Functional imaging for radiotherapy treatment planning: current status and future directions-a review.

In recent years, radiotherapy (RT) has been subject to a number of technological innovations. Today, RT is extremely flexible, allowing irradiation of tumours with high doses, whilst also sparing normal tissues from doses. To make use of these additional degrees of freedom, integration of functional image information may play a key role (i) for better staging and tumour detection, (ii) for more accurate RT target volume delineation, (iii) to assess functional information about biological characteristics and individual radiation resistance and (iv) to apply personalized dose prescriptions. In this article, we discuss the current status and future directions of different clinically available functional imaging modalities; CT, MRI, positron emission tomography (PET) as well as the hybrid imaging techniques PET/CT and PET/MRI and their potential for individualized RT.

Radiotherapy (RT) has, in the past two decades, been subject to a number of technological innovations. Modern RT techniques such as intensity-modulated RT, volumetric modulated arc therapy, as well as proton and heavy ion therapy are extremely flexible, benefiting from a high number of degrees of freedom in both the treatment planning process and in radiation delivery. As a consequence, in RT today, high levels of dose coverage of the tumour volume can be realized hand in hand with organ at risk sparing. Moreover, this flexibility gives room for escalating the radiation dose to more radioresistant areas of the gross tumour volume (GTV) with the aim of increasing tumour control probability without increasing the side effects of RT.

Modern functional imaging techniques, such as functional CT, MRI or positron emission tomography (PET) allow to visualize surrogates of a variety of pathophysiological characteristics of tumour tissue, such as metabolism, proliferation, hypoxia, perfusion etc. Consequently, integration of functional imaging seems to be very promising for individualized RT treatment planning.[1]

Functional CT techniques such as dynamic contrast-enhanced CT (DCE-CT) as well as dynamic contrast-enhanced MRI (DCE-MRI) allow us to measure surrogates of tissue perfusion. Diffusion-weighted MRI (DW-MRI) estimates the degrees of freedom that water molecules have to travel in tissue. Thus, in patients with cancer, DW-MRI is used to determine regions with abnormal extracellular space, which is used as an estimate of tumour cell density. PET imaging allows us to measure the specific uptake of radiotracers and depending on the tracer used, can allow estimation of a number of different physiological properties of tumour tissue, such as metabolism, hypoxia or proliferation which are known to be key factors for cancer treatment.[2]

Consequently, integration of functional CT or MRI and molecular PET information to (re)direct radiation dose seems to be a powerful strategy to improve modern RT planning and to overcome biology-driven radiation resistance.[2,3] In the past years, different strategies for using functional information have been proposed. First, additional complementary information from functional CT, MRI or PET imaging has been shown to improve target volume delineation in order to more accurately direct the radiation dose to the tumour region.[2-4] Furthermore, higher radiation doses can be prescribed to a subvolume of the tumour, which is more radiation resistant as identified by functional imaging. This concept is called dose painting.[5] Moreover, the functional imaging-guided increase in dose may be distributed inhomogeneously throughout the whole tumour volume according to activity distributions or, more generally, to parametric maps derived from functional imaging using a dedicated prescription function.[6,7]

In this article, the current status of functional imaging using CT, MRI and PET will be reviewed with respect to its prognostic value for RT outcome and to the possibility of potential integration into RT treatment planning. Furthermore, current developments and future directions that may be relevant for RT treatment adaptation in the future are discussed.

CT IMAGING TECHNIQUES

For several decades, CT has been the standard to base RT treatment planning and dose calculation on in order to achieve the highest dosimetric accuracy.[8] Three-dimensional high-resolution imaging to estimate linear X-ray attenuation coefficients has become a major prerequisite for high-precision RT planning and delivery. Although, research on the use of other additional imaging techniques is currently evolving, there have also been major developments in CT during recent years. For example, fast and accurate iterative reconstruction techniques are now clinically available, which allow accurate quantitative imaging of mass densities even in the presence of metal implants.[9]

A recent CT technique with special technical requirements is dual-energy CT (DECT).[10] DECT offers a more specific tissue classification[11] and therefore allows for more accurate Monte Carlo dose calculations in the RT planning process.[12-14] This may be especially interesting in the advent of particle therapy such as proton or carbon ion treatment planning and the respective dose calculations. Several studies have described and controversially discussed the potential of DECT to reduce metal artefacts, which is also very important for classical X-ray-based RT.[15-17]

Moreover, DCE-CT imaging allows us to estimate tissue perfusion by injecting iodine-based contrast agents in addition to repetitive CT imaging for approximately 40 s. Thus, DCE-CT has the potential to determine functional parameters such as blood flow, tissue perfusion and others from the contrast agent dynamics. For RT treatment planning, the availability of DCE-CT data can have enormous advantages as studies have shown that perfusion CT may improve the accuracy of RT target volume delineation.[18-20] In addition, changes in perfusion parameters measured with MRI or CT during treatment have been shown to potentially give valuable information for monitoring therapy efficacy and success.[21] A major advantage of using DCE-CT to estimate tissue perfusion lies in the linear relationship between signal intensity and contrast concentration compared with DCE-MRI, where signal-to-contrast conversion is very challenging.

A recently published study proposed a parametric method for automatically analysing perfusion information from volumetric DCE-CT data on a voxel basis in order to make this technology available as a biomarker for RT.[22] The study showed that parametric voxel-based analysis of DCE-CT data resulted in greater accuracy and reliability in measuring changes in perfusion CT-based kinetic metrics, which have the potential to be used as biomarkers in patients with brain metastasis.

DW-MRI, DCE-MRI AND MRS FOR RT OUTCOME PROGNOSIS AND TREATMENT PLANNING

In recent years, an increasing number of research projects have been initiated to explore the potential of DW-MRI, DCE-MRI and also MR spectroscopy (MRS) to assess their prognostic value, to improve RT treatment planning and delivery and to explore potential biomarkers for personalized RT in the future.

DCE-MRI—similarly to DCE-CT—allows us to assess the vascular properties of tissues by injecting a gadolinium-based contrast agent and acquiring a dynamic set of MR images.[23] Perfusion properties are then derived from the time-dependent signal by applying dedicated compartment models that are fitted to the signal curves. A recent study showed that by using DCE-MRI data, it was possible to identify subvolumes of brain metastases that were associated with therapy response.[24] In this study, the major component to determine response to therapy using principal component analysis was the area under the DCE curve. A number of different clinical studies have shown that DCE-MRI information has prognostic value with regard to RT outcome in head and neck cancer (HNC).[25-27] In a study by Halle et al,[28] the prognostic impact of DCE-MRI parameters in cervical cancer acquired prior to radiochemotherapy combined with global gene expression data was explored. In this study, DCE-MRI was successfully correlated with a hypoxia gene signature that also showed prognostic impact in an independent validation cohort of 109 patients. The results of this project are a first step towards decrypting the molecular basis of an aggressive hypoxic phenotype and suggest the use of DCE-MRI to non-invasively identify patients with hypoxia-related chemoradioresistance.[29] Furthermore, it has been shown that DCE-MRI data can successfully be used for response assessment after completion of therapy.[30] However, acquisition and also analysis and interpretation of DCE-MRI data are technically extremely challenging. Also, the choice of the correct model to be used for kinetic analysis has to be made carefully.[31]

By contrast, DW-MRI offers a contrast-free method to assess the diffusion properties of water in tissue. This functional MRI method allows us to quantify voxel-based diffusion coefficients, more exactly, signal-derived maps of the apparent diffusion coefficient (ADC) that differ from one tissue to another. Thus, DW-MRI seems to be a powerful method when it comes to identifying the diffusion properties of a tumour mass. A recent histological study showed that the ADC signal value was significantly correlated with histological characteristics of HNC, such as cellularity, stromal component and nuclear–cytoplasmic ratio.[32] Several studies have recently shown the prognostic potential of ADC to predict treatment outcome after RT in different tumour entities.[33-36] In a recent Belgian study, the prognostic value of pre-treatment ADC in a large patient population with HNC was assessed and integrated into a multivariable prognostic model with the aim of estimating the individual patient's outcome prognosis.[37] The study revealed that pre-treatment ADC value derived from high b-values is an independent prognostic factor in HNC and increases the performance of a multivariable prognostic model when used in addition with known clinical and radiological variables. Another study investigated texture analysis of ADC images in conjunction with multivariate analysis to identify pre-treatment imaging biomarkers.[38] By combining principal component analysis and texture analysis, ADC texture characteristics were identified, which seem to hold pre-treatment prognostic information, independent of known clinical prognostic factors. Further clinical trials confirmed the high value of DW-MRI for response assessment early after completion of RT.[39-41]

A French group investigated the prognostic value of MRS imaging and assessed its impact on local tumour control in glioblastoma (GBM).[42,43] Recently published results indicate that the lactate-to-n-acetyl-aspartate ratio (LNR) measured with MRS can discriminate between tumour-associated and normal LNR values with high sensitivity and specificity. Voxel areas presenting with low LNR values were spatially not colocalized with other MRI-defined volumes derived from contrast enhancement, central necrosis and fluid-attenuated inversion recovery abnormality before RT.[42] As a consequence, pre-RT MRS in GBM seems to be able to detect tumour areas that are likely to relapse. Thus, MRS lactate imaging may be a future tool to define additional biological target volumes for dose painting.

PET FOR RT TARGET VOLUME DELINEATION AND ASSESSMENT OF FUNCTIONAL TUMOUR CHARACTERISTICS

PET has been established as a major source of metabolic and functional information for use during the RT planning process. PET imaging using the metabolic tracer fluorine-18 fludeoxyglucose (18F-FDG) is now a particularly important routine imaging modality not only for tumour grading and staging but also for accurate target volume delineation. Additional information on tumour geometry and extension may lead to better treatment outcomes. A number of studies so far have shown the benefit of 18F-FDG PET imaging for staging and definition of tumour extension.[44,45] However, a study on 90 patients with oesophageal cancer reported that the value of 18F-FDG PET/CT for target volume delineation seems to be limited.[46] This study aimed at determining the proportion of locoregional recurrences that could have been prevented if RT planning for oesophageal cancer was based on PET/CT instead of CT. The result was negative. Further studies have assessed the prognostic value of dedicated PET-based parameters such as the maximum standardized uptake value (SUV) for overall survival and local tumour control.[47-50] As a consequence, robust and accurate delineation of the 18F-FDG PET-positive tumour lesion is crucial.[51] In recent years, a large number of studies and methodological research projects were carried out in order to develop and validate automatic and semi-automatic algorithms for accurate and robust delineation of RT target volumes based on 18F-FDG PET.[51-62] Validation of new segmentation algorithms in terms of accuracy and robustness is of crucial importance for the potential clinical application of (semi-)automatic PET-based contouring.[63,64] So far only a few clinical trials have been carried out in which dose escalation was prescribed on an 18F-FDG PET avid area inside the GTV.[65-67] Currently, two multicentre trials are testing the potential of redistributing the radiation dose to the metabolically most 18F-FDG PET avid part of the tumour in non-small-cell lung cancer (NSCLC)[65] and also in HNC.[66]

PET imaging is not only beneficial for improving RT treatment planning in terms of target volume definition, it also has the potential to visualize functional and biological characteristics of tissue, such as tumour hypoxia. Tumour hypoxia was known for decades to be a key factor driving individual radiation resistance.[68-70] As a consequence, non-invasive measurement of tumour hypoxia may be an important molecular marker for potential future RT adaptations.[71] PET imaging allows for the detection of tumour hypoxia using different radiolabelled tracers, such as 18F-fluoromisonidazole (18F-FMISO),[72-75] 18F-fluoroazomycin arabinoside (18F-FAZA)[76-79] or 18F-flortanidazole (18F-HX4)[80,81] among other less known tracers. A number of different studies have shown that a positive detection of tumour hypoxia using 18F-FMISO or 18F-FAZA PET is associated with a high risk of locoregional failure of chemoradiotherapy in NSCLC as well as in HNC.[72,74,77] In addition, some studies have investigated the optimal time point of hypoxia imaging during the course of RT with the aim of identifying an ideal time point for potential therapy adaptation.[73,74,77,79] Zips et al[74] analysed 18F-FMISO PET data for 25 patients with HNC examined before the start of RT as well as at Weeks 1, 2 and 5 during treatment. Similarly, Bollineni et al[77] report a study on six patients with NSCLC and six patients with HNC imaged with 18F-FAZA PET before chemo-RT and in treatment Weeks 1, 2 and 4. Both studies reported that hypoxia PET data acquired during the second week of treatment show the best correlation with observed treatment outcome and are thus most suitable and more reliable to base a potential treatment adaptation with, for example, dose painting on. However, quantitative hypoxia PET imaging is crucial for individualized RT alterations as well as for comparability of data in multicentre studies.[82] So far, a variety of different concepts for the quantification of tumour hypoxia based on PET imaging have been used in different studies ranging from visual interpretation,[72] tumour-to-background-based thresholding[79] and assessment of maximum or peak SUVs[74] to kinetic analysis of dynamically acquired hypoxia PET data.[83-86] Here, a prerequisite for a potential future hypoxia-based RT intervention is a profound validation of hypoxia quantification measures based on PET. Furthermore, the different studies published so far all suffer from low patient numbers and a high variety of the respective imaging protocols, using image acquisition times ranging from 70[83] to 240 min[74] post injection (p.i.) of the tracer. Simulation experiments and the first clinical results have shown that image contrast improves with increasing time intervals between hypoxia tracer injection and PET image acquisition, which is owing to the slow diffusion of tracer in the tissue.[87-89] Consequently, hypoxia PET image acquisition is recommended at 3–4 h p.i. for all nitroimidazole-based tracers. Direct comparison of the different clinically available hypoxia PET tracers has not been performed in a clinical setting so far. Two recent pre-clinical studies have compared the three tracers 18F-FMISO, 18F-FAZA and 18F-HX4; presenting highly inconclusive results about advantages and disadvantages and also the selection of an optimal tracer.[90,91]

A further potentially very interesting PET tracer for RT adaptation and follow-up imaging is the proliferation marker 18F-fluorothymidine (18F-FLT).[92,93] Initial clinical studies have shown that a change in 18F-FLT uptake early during RT is a strong indicator for long-term outcome in HNC and NSCLC.[92,93] 18F-FLT PET may thus be a potential imaging biomarker to guide personalized patient management and treatment modifications during an early phase of treatment.

For other tumour entities, which physiologically do not present with increased glucose metabolism, such as prostate, new tracers are currently being developed to improve diagnosis and therapy. Recently, a highly specific tracer for the diagnosis of prostate tumours, the prostate-specific membrane antigen (PSMA) ligand (68Ga)HBED-CC-PSMA was investigated.[94] Initial clinical studies using PSMA PET show very promising results in terms of a high sensitivity and specificity of PSMA PET for the identification of intraprostatic tumour foci, which would be a prerequisite for PET-based focal dose escalation in prostate cancer.[95] For RT target volume delineation in brain tumours, studies using amino acid PET tracers, such as 18F-fluoro-ethyl-tyrosine PET or carbon-11 methionine PET have shown the potential to visualize tumour areas that do not seem to be detected via MRI and could therefore yield additional, complementary information.[96-98]

HYBRID IMAGING MODALITIES AND MULTIPARAMETRIC FUNCTIONAL IMAGING

In addition to combined PET/CT[99] that has been the clinical standard for approximately 15 years now, new hybrid imaging technologies such as PET/MRI[100] have been developed in the past few years. In contrast to PET/CT, combined PET/MRI is still a matter of technological and also clinical research.[100] However, hybrid imaging modalities allow acquisition of two or more molecular, functional and anatomical image data sets either at the same time or successively in the same patient position. As a consequence, these scanners have the potential to be used for assessing different functional or biological characteristics of a tumour with only one examination, which might be highly interesting for RT personalization in the near future. Multiparametric functional imaging including new methodologies for large-scale data handling and analysis is an evolving field in RT research. A number of studies have investigated common features and correlations between different functional imaging modalities with the aim of increasing the accuracy in target volume delineation[101-103] or detection of regions with increased radiation resistance.[104-106]

Houweling et al[102] have investigated whether 18F-FDG PET and DW-MRI identify the same or different targets for dose escalation in the GTV of patients with HNC. The study found that these two imaging modalities contain different information, resulting in different RT targets, which hints at the complimentary nature of the measured biological information. Groenendaal et al[103] developed a logistic regression model for voxel-by-voxel prediction of prostate tumour presence, validated via pathology. The model defines different risk levels for tumour presence, which were then used as a basis for improved tumour delineations for focal boost therapy. Another study by van Elmpt et al[104] analysed differences in vasculature properties within NSCLC tumours measured by DCE-CT and metabolic information from 18F-FDG PET/CT. In this study, no direct correlation was observed between 18F-FDG PET and DCE-CT. Lambrecht et al[106] investigated the use of 18F-FDG PET/CT before, during and after chemo-RT and DW-MRI before chemo-RT for the prediction of pathological response in patients with rectal cancer. The results of this study showed that the combination of different time points and different imaging modalities increased the specificity of the response assessment during and after chemo-RT. Similarly, Iizuka et al[105] showed that a combination of ADC and 18F-FDG SUV was a better predictor for disease progression in NSCLC than one imaging modality alone. In analogy to dose painting concepts proposed for one single modality, more complex methods have been developed to base individualized RT treatment planning on multiparametric functional imaging information.[107]

Figure 1 shows an example of a patient with HNC presenting with multiparametric functional imaging information before the start of RT. This patient has been examined with combined PET/MR in addition to PET/CT, yielding 18F-FDG and 18F-FMISO PET data as well as anatomical MRI and also functional DW- and DCE-MRI.

Figure 1.

Multiparametric functional imaging of a 57-year-old male patient with head and neck cancer. (a) Fluorine-18 fludeoxyglucose positron emission tomography (PET) overlaid to anatomical T2 weighted MRI, (b) combined fluorine-18 fluoromisonidazole PET/MRI acquired 4 h post injection, (c) apparent diffusion coefficient map derived from diffusion-weighted MRI and (d) perfusion map showing the distribution of the parameter Ktrans derived via kinetic analysis from dynamic contrast-enhanced MRI inside the tumour volume. The radiotherapy gross tumour volume is outlined in each image slice.

A recently evolving field in functional imaging research for RT is radiomics.[108,109] Radiomics stands for the high-throughput extraction of a large amount of quantitative features from medical images, providing a comprehensive quantification of the tumour phenotype, yielding potentially complementary information to other sources such as demographics, pathology or genomics. In a large recent study, Aerts et al[109] applied a large number of quantitative image features to perform a radiomics analysis quantifying tumour image intensity, shape and texture, which are extracted from CT images of patients with HNC and patients with NSCLC. The study suggests that radiomics identifies general prognostic information, which may have a strong clinical impact to improve decision support in RT and cancer treatment in general.

REQUIREMENTS FOR THE USAGE OF FUNCTIONAL IMAGING IN RADIOTHERAPY

Integration of functional imaging in RT treatment planning requires special features in terms of image acquisition, quality and geometrical accuracy. For both functional imaging modalities, PET and MRI, a major pre-requisite for reproducible and robust image quality that can be safely taken into account during treatment planning is functional image acquisition in a dedicated RT treatment position.[3,110] Consequentially, PET and MRI should be performed on a flat table top, using vacuum mattresses and thermoplastic mask systems for patient fixation in exactly the same position as during fractionated RT treatment. If necessary, dedicated coil set-ups enabling image acquisition with those additional hardware components in the field of view may be used for MRI.[3] In the special case of combined PET/MR, dedicated PET- and MR-compatible positioning aids are required.[111] However, if imaging in RT position is not possible or in the presence of anatomical changes, dedicated methods for deformable image registration (DIR) are necessary to fuse functional imaging information on a voxel basis to the RT planning CT.[112,113] However, DIR is still a matter of research, and dedicated algorithms are only available in research software because careful clinical validation of DIR methods is still lacking. When PET imaging is used for RT target delineation or treatment planning of dose painting, the accuracy of quantitative PET information is crucial. The quality of PET data depends on a number of different factors, such as the image acquisition protocol, image reconstruction settings and also the technical characteristics of the imaging system.[114-117] Consequently, PET data used for RT planning needs to be acquired and analysed in a standardized way. Also, for the integration of (functional) MRI data into RT treatment planning, dedicated aspects in terms of patient positioning and image acquisition are required for a robust integration of the imaging data into the treatment planning process.[3] Two recent studies have investigated the geometric accuracy and the level of reproducibility in functional DW-MRI.[118,119] They found that DW-MRI can present with substantial geometric distortions[118] and also low levels of reproducibility when comparing repeated examinations.[119] Both factors are crucial for functional image-guided high-precision RT.

CONCLUSION

Functional imaging with CT, PET, MRI or hybrid imaging modalities offer a variety of possibilities to detect and visualize functional and biological processes related to tumour pathophysiology and radiation sensitivity. While initial results are promising, as discussed in this review, much research is still necessary in this emerging field of RT. However, functional image-guided RT has the potential to advance today's RT towards personalized medicine, with the realistic aim of improving cancer treatment in the near future.

FUNDING

DT is supported by the European Research Council, Starting Grant number. 335367.