing rigid registration. For PET, the CT image from the PET/CT acquisition is typically registered to the planning CT image, and the resulting spatial transformation is then applied to the respective PET image. Voxel sizes for PET images are larger than CT voxel sizes and, depending on the software, may need to be up-sampled before PET can be used for target delineation. Unless nearest-neighbor resampling is used, this resampling step will alter the quantitative SUVs; therefore, quantitative image analysis for treatment response assessment should be performed before the PET image registration. For functional MRI, anatomic MRI sequences such as volumetric T1-weighted imaging should also be acquired during imaging so that MRI and CT images can be coregistered. Special software may be required to register anatomic MRI and CT images, as the imaging values for different tissues are inherently different between the two imaging modalities.

In many cases, patient positioning will not be identical in the molecular images and in the planning CT image. This issue can make rigid registration difficult and can occasionally cause algorithms to converge to strange solutions (results should always be checked). One option, if the software allows it, is to crop the template image, leaving only the regions of interest to be registered. Another option is to use deformable registration algorithms. Like rigid registration, deformable registration algorithms maximize the similarities between two images but allow for morphologic changes beyond rotation and translation. These algorithms ultimately warp the template image to match the reference image. There are several deformable registration algorithms, each with different similarity measures or regularization constraints. Because the resulting solution will differ depending on the algorithm used, an algorithm should not be arbitrarily chosen (13). Quality assurance processes for image registration have been proposed (14), and an upcoming report from task group 132 of the AAPM will further describe the proper use and quality assurance of image registration algorithms.

Motion

Patient motion during imaging causes the image to be blurred over the path of motion, elongating target volumes. Motion also affects image quantification, effectively reducing image intensity values (15). Certain motions, such as breathing, cannot be avoided during treatment and will affect imaging of the lungs, liver, esophagus, pancreas, breast, prostate, kidneys, and other organs. Lung motion amplitudes are case-dependent but usually range between 0 and 30 mm in the superior–inferior direction, with an average of around 10 mm (16).

The degree to which motion should be compensated for during imaging depends on the degree to which motion will be accounted for during treatment. In other words, motion management for imaging should be coupled to motion management strategies during radiation delivery. In the case of PET, the simplest strategy would be to apply no motion correction to PET images, as both PET imaging and radiation therapy delivery encompass all phases of respiratory motion. Likewise, slow CT imaging can be used to create a phase-averaged CT image (17). Not accounting for motion during planning and radiation therapy delivery, however, can result in an unnecessary dose to nearby healthy organs.

There are several methods to reduce or account for respiratory motion during imaging and radiation therapy delivery. A detailed description of these methods is beyond the scope of this review but has been published elsewhere (16). These methods include shallow or tidal breathing to limit motion amplitude, respiratory gating (16,18), and real-time tumor tracking (19,20). In any case, the type of image acquisition should be closely matched to the type of radiation treatment delivery (e.g., 4-dimensional [4D] PET/CT together with real-time radiation delivery tracking).

Uncertainties

Uncertainties in molecular imaging arise from technical factors (e.g., scanner calibration), physical factors (e.g., positron range), biologic factors (e.g., patient metabolism), and analytic factors (e.g., inconsistent image processing) (21,22). Some of these uncertainties can be controlled or minimized through careful study design and quality assurance procedures (Table 1 provides examples specific to ¹⁸F-FDG PET). Understanding the limitations and uncertainties of molecular imaging is necessary to its proper implementation in radiation oncology.

For PET imaging, one limitation is its spatial resolution. There are many factors that affect PET spatial resolution, but in general, PET-avid lesions with diameters less than 5–10 mm are unlikely to be discernible from background in a typical clinical PET image (23). Consequently, small lesions, some positive lymph nodes, sharp intratumor heterogeneities, and microscopic tumor extensions may not be resolved in a PET image. The limited spatial resolution of PET has implications for its use in target volume delineation and treatment response assessment.

Uncertainties in molecular image quantification will propagate into uncertainties in treatment response assessment. Quantitative imaging for treatment response assessment requires proper calibration of the scanner equipment (including the dose calibrator for nuclear medicine) at acceptance testing and during routine maintenance. A detailed quality assurance plan, including daily, quarterly, and annual calibrations, is necessary to ensure the reliability and consistency of the scanner equipment (24,25). For MRI, quality assurance procedures that minimize geometric distortion caused by gradient field nonlinearity or static field inhomogeneity are especially important. Even with properly calibrated equipment, however, biologic variability and statistical noise cause substantial uncertainties in molecular imaging. For example, test–retest studies with ¹⁸F-FDG PET imaging suggest that biologic and statistical uncertainties contribute to an overall uncertainty (coefficient of variation) in SUV_max of about 10%–15% (26,27). Even the position of a lesion relative to the detector array geometry can cause fluctuations in PET SUV for small lesions (28).

There are also uncertainties on how to best extract meaningful quantitative imaging biomarkers from molecular images. In PET, the semiquantitative SUV is easily calculated by normalizing the measured radioactivity concentration in an image voxel by the ratio of the patient’s injected activity to body weight—either total body weight or lean body mass. Both the SUV_mean and the SUV_max of a region of interest are typically reported. However, depending on the application and radiotracer used, SUV may not represent a reliable measure of the biologic process of interest and can be very sensitive to the conditions of the acquisition, reconstruction, and analysis. In these cases, dynamic imaging with kinetic analysis may be necessary to understand the tracer kinetics and extract the most biologically relevant parameters. Dynamic scans are much more challenging to perform and analyze than static scans.

In MRI, similar challenges in image quantification exist. For example, it is becoming increasingly common for apparent diffusion coefficient (ADC) from diffusion-weighted (DW) MRI to be measured in clinical examinations, but standardized protocols for acquisition and analysis of DW MR images are lacking, especially across vendors (29). This lack of standardization also applies to perfusion MRI, which models and calculates kinetic parameters relating to blood flow. Standards for appropriate uses and analyses of quantitative imaging biomarkers, including proper characterization of their uncertainties, are being developed by the Quantitative Imaging Biomarkers Alliance (30).

Quantitative PET image values are highly sensitive to the algorithm used to reconstruct the images (31). With iterative reconstruction algorithms (which have become standard in clinical PET imaging) and a growing library of image correction methods (e.g., point-spread-function modeling), there are hundreds of possible combinations of reconstruction parameters. Furthermore, the optimal reconstruction method may depend on the region of interest (e.g., chest vs. brain). It is therefore incumbent on an institution’s nuclear medicine physicians, physicists, or technologists to determine the reconstruction method that optimizes image quality for a particular PET scanner. This generally involves scanning a PET image-quality phantom, exploring several different reconstructions, and finding the set of parameters that offers the best trade-off between spatial resolution and image noise. Furthermore, as different PET scanners can have substantial differences in image quality and quantitative accuracy, image reconstruction parameters can be tuned such that the different PET scanners involved in a multicenter study produce comparable image quality. This process of scanner harmonization can improve the statistical power of multicenter clinical trials investigating imaging biomarkers (32).

Overall, many uncertainties in molecular imaging and in its application to radiation oncology can be minimized using a detailed and unambiguous protocol for acquisition and image analysis. One such protocol is the ¹⁸F-FDG PET imaging protocol developed by the Uniform Protocols for Imaging in Clinical Trials working group (33). Strict adherence to protocols ensures that for baseline and follow-up imaging, identical procedures will be followed for patient preparation, patient immobilization, radiotracer or contrast injection, acquisition settings, image reconstruction/corrections, image processing, and image analysis. Minimizing these quantitative imaging uncertainties will allow for more precise quantification of treatment response (34).

Manual Tumor Segmentation

Currently, the most common method for molecular imaging–based gross tumor volume delineation is manual segmentation. A prerequisite for all molecular imaging–based segmentation methods is an understanding of the normal biologic processes that will be measured in the images, so that normal tissue is not mistaken for diseased tissue during segmentation. Potential sources of false-positives and false-negatives should be well understood. For example, in the case of ¹⁸F-FDG PET, brown adipose tissue, granulous tissue, areas of inflammation, and muscle often have elevated levels of uptake and can be mistaken for tumor (1,36). PET images should be interpreted by nuclear medicine physicians and MR images by radiologists before being used for radiation therapy purposes so that, if necessary, previously unknown lesions can be detected and potential image artifacts can be identified.

After image registration, the molecular and the anatomic images should be fused by the visualization software (37). All available information should be used for defining the target volume. Manual segmentation methods, however, are known to suffer from inter- and intraobserver variability in target volume delineation. To reduce this variability, it is recommended that institutions develop a strict delineation protocol that describes exactly how the tumor should be visualized (e.g., window level and color settings) and details any other parameters that can affect the physician’s choice of contouring (38). Because of the significant inter- and intraobserver variability, manual segmentation is generally inferior to automatic and semiautomatic segmentation and as such is discouraged. However, it will take some time for the manual segmentation to be entirely replaced by automatic and semiautomatic segmentation tools, which still require more rigorous clinical validation.

If DW MRI is used for tumor segmentation, special attention should be paid to the potential for spatial distortions in the ADC maps. Studies have found that substantial geometric distortions can occur in DW MR images, even on the order of centimeters (39). Such distortions, both global and local, can also occur for other imaging sequences, including sequences used for perfusion MRI. If not accounted for, these distortions can substantially affect the accuracy of the target volume delineation.

Automatic and Semiautomatic Tumor Segmentation

Computer algorithms that automate tumor volume segmentation based on molecular images can improve workflow and reduce physician variability in target definition. In PET, the most common method for automatic segmentation is thresholding based on SUV. There are different ways by which thresholding can be performed. A threshold based on SUV_max is the most commonly used method and usually the most simple. The threshold value for ¹⁸F-FDG PET is often set at around 40% of SUV_max but varies from 20% to 50%. Other types of thresholding include using a fixed SUV threshold (e.g., SUV ≥ 2.5 (40)) or using a background-subtracted threshold (41). A clear disadvantage of thresholding is the uncertainty about which threshold is optimal, which can vary depending on the scanner hardware, reconstruction and acquisition parameters, presence of motion, tumor size, and patient biology (42). Furthermore, thresholding cannot distinguish normal uptake from disease, and results will need to be corrected manually, preferably by an experienced nuclear medicine physician.

Numerous advanced automatic and semiautomatic PET segmentation algorithms have been developed (43,44). Broadly, these can be categorized as iterative/adaptive methods, statistical modeling methods, machine learning–based methods, and image filter–based methods (e.g., using gradients or texture features). Currently, it remains unclear which methods perform best and how best to assess the performance of different algorithms. Various performance evaluation methods have been developed for testing segmentation algorithms (44,45). It is recommended that institutions thoroughly evaluate segmentation algorithms, including any tuning parameters, and benchmark their performance against both phantom and clinical images before clinical implementation. Even when automatic and semiautomatic tools are used, manual review of the contours is highly recommended.

For brain tumor segmentation, the utility of ¹⁸F-FDG PET is limited by high background levels of ¹⁸F-FDG uptake in healthy brain tissue. Other radiotracers have been shown to perform better in the identification and delineation of brain tumors. For example, the amino acid radiotracers O-(2-¹⁸F-fluoroethyl)-l-tyrosine and l-[methyl-¹¹C]-methionine are taken up by cancer cells because of overexpression of amino acid transporters (46,47). 3,4-dihydroxy-6-¹⁸F-fluoro-l-phenylalanine (¹⁸F-DOPA) is another amino acid PET radiotracer that has produced promising results (48). MRI, however, remains the standard for delineation of most brain tumors. Other tumor sites may also benefit from the use of nonstandard PET radiotracers for target definition, such as 16α-¹⁸F-fluoro-17β-estradiol in breast cancer, and somatostatin-analogs DOTATOC and DOTATATE for neuroendocrine tumors.

Nodal Involvement

The ability of PET imaging to detect involved lymph nodes that appear benign on CT images has been demonstrated (49). However, its improvement over CT imaging for nodal detection is dependent on the tumor site. A series of review papers has been published on PET for radiation treatment planning for various tumor sites, including the brain (50), head and neck (51), lung (52), gastrointestinal tract (53), prostate (54), and cervix/endometrium (55). For some situations, such as mediastinal lymph node staging for lung cancer, the sensitivity of ¹⁸F-FDG PET/CT appears to be superior to CT alone at detecting involved lymph nodes (52). In others, such as colorectal cancer, PET does not perform as well (56). It remains to be seen whether PET/MRI scanners will improve nodal detection, particularly in some anatomic sites such as the abdomen.

Tumor Boosting

One appealing future application of molecular imaging in radiation therapy planning is in the delineation of biologic tumor subvolumes for dose escalation. This method has been termed biologically conformal radiation therapy, or dose painting (57). The rationale for dose painting comes from the fact that tumors are biologically heterogeneous in composition and often show nonuniform patterns of response to radiation therapy (58,59). Molecular imaging may be able to identify spatial patterns of radioresistance, which can be used to guide and shape the dose distribution. Indirect evidence of this principle comes from studies that have found molecular imaging biomarkers to correlate with patient outcome after radiation therapy (60,61). Possible biologic targets include tumor hypoxia (measured with PET hypoxia radiotracers or with dynamic contrast-enhanced [DCE] MRI), cellularity (measured with ¹⁸F-FDG PET or DW MRI), and others. Identifying effective and robust imaging-based targets for dose painting is currently an ongoing research problem (61,62). Dose painting can be applied either on a region-of-interest level (dose painting by contours) or on a voxel level (dose painting by numbers), in which each tumor voxel receives a unique dose prescription proportional to the voxel’s image intensity value (63). Studies have demonstrated the feasibility of creating and delivering dose-painting plans—both subvolume boosting and voxel-based dose painting—using existing clinical software and therapy systems (64).

Dose painting has yet to be validated as an effective treatment option, but clinical studies have begun to investigate the efficacy of dose painting. Studies have been performed on the brain using ¹⁸F-FDG (65) and O-(2-¹⁸F-fluoroethyl)-l-tyrosine PET (66), on head-and-neck tumors using ¹⁸F-FDG PET (67–69), on prostate tumors using ¹⁸F-fluorocholine PET (70), and on lung tumors using ¹⁸F-FDG PET (71). Further clinical studies are ongoing.

Normal-Tissue Sensitivity

The potential for using molecular imaging to discriminate radiosensitive tissue from radioresistant tissue in tumors may be equally valuable when applied to the surrounding normal tissue. Certain regions in an organ, particularly the lung, appear to be more sensitive to radiation therapy than other regions in the same organ. Or some regions may be more functional than other regions and therefore more important to spare. For example, in ¹⁸F-FDG PET imaging of the lungs, studies have found that regions of uptake before radiation therapy are more likely to experience radiation-induced toxicity after radiation therapy (72). Ventilation imaging may identity lung regions that are blocked—either by obstructive lung disease or by tumor burden—and may contribute less to overall lung function than healthy, functioning lung. Information on functional lung can be acquired through a variety of imaging methods. For MRI, hyperpolarized gases or gadolinium aerosol can be used in combination with serial imaging to produce 4D images of lung ventilation (73). Ventilation maps can also be derived from 4D CT imaging, in which voxel Hounsfield units correlate with the fraction of air in the voxel volume (74). In SPECT, ^99mTc-labeled macroaggregated albumin can also be used to create 3D ventilation maps (75). The feasibility of incorporating lung ventilation maps into treatment planning for conformal avoidance has been demonstrated (76). Although promising, the clinical benefit of using functional lung images for conformal avoidance has yet to be validated.

Overall, there are many potential uses of molecular imaging in target definition for radiation therapy. The various methods by which molecular imaging can be used for target definition and their respective degrees of complexity are outlined in Table 2. Considering the extensive ongoing clinical research and continuous technologic advances, it is likely that the methods presently considered to be advanced will soon become routine practice.

Prognostic Role of Molecular Imaging in Radiation Therapy

High clinical availability and wide applicability are why ¹⁸F-FDG PET/CT is most commonly investigated as a prognostic and predictive imaging biomarker in radiation oncology. Relative changes in ¹⁸F-FDG SUVs are commonly investigated for being representative of clinical response, as suggested in the guidelines of the European Organization for Research and Treatment of Cancer and the National Cancer Institute, as well as PERCIST, for the measurement of treatment response (78–80). These guidelines also provide criteria for classifying patients into various response categories (complete response, partial response, stable disease, progressive disease). The response thresholds have not been generally validated in larger clinical trials and therefore serve mostly as recommended guidelines for standardized reporting. In addition, ¹⁸F-FDG PET/CT–based therapy response assessment requires proper selection and control of some factors that are highly important for therapy response assessment. One of them is the timing of sequential ¹⁸F-FDG PET/CT imaging for response assessment. Because of the dynamic nature of inflammatory and tumor-response processes, the timing of imaging relative to the treatment schedule has a profound effect on how early it is possible to assess treatment response. Some key studies representing the current evidence for various tumor types are highlighted as follows.

Cervical Cancer

In locally advanced cervical cancer, the generally accepted method of response assessment is ¹⁸F-FDG PET/CT 3 mo after the completion of concurrent chemoradiotherapy; reliable long-term prognostic information can be obtained (96). Despite the high predictive value of ¹⁸F-FDG PET/CT–based response assessment, relapses remain problematic. Therefore, other disease characteristics such as bulky tumors (>5 cm), high stage (>IIB), or pelvic or paraaortic lymph node metastasis can be used to direct more aggressive treatment or adjuvant chemotherapy regimens (97). In cervical cancer, apart from being used for posttherapeutic response assessment, ¹⁸F-FDG PET/CT has been investigated for monitoring of response during radiotherapy. Despite some promising results, future clinical trials are warranted (98).

Although some limitations of ¹⁸F-FDG PET/CT–based therapy response assessment in radiation oncology can be overcome by careful design and implementation of imaging protocols, the inherent limitation caused by radiation-induced inflammation makes interpretation of treatment-response results difficult. To overcome this limitation, the focus of PET imaging is shifting to more tumor-specific characteristics (e.g., cellular proliferation, hypoxia, angiogenesis, and apoptosis). Currently, the most established of these is imaging of cellular proliferation using ¹⁸F-FLT PET, a surrogate of cellular proliferation. The characteristics of ¹⁸F-FLT make it suitable for static PET imaging, with a reasonable uptake period and dynamic PET imaging that can be subsequently analyzed for kinetics (99,100). The potential of ¹⁸F-FLT PET/CT for early assessment of chemotherapy has been demonstrated in a variety of cancers (101–103). In the context of radiotherapy, ¹⁸F-FLT PET/CT first proved its sensitivity to ionizing radiation in preclinical tumor models (104,105). After promising results from ¹⁸F-FLT PET/CT–based assessments of radiation response in these models, an increasing number of studies have shown that decreased ¹⁸F-FLT uptake early (as early as 5 d/10 Gy) in the course of radiotherapy or chemoradiotherapy is a strong indicator of long-term outcome in a large-animal preclinical model (106) and humans (107–110) (Fig. 2). Although the data on ¹⁸F-FLT PET/CT response assessment in radiation therapy are still emerging, its potential is indisputable, and it should be considered an alternative to ¹⁸F-FDG PET/CT. One alternative PET surrogate marker of cellular proliferation is ¹¹C- and ¹⁸F-choline, which appears relatively sensitive for detecting recurring prostate cancer (111). Another is l-[methyl-¹¹C]-methionine, which may be useful for late response evaluation in brain cancer (112).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2.

Multimodality multitracer molecular images of ⁶¹Cu-diacetyl-bis(N4-methylthiosemicarbazone) (Cu-ATSM) PET, ¹⁸F-fluorothymidine (FLT) PET, and DCE CT in patient before bevacizumab monotherapy (time point 1), after 3 wk of bevacizumab (time point 2), and after 1–2 wk of chemoradiation therapy (time point 3), indicating potential of molecular imaging to assess complex radiation treatment regimens. (Reprinted with permission of (116).)

Apart from cellular proliferation imaging, hypoxia is another tumor-specific characteristic that has been successfully imaged and used for radiotherapy response assessment in clinical studies. Although tumor hypoxia is widely recognized as a negative prognostic factor in radiation oncology, hypoxic tumor subvolumes often change during radiotherapy (113,114). Treatment outcome may correlate more with hypoxia assessed during radiotherapy than before radiotherapy (115) (Fig. 2 (116)).

Similar to PET-based radiotherapy response assessment, SPECT imaging with ^99mTc-HL91 radiotracer allows for detection of tumor hypoxia and has significant predictive power for tumor response and patient survival (117). SPECT imaging with the ^99mTc-hydrazinonicotinamide-rh-annexin V radiotracer also allows for evaluating apoptosis in tumors. Apoptotic SPECT imaging has shown radiation dose–dependent uptake in the parotid glands within 2 d after the first course of chemoradiotherapy (6–8 Gy) in head and neck cancer patients; such uptake was indicative of early apoptosis during radiotherapy (118).

In addition to nuclear medicine molecular imaging techniques, MR-based molecular imaging has long been investigated for early radiotherapy response assessment. Research has focused primarily on DCE and DW MRI. For example, increased perfusion early during the course of radiotherapy (within the first 2 wk) assessed with DCE MRI has been shown to be a strong predictor of tumor regression and local control in cervical cancer treated with conventional radiotherapy (119). Also, pretherapy DCE MRI perfusion, permeability, or blood volume measurements have been associated with radiation treatment response in rectal cancers (120) and brain cancers (121). Water diffusion in the tissue can be assessed with DW MRI through the ADCs. Low ADCs measured before radiotherapy have been shown to be associated with response to radiotherapy for brain lesions (122), cervical cancer (123), and head and neck cancer (124). Similarly, increases in ADCs measured during radiotherapy in patients with brain (122), cervical (123), liver (125), rectal (126), and head and neck cancer (124) have been shown to be associated with a favorable outcome. Obstacles toward more successful use of DCE or DW MRI for treatment response assessment includes the unknown optimal timing of DCE MRI (which is most likely disease-dependent (127)), and the lack of reproducibility in ADCs between vendors and institutions (128).

Spatially Resolved Treatment Response Evaluation

Solid cancers have intrinsic spatial heterogeneity in biologic characteristics, which limits the efficacy of point measurements such as biopsy-based molecular assays or point measures derived from PET images (e.g., SUV_max). Methods to incorporate spatial heterogeneities into treatment response assessment are an active area of research. An obvious approach to incorporate the intratumoral heterogeneity into treatment response would be to evaluate spatial therapeutic response. Therapeutic response can be evaluated at a voxel level as absolute change in the PET values (129,130) or as relative change from the baseline image (131). Evaluation of relative changes on a voxel level is unreliable for tumor subvolumes with a very low baseline PET value, for which relative changes may be high despite both baseline and posttherapy PET image values within the background range. Evaluation of absolute change in PET values on a voxel level (129,130) might be less sensitive to cases of very low baseline PET values. An evaluation of the therapeutic response on a voxel level is also strongly affected by the selection of registration algorithm (131).

Intratumoral heterogeneity can also be incorporated into treatment response assessment by the extraction of various textural features from molecular images (popularly termed radiomics). These features are subsequently used for therapy response assessment instead of or in addition to point SUV measures (i.e., SUV_peak and SUV_max) or overall SUV measures (i.e., SUV_mean and SUV_total). Although the use of texture analysis in molecular imaging is still in its infancy, some successful applications in radiation oncology have already been published. Various textural features that were extracted from pretreatment ¹⁸F-FDG PET images have been correlated to treatment outcome in breast, cervix, esophageal, head and neck, and lung cancer tumors (132). Changes in textural features of ¹⁸F-FDG PET images during and after chemoradiotherapy have also been significantly correlated to time to progression and survival in rectal cancer patients (133).

A common technical limitation of all approaches for incorporating intratumoral heterogeneity into treatment response assessment is the influence of image acquisition and reconstruction parameters on image heterogeneity measures. Image acquisition and reconstruction intrinsically implement some smoothing and possibly other distortions of the data, in comparison to the real data, the degree of which depends on the imaging hardware and software, and also on the acquisition and reconstruction settings. Therefore, sequential imaging for voxel-based treatment response assessment and for extraction of imaging features would have to be performed on the same scanner with the same acquisition and reconstruction settings. Alternatively, different scanners could produce comparable results if they were carefully harmonized. However, interchangeability of textural features from different but carefully harmonized PET/CT scanners has yet to be proved.

Adaptation

The main goal of early molecular-imaging response assessment during radiation therapy is to allow therapy to be modified to improve the clinical outcome—the so-called biologically adaptive radiotherapy. The radiation is adapted during therapy to match corresponding changes in tumor physiology as derived from molecular imaging. Modeling studies have demonstrated potential for significant improvement of clinical outcome using biologically adaptive radiotherapy (134), whereas the clinical evidence has yet to be provided.

Although ¹⁸F-FDG PET/CT is the molecular imaging technique with the highest clinical acceptance and availability (68,135), treatment adaption based on ¹⁸F-FDG PET/CT is hampered by the inflammatory signal after radiotherapy. The potential use of ¹⁸F-FDG PET/CT for adaptive tumor dose escalation or normal-tissue sparing has been investigated in patients with NSCLC (135) and even coupled with dose painting in patients with head and neck cancer (68). Both studies found significant changes in target volumes based on mid-therapy ¹⁸F-FDG PET/CT images, warranting replanning of the treatment target.

In addition to ¹⁸F-FDG PET, more specific (e.g., cellular proliferation, hypoxia) imaging biomarkers can be used for radiotherapy adaption. As a rapid decrease in ¹⁸F-FLT uptake has been demonstrated as early as 5 d/10 Gy in the course of radiotherapy (107–110), early radiotherapy adaption based on ¹⁸F-FLT PET/CT imaging appears particularly attractive. The clinical utility of early radiotherapy adaption based on ¹⁸F-FLT PET/CT has yet to be demonstrated.

Normal-Tissue Response Evaluation

The main application of molecular imaging for normal-tissue response evaluation is the detection of radiation-induced injuries to these tissues. Monitoring induction, resolution, and mitigation of radiation-induced toxicity is essential in the development of clinically successful strategies to preserve normal tissues. Molecular imaging can be particularly useful to study and monitor these changes.