Emerging Role of MRI in Radiation Therapy

Abstract

Advances in multimodality imaging, providing accurate information of the irradiated target volume and the adjacent critical structures or organs at risk (OAR), has made significant improvements in delivery of the external beam radiation dose. Radiation therapy conventionally has used computed tomography (CT) imaging for treatment planning and dose delivery. However, magnetic resonance imaging (MRI) provides unique advantages: added contrast information that can improve segmentation of the areas of interest, motion information that can help to better target and deliver radiation therapy, and posttreatment outcome analysis to better understand the biologic effect of radiation. To take advantage of these and other potential advantages of MRI in radiation therapy, radiologists and MRI physicists will need to understand the current radiation therapy workflow and speak the same language as our radiation therapy colleagues. This review article highlights the emerging role of MRI in radiation dose planning and delivery, but more so for MR-only treatment planning and delivery. Some of the areas of interest and challenges in implementing MRI in radiation therapy workflow are also briefly discussed.

Introduction to Radiotherapy for Cancer Treatment

The cancer mortality rate has decreased over the last century in men and women for almost all disease sites except glioblastoma and pancreatic cancer.¹ This is in part related to the advances in treatment resulting in better outcome and survival. In this context, radiation therapy (RT) is playing an increasing role as an important modality along with chemotherapy for the management of cancer patients in most disease sites.^2–4 It is estimate that 60% of cancer patients are treated with radiation in the management of their disease during their life span.⁵ Even though cost effectiveness of radiation therapy is debated due to heavy initial cost of modern machines and use of national resources,⁶ a true analysis provides a different picture, where radiation is cost-effective in the management of cancer patients.⁷

Advances in imaging, providing accurate information of the target volume (that is, the volume to be irradiated), and the adjacent critical structures or organs at risk (OAR), has made significant improvements in delivery of the external beam radiation dose. Radiation oncology has transitioned from a 2D approach to 3D conformal therapy (3DCRT) in 1990s. Additionally, modern advances with intensity-modulated radiation therapy (IMRT) and volumetric-modulated radiation therapy (VMAT) combined with inverse optimization in treatment planning has demonstrated a reduction in toxicity, as noted in a number of disease sites.^8–12 These techniques rely on imaging to appropriately plan and deliver radiation dose. Furthermore, more focused and hypofractionated beams for small targets such as stereotactic radiosurgery (SRS) and stereotactic body radiation (SBRT) has improved patient care by pinpointing the radiation delivery to the area of interest. Dose calculation from simple to more complex including inhomogeneity corrections have further tailored dose to the required sites. There are many aspects of progress in RT that are closely intertwined with imaging. These advances have improved planning and delivery of RT, resulting in superior outcome and reduction in toxicity with better quality of life for the cancer patients.

Current Workflow for RT Planning and Delivery

External beam radiation treatment involves a set of steps (Fig. 1) for precise and accurate dose delivery. The processes are immobilization, imaging for radiotherapy planning (computed tomography [CT], magnetic resonance imaging [MRI], positron emission tomography [PET]), image fusion, contouring (target volume and structure delineation), treatment planning including optimization, patient-specific quality assurance (PSQA), data transfer to the treatment device, pretreatment verification, and dose delivery. These steps could be modified or altered for some disease sites. A short description of the immobilization and imaging steps is described below.

FIGURE 1:

Radiation treatment planning and delivery workflow.

Immobilization

As radiation treatment lasts several minutes, patient immobilization is critical for imaging and treatment. The success of radiation treatment is heavily dependent on the quality of immobilization. Immobilization devices play an important role in the disease-specific and site-specific RT such as prostate, lung, breast, and head and neck (Fig. 2). Immobilization devices are made out of materials with low electron density (such as plastic) that do not attenuate the beam, are nontoxic to the skin, and are comfortable for the patients for the duration of the treatment time, which can last up to 30 minutes. There is a wide array of immobilization devices for disease-specific sites and they are made either on-site or are standard devices that can be retrofitted or adjusted to the patient’s body (such as a breast board). These devices are kept for the entire duration of the treatment, lasting a few days to as long as 45 days.

FIGURE 2:

Various types of immobilization devices. Upper panels are reusable and customizable to patient (a) breast and thorax board, (b) Leg support used in prostate, (c) prone breast board. Lower panels are patient specific fixation devices (d) aquaplast mold for head and neck, (e) Vaclock fixation used for trunk and (f) solid aquaplast device for pelvic immobilization.

Imaging

Imaging is an essential component of the external beam RT planning for treatment delivery. Currently, most treatments use CT data for treatment planning, as it provides patient- and tissue-specific attenuation and electron density information. The emergence of CT-simulation¹³ has changed the paradigm for treatment planning by providing patient-specific volumetric imaging data. It has replaced the regular radiographs with digital reconstruction radiographs (DRRs), which are created from the CT data.¹⁴ Before advances in CT imaging, during treatment planning the radiation dose was computed with all tissues considered to have the same density (that of water). However, modern treatment planning and dose calculation relies on CT data that provides spatial 3D map of tissues attenuation values (Hounsfield Units, HU). The HU is converted into electron density, which is used to calculate tissue- and organ-specific doses.^15–19

In current clinical practice, CT data are sometimes augmented with MRI and PET in the treatment of various malignancies. These images from other modalities are fused with CT data to take advantage of the superior soft-tissue contrast or metabolic information provided by these modalities. MRI provides unique and multiple contrast information. Furthermore, MRI contrast (such as T₁, T₂, and diffusion) could be tailored to highlight a specific organ or tumor. The sections below will briefly highlight the role of MRI in treatment planning and delivery. However, there are a number of challenges for MRI-only radiotherapy planning and delivery, including lack of electron density information and geometric distortion, which will also be addressed in this review.

MRI for Pretreatment Planning

The increased dosimetric conformity of modern treatment techniques, such as IMRT and VMAT, has generated new constraints on the accuracy of target delineation through imaging in RT. In order to deliver highly conformal treatments accurately, a precise definition of tumor and OAR is needed. This is the main reason MRI has seen increased usage in radiation oncology departments. Although in many cases CT still acts as the master reference scan, MRI provides superior soft-tissue contrast compared to CT, as well as a myriad of information on tumor characteristics that aid the delineation of both the tumor and OARs. The flexibility to acquire multiple contrasts has shown advantages for accurate tumor delineation in a large body of literature over the recent years. The American Association of Physicists in Medicine (AAPM) Task Group (TG-101)²⁰ on stereotactic body RT states that MRI is a gold-standard for visualization of brain tumors and “is increasingly used in SBRT applications including prostate, spinal tumors, chest, and solid abdominal tumors.” MRI is routinely utilized in a number of malignancies (Table 1) for treatment planning (Figs. 3–4). T₂-weighted imaging, for example, is able to distinguish tumor from normal tissue and fat in rectal and esophageal cancer,²¹ whereas T₁-weighted imaging provides good tumor contrast in squamous cell carcinoma in head and neck cancer.²² Besides native T₁ and T₂ contrast, physiological contrasts such as dynamic contrast-enhanced (DCE), blood oxygen level-dependent (BOLD), and diffusion-weighted imaging (DWI) have been shown to have added value in defining tumor boundaries.^23–26 Physiological information from DCE, DWI, and BOLD imaging (often all referred to as functional imaging in the RT community) has been successfully used to derive tumor probability maps in prostate cancer, a major male cancer in the Western world.²⁷ This information was later used in a large Phase 3 trial, called the FLAME trial,²⁸ in which a focal micro boost up to 95 Gy in 35 fractions was given to the tumor (ie, GTV), while the prostate gland (ie, CTV) received a standard dose of 77 Gy in 35 fractions.

TABLE 1.

Examples of Common Malignancies for Which MRI is Routinely Utilized for Pretreatment Planning

Organ
Brain	Better delineation of brain tumors on MRI as these tumors are not conspicuous on CT exam
Nasopharynx	Nasopharyngeal tumor are contoured on MRI and routinely fused with CT as shown in Fig. 3
Liver and pancreas	MRI is being increasingly utilized for hepatobiliary and pancreatic malignancy
Spine	Spinal tumor are visualized on MRI due to higher contrast resolution on Dixon and T2W images
Prostate	Prostate anatomy is not well visualized on CT and it can be difficult to delineate prostate from rectum. T2W MRI is helpful for visualization of prostate anatomy and tumor (Fig. 4)

FIGURE 3:

Gross tumor volume (GTV) for head and neck cancer is delineated on pretreatment (a) MRI image which is registered to corresponding (b) CT dataset. MRI has higher contrast resolution, which enables tumor visualization and accurate GTV delineation, whereas CT images provide electron density information and are used for on-board registration with cone beam CT.

FIGURE 4:

(a) Treatment planning CT of the pelvis for prostate cancer. (b) T₂WI of the prostate. (c) Fused CT and MRI for external beam radiation therapy for prostate cancer.

In a standard MRI-aided workflow, the acquired MRI data are registered to the planning CT on which the treatment plan is simulated. In order to minimize coregistration errors between the MRI and CT datasets, it is important that most imaging be performed in the treatment position (eg, flat tabletop with immobilization, as discussed in the section above). This, however, requires adaptations to the MRI workflow and the use of specialized imaging hardware. Most major MRI vendors themselves or in partnership with third-party suppliers offer special editions of their flagship 1.5T and 3T scanners that are equipped with tailored RT hardware and software. These options include more accurate laser positioning devices, flat tabletops that match the treatment table, coil bridges to prevent deformation of the patient’s body contour, and fixation devices such as thermoplastic masks and arm supports, as shown in Fig. 2, but compatible in the magnetic field. The imaging protocols are also adapted to the specific requirements of treatment simulation. In comparison with diagnostic imaging, much more emphasis is put on the geometric accuracy of the imaging. The geometric accuracy of the preparatory scans determines the required safety margins,^29–32 and thus the amount of healthy tissue that is irradiated. Therefore, it is important to achieve the highest geometric accuracy possible. For this reason, RT scans are typically acquired at higher resolution and higher readout bandwidths, at the expense of signal-to-noise ratio (SNR). The push for 3D acquisitions is also higher than in diagnostic imaging, for two reasons: 1) the need to acquire isotropic resolution, and 2) the need to correct for gradient nonlinearity along all three dimensions. All these adaptations have to ensure that the MRI matches the planning CT as well as possible, because any misregistration would introduce a systematic error that propagates through the entire treatment.

A more recent trend is the use of MRI as the sole modality for RT: the so-called MRI-only workflow. A workflow in which all the preparatory steps are carried out on the MRI is favorable from a logistic point of view and removes the need to register the images to a separate planning CT, which potentially minimizes the risk of systematic error due to misregistration.^33–35 The major challenge for such a workflow is the assignment of electron densities to MRI scans for the dose calculation by the treatment planning system. However, partly due to the similar challenges that exist in PET/MR, a number of methods have been proposed in the literature that allows the generation of synthetic CT from MRI data. The methods range from voxel-based approaches that primarily use the information about voxel intensities, atlas-based approaches that register the images to a known (segmented) atlas, or hybrid approaches that use both. A brief discussion will follow but a comprehensive review has been published by Edmund and Nyholm.³⁶ More recently, deep learning approaches are gaining momentum and showing potential for success in converting MRI data to synthetic CT, especially where bone and air are present.^37–39

Synthetic CT From MRI Data

Generation of Synthetic CT

The MR-only treatment planning can avoid potential error in MR-to-CT registration, and spare the cost and radiation from dedicated CT simulation. The emerging hybrid MR and linear accelerator technique⁴⁰ also motivates MR-only treatment planning to simplify the process of MR guidance in RT. However, unlike CT, there is no simple conversion from MR signal intensity to the electron density value that is required for accurate radiation dose calculation. Various methods have been proposed to generate pseudo-CT or CT-like images, also called synthetic CT (synCT) from MR images to replace CT data for radiotherapy treatment planning. These methods mostly utilize two types of information to establish a relationship between MR intensities and CT HUs: 1) tissue information derived from MR images, and 2) MR-to-CT correlation and transformation from a paired MR and CT dataset.

Tissue classification methods derive tissue content from MR images and apply the knowledge of bulk density of various tissue types to assign a CT number. The classification could be done by manual segmentation of T₁- or T₂-weighted MR images^41,42 or automatic intensity-based classification on multiple MR sequences.^43–45 Bone and air have low signals on conventional MR sequences. Thus, discriminating bone- and air-containing tissues remains a major challenge for intensity-based methods. The typical solution has been to include an ultrashort echo time (UTE) or zero-TE sequence data in the classification of bone and air.^44–47 However, it is important to note that including UTE and using multiple MR sequences increase scanning time, which may lead to motion artifacts and misalignment between images from different sequences. Other approaches that do not utilize additional imaging with UTE sequences include image segmentation methods such as a bone shape model⁴⁸ and active contour⁴⁹ to segment bone from T₁-weighted or DIXON MR images before classification of the remaining voxels for nonbone tissue. However, the accuracy of bone segmentation may still suffer from nearby air and artifact.

Although the appearances of MR and CT vary significantly, the spatial correlation between paired MR and CT could provide clues to constructing CT-like images from MRI data. Using MR images as signatures of associated CT, atlas-based methods generate synCT for patient MR by finding optimally matched atlas images.^50–52 These methods use conventional MR sequences but depend on the accuracy of deformable registration between atlas and patient MR. It is important to note that deformable registration is a challenging task, especially if the patient has pathological and/or anatomical differences from atlas images. To alleviate the inherent registration errors, the multi-atlas methods register patient MR to multiple atlases and fused associated CTs to generate synCT. Various multi-atlas methods have been proposed that use different approaches for fusing multiple atlas CTs, including voxelwise median,⁵³ probabilistic Bayesian analysis,⁵⁴ local image similarity,⁵⁵ and/or regional errors in the registration.⁵⁶ The multi-atlas registration and fusion was designed to increase robustness to registration errors, but nevertheless adds complexity and computation burden to the treatment planning workflow.

Instead of simply using spatial similarities in an MR-CT dataset, learning-based methods derive a map function to associate MR voxel intensities or image patches with HU numbers by supervised training on the atlas dataset. The mapping could be a regression function,^44,57,58 statistical decomposition,⁵⁹ random forest modeling,⁶⁰ or pattern recognition technique.⁶¹ Recently, machine learning, especially deep learning and convolutional neural networks (CNNs),⁶² has shown potential in this task of generating synthetic CT from MRI data. Han³⁷ built a U-Net-based CNN model⁶³ that consisted of an encoding part to learn from an input 2D MR slice and a decoding part to generate a corresponding 2D synCT slice. CNNs automatically learn multiple levels of information from a large set of MR-CT datasets. A CNN model⁶⁴ learned the mapping from a 3D multiple-parametric MR patches input to a same-size 3D synCT patch on a U-Net architecture. The model training takes a long time and a large amount of data, but synCT generation after training could be faster than classification or atlas-based approaches.

The aforementioned methods are more complementary than competitive. By using image classification for soft tissues and atlas registration for bone delineation, the hybrid method^64,65 generated synCT from DIXON MR without the need of a UTE sequence (Fig. 5). Gudur et al⁵⁴ combined T₁-weighted MR intensities and atlas-based geometry information to build a unified posterior probability density functions (PDF) for assigning the CT number. Continued advances will result in improved construction of synthetic CT from the MRI data.

FIGURE 5:

Workflow of a hybrid method for synCT generation from Dixon MR images. These methods were initially proposed for PET/MR attenuation correction. μ is linear attenuation coefficient (LAC). SynCT is generated by converting the μ-map to HU numbers.

Evaluation of Synthetic CT

SynCT have been developed and evaluated in support of MR-only treatment planning in a variety of anatomical sites,⁶⁶ including the brain,^67–69 head and neck,^45,70 and pelvis.^42,71,72 The majority of the studies were retrospectively performed on a small number of patients, usually in the range of 10–20. However, a recently performed prospective multi-center study of over 150 patients⁷³ validated MR-only prostate treatment planning using commercially available software. To replace CT in the treatment planning workflow, synCT is typically evaluated for the HU similarity between synCT and conventional planning CT and the equivalence of dose distributions calculated from the two datasets.

The mean absolute error (MAE) that is average of voxelwise absolute HU differences between synCT and CT is commonly used for synCT evaluation. The MAE should be calculated within the body contour but is often reported in different tissue regions, as there are great variations in HUs between different tissues. Instead of directly comparing voxelwise HU differences, geometric similarity between synCT and CT could be assessed by overlap of tissue volumes between the two images. The Dice similarity coefficient (DSC) for bone volumes has been reported for different methods and anatomical sites.³⁶ The external body contour is often generated automatically in treatment planning software using a threshold technique. Its spatial accuracy can be very important in dose calculation, especially for superficial tumors.⁷⁴ The DSC metric has been reported for geometrical accuracy of body volume, but it may be insensitive to the differences in the contours.⁵⁵

Many studies have demonstrated the dosimetric equivalence of plan doses calculated on synCT and CT. These studies either recalculated a clinical plan generated from CT on corresponding synCT, or created a plan on synCT and recalculated it on CT to simulate an MR-only workflow. The synCT and CT-calculated doses are then compared for 3D dose distributions and dose-volume metrics. Absolute dose difference maps describe dosimetry agreement in 3D space,⁷² but relatively large deviations are present at the boundaries of the body and target volume due to possible synCT-to-CT misalignment at the high dose gradient area. Gamma analysis composites both dose difference in low-dose gradients and distance to agreement (DTA) in high-dose gradients,⁷⁵ and is nearly unanimously reported in the literature for dose agreement assessment. The gamma index could be calculated on 3D dose volumes or 2D dose planes that may yield more stringent results.⁷⁶ Overall, gamma analysis across different methods and anatomical sites showed a >95% passing rate with clinically used criteria of 2% and 2 mm (dose-difference/DTA).^{36,42,65,67,68,72}

Clinical assessment of a plan for treatment is largely based on dose-volume metrics for target volume and related OARs. The metrics are calculated from cumulative dose volume histograms of structures based on the guidelines of Quantitative Analysis of Normal Tissue Effects in the Clinic (QUANTEC).⁷⁷ The differences in reported metrics mostly were less than 2% and statistically insignificant^{36,42,65,67,68,72} (Table 2). Wang et al⁷² further applied the graphic technique to assess the equivalence of dose-volume metrics between the two image modalities. The clinically acceptable agreements in the available literature suggest that possible residual distortions in MR, synCT-to-CT local misalignment and HU deviations are of minor importance for dosimetric accuracy, likely due to the fact that doses of high-energy photons are relatively insensitive to small local electron density variations.^78,79

TABLE 2.

Metrics Routinely Used for Evaluation of Dosimetric Agreement Between synCT and CT Calculated Plans

Tumor type	PTV dose metrics	OAR dose metrics
Brain tumor67	D95%, D5%, Dmax	Brainstem, optical nerves, eyes, lenses, chiasm, cochlea: Dmax
Head and neck cancer55,56	D98%, D2%, Dmean, Dmax	Parotid glands, submandibular glands, brain stem, spinal cord: D2%, Dmean, Dmax
Lung cancer65	D95%, D98%, D100%	Lung: V10Gy, V20Gy; Heart: V40Gy; Spinal cord: Dmax
Liver cancer49	D99%, D95%, D5%, D0.1cc	Colon, spinal cord, duodenum, esophagus, heart, stomach: D0.5cc
Prostate cancer41,42	D99%, D98%, D95%, D2%, Dmean, Dmax	Bladder, rectum: D35%, D25%, D15%, D2%, Dmean; Penile bulb: D90%
Pelvic cancer48	D99%, D0.5cc	Femur: V30Gy; Pelvis: V10Gy, V20Gy, Dmean; Rectum: V45Gy, Dmean; Sacrum: V10Gy, V20Gy; Bowel: D1cc, D5cc, V55Gy
Colorectal cancer72	D100%, D95%, D2%, Dmean	Bladder: V40Gy, Dmean; Bowel: V45Gy, Femoral head: V30Gy

PTV: planning target volume; OAR: organ at risk. A dose metric for a structure is calculated from dose volume histogram (DVH).

Geometric Distortion

The essence of imaging for radiotherapy treatment guidance is geometric fidelity, as geometric distortions could directly lead to a misplacement of the radiation dose, thereby decreasing the effectiveness of the treatment and potentially increasing toxicity for nearby OAR. In diagnostic imaging, geometric distortions are normally only considered for acquisitions with long imaging readouts like echo-planar imaging (EPI) and distortion along a single readout line is usually ignored. For real-time MRI-guided RT, however, precision of <2 mm is required, so imaging is performed at much higher readout bandwidths (and thus lower SNR) compared to diagnostic imaging.^26,80 To further minimize off-resonance, corrections along the readout direction has been explored for treatment preparation scanning,⁸¹ and for online MRI guidance.⁸² To date, however, these methods have not yet been clinically introduced. DWI is a powerful imaging tool in oncology, and is often used for delineation guidance and treatment response monitoring. Due to the sensitivity to off-resonance distortion, EPI-based DWI is severely limited in head and neck and thoracic regions, even after distortion correction. For this reason, turbospin echo (TSE)-based DWI has seen renewed interest. DWI with a modified fast spin-echo acquisition (DW-SPLICE)⁸³ has been shown to be a viable, distortion-free alternative for imaging of head and neck patients at 3T,⁸⁴ but comes at the price of reduced SNR and prolonged imaging time due to the TSE readout, which hampers the transitioning to 1.5T.

Apart from off-resonance-induced distortion, distortions caused by gradient nonlinearities are also an important consideration. For diagnostic imaging, the vendor-provided corrections, performed during image reconstruction, may be adequate. For therapy guidance, however, these residual distortions need to be carefully characterized⁸⁵ and ideally further mitigated. Particularly for single-slice acquisitions, for which the gradient nonlinearity is only corrected in-plane, is still an unsolved issue for real-time image guidance. This through-plane slice distortion (referred to as potato chipping) could lead to local displacements of up to a few centimeters at locations 20 cm away from the isocenter.⁸⁶ Further work is needed to improve the geometric accuracy of various MR acquisition schemes and is an active area of investigation.

MR Guidance During Treatment

Over the last two decades the quality of external beam radiotherapy has advanced tremendously due to advances in pre- and postimaging such as on-board imaging methods like cone beam computer tomography (CBCT)⁸⁷ and electronic portal imaging devices (EPID).^88,89 The major limitation of the x-ray-based position verification is the poor soft-tissue contrast of these imaging modalities. Apart from lung tumors, tumor visualization is barely possible on clinical CBCT images, as shown in Fig. 6. Patient positioning is therefore often based on the bony anatomy in the vicinity of the tumor. Since a fixed relationship with bony anatomy (which can be well visualized on CBCT) cannot be established with great confidence, these uncertainties with respect to tumor position are dealt with by margins: the tumor volume is expanded to a much larger planning target volume, to make sure the tumor is always in the prescribed radiation beam. To stress this limitation: for these tumors it is accepted clinical practice that radio opaque markers (ie, fiducials) are implanted surgically, prior to radiotherapy, in order to allow “tumor-based” image registration on the linear accelerators (Fig. 7). This surgical procedure, however, is a substantial burden and risk to the patient.

FIGURE 6:

Planning CT (top row) and cone beam CT (bottom row) of a liver tumor in axial, coronal, and sagittal views. Cone beam CT images are of inferior quality with poor visualization of the tumor. Red curve is the contour of the tumor obtained on the planning CT and copied to the Cone beam CT after image registration.

FIGURE 7:

Example images of a pancreatic tumor. The poor soft-tissue contrast in (a) CT and (b) Cone Beam CT necessitates the use of implanted fiducials in order to allow accurate tumor positioning on the LINAC. MRI on the other hand allows (c) direct visualization and (d) motion tracking of the tumor as well as organs at risk, which is essential for online treatment monitoring. No implanted fiducials are needed.

The ultimate integration of MRI in modern radiotherapy is the use of MRI during the treatment session. Several groups are working on integrating MRI into the linear accelerator, resulting in hybrid MR-LINAC devices (Table 3). On-board MRI will allow position verification to be performed directly on the tumor, instead of nearby bony structures or implanted fiducials. Moreover, real-time MRI allows continuous tracking of the tumor position (Fig. 7) during radiation delivery. Depending on the type of motion, the treatment can be delivered in a gated fashion (eg, treatment delivery only during a certain respiratory phase), or fully tracked (in which the treatment beam follows the entire respiratory pathway).

TABLE 3.

Overview of Some of the MR-LINAC Systems Either Currently Available Commercially or Being Actively Investigated in a Research Setting

ViewRay MRIdian,** Cleveland, USA	0.35 T MRI 3 Co sources / 6 MV Linac
Elekta Unity,* Stockholm, Sweden	1.5 T MRI 7 MV Linac
Aurora-RT, MagnetTx, Edmonton, Canada	0.5 T MRI 6 MV Linac
Australian MRI-Linac, Ingham Institute, Liverpool, Australia	1.0 T MRI 4 & 6 MV Linac

^*CE marked;

^**CE marked and FDA cleared (June 2018).

The MRI-only pathway for RT has been suggested and implemented by many groups.^{26,39,90–95} Such approaches have patient comfort, time delay, and financial burden for the multiple imaging in mind. However, many technical hurdles remain unsolved, such as geometric distortion and creation of universal synthetic CT.^25,96–98 An MR-only workflow, with or without MR-LINAC, will require suitable MRI-compatible immobilization devices, MRI sequences for disease-specific imaging for target and OAR delineation, synthetic CT generation, treatment planning, and then finally treatment. With a dedicated and integrated system, repeated MRI can play an increasing role in evaluation of treatment response, as suggested by number of research groups,^99–102 and briefly discussed below.

MRI for Treatment Assessment

The treatment response is usually evaluated by measuring the change in size of the tumor either in one dimension (RECIST and RECIST 1.1) or in two dimensions (WHO criteria). It is well understood that a change in size is a late measure of treatment response, and hence there is tremendous interest in developing biomarkers of early treatment response, especially in this era of targeted chemotherapies, immunotherapies, and combination of chemotherapy and RT.

MRI provides a unique opportunity to explore different contrast mechanisms to assess early treatment response. DWI is a functional MRI technique that is sensitive to the random microscopic motion (also known as Brownian motion) of the protons associated with water molecules. Highly cellular and complex tumor tissue impede mobility of the water molecule, resulting in high signal on DWI and corresponding low apparent diffusion coefficient (ADC). In response to therapy, tumor tissues will have decreased overall cellularity with increased necrosis. This will result in less restricted water molecules, resulting in low signal on DWI and correspondingly high ADC.

A number of studies have shown that changes in diffusion signal and ADC are helpful in predicting pathologic response to various tumor types, including locally advanced rectal cancer as well as cervical cancers undergoing combined neoadjuvant chemotherapy and radiation therapy (CRT).^103,104 In a recent study, post-CRT skewness of the ADC histogram and percentage change in ADC were useful for predicting a favorable response to neoadjuvant CRT in cervical cancer.¹⁰³ Similarly, in a small study of pancreatic cancer, there was an increase in ADC values of the pancreatic tumor after neoadjuvant chemoradiation.¹⁰⁵ Furthermore, posttreatment ADC values were correlated with degree of pathologic response.

DCE perfusion-weighted imaging (PWI) has also been explored in evaluation of treatment response after RT. However, this requires injection of exogenous gadolinium contrast agent. Other MRI endogenous contrasts such as arterial spin labeling (ASL), spectroscopy, T₁ and T₂ relaxation rates are promising and need further evaluation.

Conclusion

MRI is increasingly utilized in radiotherapy treatment planning due to improved contrast resolution of MRI compared to conventional CT. Furthermore, MRI will play an important role in the delivery of radiotherapy and in the assessment of treatment response in the near future, especially with introduction of the MR-LINAC systems. It will require close collaboration between radiation oncologists, radiologists, MR physicists, and RT physicists to take advantage of the unique capabilities of theses combined systems. Radiology and the MR community will need to understand the basics of RT planning and delivery and how MRI currently plays a role in radiotherapy treatment. This will enable development of novel MRI methods to tackle the unsolved problems and unmet need.