Emerging Role of MRI for Radiation Treatment Planning in Lung Cancer

Magnetic resonance imaging is expected to play an increasingly important role in differentiating between malignant and benign nodules. Zou et al showed that for dynamic contrast-enhanced MRI scans (DCE-MRI), the uptake pattern of contrast agents varied for the 3 pathologies.⁴⁰ Benign nodules showed a low overall uptake, whereas the uptake in both malignant and inflammatory nodules was initially high. However, 4 minutes after administration of the contrast bolus, the uptake in the inflammatory nodules was significantly higher than in the malignant nodules, possibly caused by differences in the vascular integrity. Besides DCE-MRI, diffusion-weighted imaging (DWI) may be used to differentiate malignant from benign nodules. This was confirmed in a meta-analysis, which also highlighted the need for standardization in future studies.⁴¹ Furthermore, a study by Zhang et al⁴² showed that the apparent diffusion coefficient (ADC) value, which can be derived from diffusion-weighted MRI (DW-MRI), is a significantly better discriminator of malignant and benign nodules than the maximum standardized uptake value in PET imaging (SUV_max). Similar results were found in other studies.^43-45 Due to the high DWI contrast between the malignant nodules and the other tissues, Chetley Ford et al claimed that DW-MRI may be a good tool for involved lymph node delineation⁴⁶ (Table 1).

Magnetic resonance imaging may also be used to discriminate atelectasis from the primary lung tumor. In T2-weighted (T2w) scans, a contrast difference can be observed between the tumor and the atelectasis.⁴⁷ This contrast difference is less well defined in T1-weighted (T1w) scans. Furthermore, atelectasis may be visualized with DWI. Yang et al performed DWI on patients with pathologically proven lung cancer and presence of atelectasis.⁴⁸ Region of interest analysis showed that the ADC of lung tumors was significantly lower than for atelectasis. An overview of T1w, T2w, DWI, and ADC images is shown in Figure 1. Unfortunately, neither ¹⁸FDG-PET-CT nor MRI has been (prospectively) evaluated with histopathological conformation to distinctively discern between malignant tumor and atelectasis.

The extent of primary tumor invasion into mediastinal structures, the thoracic wall, or neurovascular structures is important for primary tumor staging and for the definition of the GTV in subsequent radiation treatment planning. The “standard” anatomical sequences used are T2w, T1w with contrast, and magnetic resonance angiography (MRA).^49,70 However, MRI can also be used to assess tumor invasion into mediastinal structures, either by enhancing existing sequences, such as MRA, or new sequences such as respiratory dynamic MRI (RD-MRI) and thin-section single-shot turbo spin echo with half-Fourier acquisition (SS-TSE-HF or HASTE) as demonstrated in 3 recently published articles.^50-52 More recently, imaging sequences with ultrashort echo time (UTE) are being explored to accurately discriminate between the tumor and the lung tissue³⁰ (Table 1).

Interpretation of thoracic MRA images is hampered by cardiac motion artefacts. Ohno et al enhanced the image quality of MRA using electrocardiogram (ECG) triggering.³³ In 20 of the 50 patients having NSCLC with tumor invasion into the mediastinum or hilum, image quality and diagnostic accuracy of the conventional MRA and ECG-triggered MRI were compared with contrast-enhanced CT and the surgical findings. The image quality of the ECG-triggered MRA images was significantly better than that of images acquired with conventional MRA or contrast-enhanced CT. In addition, ECG-triggered MRI showed a higher diagnostic accuracy for invasion of the pulmonary artery or vein of 86%, whereas contrast-enhanced CT, conventional MRI, and MRA demonstrated an accuracy of 68%, 73%, and 82%, respectively. In this study, however, a single-detector helical CT was used, which is not the current standard.

In 3 recently published articles, new MRI scan protocols were investigated.^51,52,71 Patients with resectable NSCLC underwent preoperative MRI and CT scans. The first 2 articles investigated RD-MRI, whereas the third article explored the use of SS-TSE-HF.^51,52,71 In each article, the new MRI technique was either added to conventional MRI sequences and compared with CT or directly compared with CT. Next, the radiological findings were then compared with the pathological findings of the surgical specimens.

The RD-MRI method visualizes the natural sliding motion of the lung tissue along the adjacent tissue. Patients breathed slowly during image acquisition. In patients with tumors showing no or restricted sliding motion, invasion of the tumor in the adjacent tissue was suspected.^50,52 Both CT and conventional MRI already showed sensitivity and negative predictive value of 100% for tumor invasion. However, the specificity of CT was generally low. The advantage of RD-MRI is that it significantly lowers the number of false-positive findings and thereby increases the specificity as compared to CT and/or conventional MRI sequences. This led to an increase in specificity in both studies of 29% to 71% and 44% to 69%, respectively.^52,71

The RD-MRI is particularly accurate for tumors that are less than 5 cm in diameter and tumors close to the mediastinum. There are 3 possible explanations.⁵⁰ First, the lower lobes show more respiratory movement than the upper lobes and therefore possess more “sliding” motion potential with the tumor. Second, a tumor larger than 10 cm and close to the chest wall exhibits little motion, for the weight of the tumor hinders motions between the tumor and the adjacent tissue. Third, in the upper lobes, especially the left upper lobe, there is little room for movement because of the tight anatomical configuration.

The SS-TSE-HF MRI has a tissue contrast comparable with turbo spin echo. However, by obtaining 3-mm slices with parallel imaging perpendicular to the tumor, the spatial resolution can be increased, a technique also known as SS-TSE-HF MRI (Figure 2).⁵¹ A breath hold of 12 seconds is needed to acquire a full data set. With this technique, the sensitivity for local invasion was 85%, which was comparable with CT and “conventional” MRI (RD-MRI, conventional axial SS-TSE-HF MRI, and T1w high-resolution isotropic volume excitation). Remarkably, the specificity was 89% and thus superior to the rates of 26% and 61% for CT and conventional MRI sequences, respectively. Furthermore, the 2 reviewers who independently reviewed the imaging modalities, both blinded for clinical information and the findings of the other imaging modalities, had an excellent interobserver agreement (κ = 0.81) when using thin-section SS-TSE-HF MRI as compared to conventional MRI (κ = 0.68) and CT (κ = 0.54).

Noteworthy, most patients in the 3 aforementioned studies had a tumor either with chest wall invasion or with invasion of the aorta. To our knowledge, little is known about sensitivity and specificity of (those) MRI sequences with regard to invasion of the other mediastinal structures or the invasion in bony anatomy.

The last sequence, UTE, enables signal retrieval close to air–tissue transitions, which is problematic using other sequences.³⁰ With the UTE technique, the time between MRI signal generation and acquisition is extremely shortened in order to capture as much signal as possible. In one study, the UTE technique was optimized in healthy volunteers showing improved visualization of the lung tissue and airways.³⁰ Unfortunately, validation of UTE imaging against pathological specimens is still lacking. We anticipate that UTE imaging may be of special interest for radiation treatment planning as it can potentially lead to a more accurate depiction of the tumor borders.

Four-dimensional CT is the standard imaging modality for motion detection of primary lung tumors. A disadvantage is that it reconstructs motion as a function of the respiratory phase instead of time. Therefore, it does not reflect tumor motion accurately in real time.²⁶ Since tumor motion can differ significantly depending on the tumor location, it is important to have accurate motion characterization for accurate definition of the GTV and subsequent PTV margins.²⁶

Motion characterization with MRI can be performed with cine MRI. With this technique, multiple 2-dimensional (2D) volumes per second can be acquired along any slice orientation, and image data are very rapidly processed. Cine MRI can be complementary to 4D-CT. Cine MRI sequences show the motion pattern per respiratory cycle, whereas CT shows the average motion pattern over a number of respirations.⁷² Therefore, cine MRI is a more direct measure of respiratory motion and can also be used to gain insight into the variation in the respiratory pattern. This technique has already shown to accurately depict impaired respiratory mechanics associated with pulmonary emphysema.²⁷ Important to remember is that dynamic sequences always have a tradeoff between temporal and spatial resolution. The search for new dynamic sequences is to find an ideal balance between temporal and spatial resolution, which enables the radiation oncologist to accurately define the tumor as well as to determine the individual movement pattern of the tumor with respiration and possibly with cardiac motion (Table 1).

Two sequences for fast cine MRI have been investigated frequently: balanced steady-state free precession (bSSFP) and fast low-angle shot (FLASH) imaging.²⁸ In patients with stage I NSCLC, 2 dynamic MRI techniques were compared: 1 sequence with a high-spatial resolution (bSSFP, 3 images/s) and 1 with a high-temporal resolution (FLASH, 10 images/s).²⁸ The data acquisition rate was increased by parallel acquisition techniques. The bSSFP showed the best signal-to-noise ratio (SNR), with still an acceptable good temporal resolution. However, in this study, only the movement in the craniocaudal direction was investigated in the coronal plane, and a slice thickness of 10 mm was applied.

The use of 2D and 3-dimensional cine MRI for offline and online soft-tissue–based image guidance in patients with stage I to IV NSCLC is currently under investigation at the University of Texas Southwestern Medical Center (NCT01421784, ClinicalTrials.gov). This investigation focuses on monitoring moving and deforming tumors.

Contrast-enhanced CT, registered to the ¹⁸FDG-PET-CT, is the imaging standard for the delineation of the lymph nodes. However, accurate delineation of metastatic lymph nodes is hampered by 2 limitations. First, the movement of the lymph nodes, and second, the technical limitations of these imaging modalities (Table 1).

Several processes influence the movement of the lymph nodes. The respiratory pattern, cardiac motion, location of lymph nodes (eg, mediastinal vs hilar region), and changes in the lymph node volume during treatment influence the mobility of the lymph nodes. Furthermore, movement of the metastatic lymph nodes hardly correlates with the movement of the tumor.^77,78 The movement of the lymph nodes aggravates the limitations of ¹⁸FDG-PET-CT and contrast-enhanced CT: (small) nodes may be missed due to the low resolution and motion blurring.³⁸

Cine MRI has the potential to characterize the motion of the lymph nodes. However, to our knowledge, no research on this subject has been published yet. We believe that the motion characterization of the tumor and lymph nodes will be essential in creating individualized PTV margins, enabling enhanced sparing of the OARs, and safer dose-escalation studies and stereotactic radiotherapy of centrally located tumors.

The mobility of several OARs has been investigated with 4D-CT. An early study by Giraud et al showed that thoracic organs move considerably due to respiratory motion, especially at the diaphragm (up to 35 mm), whereas the smallest motion was recorded at the lung apices (5 mm).⁸⁰ The mobility of the esophagus, especially of the distal esophagus, can be as large as 9 mm in the dorsoventral or the lateral direction.⁸¹ The motion of the heart has been investigated recently and can be as large as 13 mm in the lateral direction.²¹

No specific MRI sequences have been investigated to characterize the motion of the OARs for radiation treatment planning. This is unfortunate since acute and late cardiac and lung toxicity in thoracic radiotherapy is substantial, especially in patients with cardiac and pulmonary toxicity. Furthermore, SBRT of (central) lung tumors, dose-escalation studies, and new systemic therapies will probably lead to a longer survival and therefore also to an increase in observable late toxicity. Cine MRI will have an important role to characterize the real-time motion of the OARs, which may lower the dose in OARs when taken into account properly in treatment planning (Table 1). In addition, knowledge of the motion patterns will enable more realistic registration of the dose delivered to the OARs.

The current standard of dealing with thoracic motion in (stereotactic) radiotherapy of lung tumors is an internal target volume (ITV)-based approach. The ITV is typically generated from the thoracic motion information obtained by 4D-CT. Although no vendor solution exists as of yet, several 4D methods for MRI have been developed in a research setting.^66,67,88 Apart from a few adaptations to cope with the technical differences between CT and MRI, these methods are comparable to the 4D-CT methods by the fact that they all sort the data based on the respiratory trace retrospectively. The advantage of 4D-MRI over 4D-CT is its flexibility in acquisition strategy, which can avoid some of the artifacts typically seen in 4D-CT such as motion-induced blurring and incomplete structure artifacts.⁸⁹ In addition to this, MRI has improved soft-tissue contrast, which can be tuned to either the tumor or the surrounding OARs. Because of these advantages, we foresee that vendors will also turn 4D-MRI into product within a few years’ time.

Unfortunately, retrospective 4D methods hold 2 clear disadvantages: (1) they only provide a representation of the “average” tumor motion (ie, they do not provide information on the intercycle variability) and (2) the motion during simulation may be different from motion during the treatment fractions.⁹⁰ To mitigate these disadvantages, a few research groups have endeavored to develop techniques that link the simulation scans with the real-time imaging available during treatment.^88,91 The goal is to combine the high-resolution 4D information obtained during simulation with the fast, but often 2D, imaging during radiation in through physiological or mathematical motion models. In the future, this will lead to accurate dose optimization, taking 4D information into account.

Approaches have been explored in a conventional setting by linking 4D-CT to 2D electronic portal imaging device imaging or for future MRI-linac–based treatments linking 4D-MRI to real-time 2D cine MRI. All these methods aim to solve the fact that MRI is not fast enough to provide true 4D real-time imaging. However, over the last few years, tremendous advances have been made in imaging speed. These developments are likely to continue and promise even greater imaging speeds in the future and will hopefully bring true 4D real-time MRI within reach of clinical use.

Another interesting application of the recently developed motion models is the reconstruction of PET images in hybrid PET-MRI scanners. Using the simultaneously acquired MR images, it is possible to correct for motion blur of the PET acquisition, thereby increasing the PET resolution, as shown by several groups.^92,93 In the future, this could have a significant impact on the detection of small mediastinal lymph nodes.

Apart from combined PET-MRI imaging, we anticipate further advances in multiparametric MRI. In other areas, such as prostate and brain imaging, it is common to acquire multiple quantitative MRI scans, such as DWI-MRI and DCE-MRI. Instead of utilizing each imaging sequence for individual purposes, the current efforts focus on advanced modeling techniques on multiparametric imaging, where the combined information from multiple contrasts is used for diagnosis, staging, and radiation treatment planning.^94-96 The complimentary information obtained by conventional T2w imaging (morphological information), combined with functional techniques such as DWI-MRI, and DCE-MRI has been very successful in imaging of the prostate. We anticipate that these sequences, in combination with RD-MRI, will be prime candidates for multiparametric MRI of the lung. However, a prerequisite for multiparametric MRI is accurate image registration between the different scans. When imaging the lung, conventional DW-MRI and other functional sequences have large geometric distortions, making accurate registrations difficult. This is one of the technical challenges that have to be addressed before the added value of multiparametric MRI can be assessed in the context of lung imaging.

In order to overcome the low proton density in the lungs, HP gas MRI can be used. In HP-MRI, MRI-sensitive nuclei (eg, 3He or ¹²⁹Xe) are being polarized and administered to the patient during the scan. The hyperpolarization enables direct detection of the gas on MRI, which dramatically improves the SNR in the lung parenchyma and allows function assessment of the lungs. Several studies have already reported that HP-MRI provides additional information on healthy lung tissue in patients with NSCLC, and studies are being designed to investigate functional lung avoidance treatment planning based on HP-MRI.^62-64 If these trials are successful, this will eventually translate into the clinic.

In the past decades, USPIO particles have been developed for diagnostic imaging, especially nodal staging.⁵⁸ One such particle is ferumoxytol. Under normal circumstances, ferumoxytol is taken up by macrophages in nonmalignant lymph nodes, which results in a change from hyperintense (white) into a hypointense (dark) lymph node on T2w (contrast-enhanced) images. If the lymph node contains metastatic cells, little or no ferumoxytol will be taken up and the (metastatic part of the) metastatic lymph node will remain hyperintense (white; Figure 5).^58,59

Most research has been performed for the nodal staging in prostate cancer and shows a sensitivity between 65% and 92% and specificity between 93% and 99%.^98,99 To our knowledge, only one publication exists, in which MRI and ferumoxtran-10 were used for mediastinal staging in 18 patients with NSCLC.¹⁰⁰ The MRI findings were correlated with histological or cytological findings, and the authors reported an overall sensitivity of 92% and specificity of 80%. The sensitivity and specificity were dependent on the size of the lymph nodes, with a sensitivity and specificity of 67% and 88% in nodes <15 mm and 100% and 71% in nodes >15 mm, respectively. This may be explained by the limited resolution and other technical limitations of the MR scanners of that time period. Future research should focus on the comparison of ferumoxytol-enhanced MRI sequences with PET-CT and histological findings.

Currently, the treatment response is assessed with the response evaluation criteria in solid tumors (RECIST version 1.1), which were last revised in 2009.¹⁰¹ These criteria determine treatment response by changes in tumor size on CT in 2 dimensions. Three-dimensional or functional changes in the tumor (eg, measured with DWI-MRI, DCE-MRI, or PET-CT) were not incorporated in the revised RECIST criteria. This is a missed opportunity for 2 reasons. First, future evaluation tools should be aimed at detecting other signs of treatment response, such as functional imaging, since the use of targeted agents may not lead to significant reduction in tumor size and volumetric responses to chemoradiation do not correlate with pathological complete response.^102-104 Second, recent publications in patients with NSCLC indicate that metabolic volumetric information of the tumor (and metastatic lymph nodes and distant metastases), derived from either repeated ¹⁸FDG-PET-CT for therapy evaluation or solely from diagnostic ¹⁸FDG-PET-CT, can be highly predictive of survival even in patients with the same tumor, node, and metastasis stage.^105-108

Although functional changes in tumor metabolism during treatment have been incorporated in the PET response criteria (PERCIST), PERCIST has only been validated for the treatment evaluation in lymphoma.^103,109 The main reasons for not using ¹⁸FDG-PET-CT for treatment evaluation more widely are the limited availability of PET-CT and the lack of sufficient standardization.¹⁰⁹ International multicenter studies will probably accelerate the process of standardization.

Magnetic resonance imaging is increasingly investigated for therapy evaluation of neoadjuvant therapies to guide organ-preserving surgery in breast, rectal, and esophageal cancers.^79,110-113 Most clinical MRI studies concerning therapy evaluation for lung cancer used DW-MRI and demonstrated the potential of (early) therapy response evaluation and prediction.^114,115 Two studies used DW-MRI to assess the early treatment response and predict prognosis early during the treatment.^114,115 Yabuuchi et al analyzed 28 patients with stage IIIB and IV NSCLC who received platinum-based chemotherapy, performing a DW-MRI at baseline and after 1 course of chemotherapy. The ADC value increased during treatment and correlated significantly, with a decrease in tumor size on CT (r2 = .41). More importantly, the increase in ADC identified a good and poor prognostic group, which significantly correlated with the OS and progression-free survival (PFS).¹¹⁵ Tsuchida et al¹¹⁴ showed similar results in 28 patients with stage IIIB and IV NSCLC (n = 17) and small-cell lung cancer (n = 11) who received platinum-based chemotherapy and underwent DW-MRI and 18FDG-PET before and after 1 course of chemotherapy. The increase in ADC and SUV_mean was significantly correlated with responders and nonresponders (based on PFS and OS), as opposed to no significant correlation to the change in tumor diameter on CT. The results of these studies suggest that the ADC change precedes morphological changes (eg, “tumor shrinkage”).

The first steps for the implementation of MRI for radiation treatment planning in lung cancer have been taken with the current MRI techniques. However, the focus of future research should be on image registration, image guidance, and treatment adaptation.