Magnetic resonance imaging in lung: a review of its potential for radiotherapy

Abstract

MRI has superior soft-tissue definition compared with existing imaging modalities in radiation oncology; this has the added benefit of functional as well as anatomical imaging. This review aimed to evaluate the current use of MRI for lung cancer and identify the potential of a MRI protocol for lung radiotherapy (RT). 30 relevant studies were identified. Improvements in MRI technology have overcome some of the initial limitations of utilizing MRI for lung imaging. A number of commercially available and novel sequences have shown image quality to be adequate for the detection of pulmonary nodules with the potential for tumour delineation. Quantifying tumour motion is also feasible and may be more representative than that seen on four-dimensional CT. Functional MRI sequences have shown correlation with flu-deoxy-glucose positron emission tomography (FDG-PET) in identifying malignant involvement and treatment response. MRI can also be used as a measure of pulmonary function. While there are some limitations for the adoption of MRI in RT-planning process for lung cancer, MRI has shown the potential to compete with both CT and PET for tumour delineation and motion definition, with the added benefit of functional information. MRI is well placed to become a significant imaging modality in RT for lung cancer.

INTRODUCTION

Radiotherapy (RT) plays a significant role in the treatment of lung cancer¹ and relies on accurate imaging for precise treatment delivery. Improvements in imaging and the use of multimodality imaging have improved tumour delineation for RT planning and treatment of lung cancer.^2,3

CT is the standard imaging modality in RT with a relatively high spatial resolution, but limited specificity. Tumour definition on CT can be obscured in the presence of adjacent lung collapse or consolidation. The incorporation of positron emission tomography (PET) with glucose analogue flu-deoxy-glucose (FDG) tracer has significantly improved the discrimination between benign and malignant tissue.⁴ The use of FDG-PET has reduced gross tumour volumes (GTV) owing to the improved differentiation between benign and malignant tissue^5–7 and reduced interobserver variability in the delineation of GTV.^4–13 While FDG-PET with CT is currently the standard of care for tumour delineation, there are limitations. The spatial resolution of PET is poor, ranging between 5 and 7 mm compared with 2 mm for CT.¹⁴ This low spatial resolution results in blurred edges, and tumours <4 mm may be falsely negative on FDG-PET scans. There is also a lack of consensus on the FDG-PET visualization method within RT, with a number of different methods reported in the literature.^6–9,13,15 Use of ionizing radiation to acquire images for PET and CT can be a limiting factor in repeated examinations.

The mobility of tumour and normal anatomy during respiration can lead to a large degree of uncertainty in tumour position. In order to visualize and quantify tumour motion, a number of options have been utilized including fluoroscopy, slow CT scans and breath-hold devices.¹⁶ However, these have largely been superseded by respiratory-correlated CT or four-dimensional CT (4DCT).¹⁷ 4DCT image acquisition is based on acquiring CT images with an external respiratory trace while the patient is free breathing. During post processing, the acquired images are correlated with an external respiratory signal usually either in the form of a reflective marker and camera system or a pressure-sensing belt around the abdomen. The advent of 4DCT in RT planning has overcome some of the problems associated with imaging the thorax¹⁸ and has allowed for the definition of patient-specific margins for tumour motion.^19–21 However, accurate definition of motion on 4DCT is reliant on a consistent respiratory cycle.²²

MRI is a well-established diagnostic tool in oncology.²³ Unlike CT where tissue contrast primarily depends on electron density, MRI contrast can be varied extensively by imaging other intrinsic properties of the tissue (e.g. spin lattice and spin–spin relaxation time, proton density, diffusion etc.). Typically, a MRI examination will consist of multiple series of scans in several imaging planes using different pulse sequences which exploit these properties.²⁴ This allows flexibility in facilitating optimal tumour visualization and evaluation.²³ The benefit of MRI in delineating soft tissues has been demonstrated for a number of disease sites in RT^23,25,26 and is being incorporated into the treatment-delivery process with the development of MRI linear accelerators,^27–29 making it a significant imaging modality in future. The use of MRI in the lung has been complicated by firstly by respiratory motion and also low proton density of the lung tissue, which can reduce the signal-to-noise ratio and increase magnetic susceptibility effects.³⁰

Improvements in technology (most notably parallel imaging) resulting in faster acquisition times and better respiratory-gating techniques (e.g. navigator echoes) have significantly improved the quality of lung MRI.^31,32 Development of any MRI protocol for the lung needs minimization of the impact of susceptibility and motion artefacts on the image quality. A lung protocol for diagnostic imaging has been described^31,33 and it consists of a combination of T₂ and T₁ weighted images (Figure 1). T₂ weighted images highlight tumour infiltration and nodular lesions or masses with high fluid content, and T₁ weighted images cover high-signal pulmonary nodules and masses. For the detection of mediastinal lymph nodes, a T₂ weighted scan with fat saturation is recommended. A healthy lung generally has low signal intensity on MRI, and the presence of nodules or masses with increased proton density improves signal and contrast with the surrounding lung and therefore potential detection.^33,34

Figure 1.

Comparison of 3-T axial T₁ weighted breath-hold at inspiration gradient-echo sequence (volumetric interpolated breath-hold) (a) and T₂ weighted turbo spin-echo navigator sequence (half-Fourier acquisition single-shot turbo spin echo) at the exhale phase (b). The image in (b) demonstrates tumour infiltration into the mediastinum abutting the superior aspect of the aortic arch.

OBJECTIVE

The aim of this study was to review the current status and developments in lung MRI in order to evaluate its potential role in: (i) target volume delineation, (ii) tumour motion quantification (iii) and functional imaging for lung cancer RT and to identify the sequences necessary to achieve these.

METHODS AND MATERIALS

A literature search was performed using PubMed, Medline and Google Scholar using the terms listed in Table 1 for articles published between 1990 and September 2014. The keyword combinations, A–H, listed in Table 1 were entered into the search strategies for each of the databases defined above to identify appropriate literature. The literature review was limited to articles written in English and on human subjects. The results were grouped according to three major areas: (1) MRI-based anatomical imaging with the potential for tumour delineation, (2) MRI-based tumour motion analysis and (3) MRI-based functional imaging.

Table 1.

Search keywords

Keywords	Keyword combinations
(1) Lung (2) Cancer (3) Thorax (4) Thoracic (5) Radiotherapy (6) Radiation therapy (7) MRI (8) MRI (9) MR (10) Anatomical (11) Motion (12) Functional	A: (1 and 2), (5 or 6) and (7 or 8 or 9) B: (1 and 2) and (7 or 8 or 9) C: (3 or 4) and (5 or 6) and (7 or 8 or 9) D: (3 or 4) and (7 or 8 or 9) E: (A) and (10 or 11 or 12) F: (B) and (10 or 11 or 12) G: (C) and (10 or 11 or 12) H: (D) and (10 or 11 or 12)

RESULTS

30 publications were identified which met the selection criteria, 3 publications on the anatomical detection of tumour (Table 2), 9 publications on motion analysis using MRI (Table 3) and 18 publications on functional MRI for lung tumour and pulmonary nodules including functional imaging of healthy lung (Table 4). The majority of the studies were conducted on 1.5-T scanners using a combination of breath-hold and free-breathing scans with and without respiratory and/or cardiac gating. Most protocols used fast-imaging-sequence variants of both the gradient-echo (GRE) or turbo-spin echo techniques. In most cases, parallel imaging—using radiofrequency coil encoding—was applied to further reduce the scan time and limit the breath-hold duration required. Table 5 highlights the basic sequences used and their applications and limitations in lung imaging. While there was evidence available for lung cancer imaging from a diagnostic perspective, there was limited evidence in its application in radiation oncology imaging.

Table 2.

Literature summary of scan protocols for studies evaluating MRI-based anatomical detection of lung cancer

Reference	Scanner	Protocol	Acquisition plane	Breathing manoeuvre	Physiology assessed
Biederer et al35	1.5T (Siemens)	3D GRE VIBE	Coronal	Breath-hold 20 s	Vessels and airways
Bruegel et al36	1.5T (Siemens)	T2 HASTE, T2 IR-HASTE, T2 TSE, STIR, VIBE	Axial	Breath-hold 14–19 s at end inspiration	Pulmonary lesions
		STIR		Respiratory and pulse triggered
Chin et al37	3T (Phillips)	T2 triple-inversion black-blood TSE, T1 3D TSE	Axial	Breath-hold 16 s	Pulmonary nodules

3D, three-dimensional; GRE, gradient-recalled echo sequence; HASTE, half-Fourier acquisition single-shot turbo spin echo; IR-HASTE, inversion recovery HASTE; STIR, short-tau inversion recovery; TSE, turbo spin echo; VIBE, volumetric interpolated breath-hold.

GE; Milwaukee, WI; Phillips, Amsterdam, Netherlands; Siemens, Erlangen, Germany.

Table 3.

Literature summary of studies evaluating MRI-based tumour motion

Reference	Scanner	Protocol	Acquisition plane	Breathing manoeuvre	Physiology assessed
Biederer et al38	1.5T (Siemens)	3D GRE	Coronal	Phantom study	Porcine heart and lung collocated into a chest phantom
Cai et al39	1.5T (Siemens)	TrueFISP	Sagittal	Quiet breathing	Tumour and lung motion
Cai et al40	1.5T (Siemens)	TrueFISP	Sagittal	Normal breathing cycle—300-s continuous scan	Tumour and healthy lung
Koch et al41	1.5T (GE)	FSE fGRE	Sagittal coronal and axial	NA	Phantom
Liu et al42	1.5T (GE)	fGRE—modified	Axial, sagittal and coronal	Free breathing	Pulmonary vessels
Plathow et al43	1.5T (Siemens)	TrueFISP	Lung motion—coronal; tumour motion—sagittal, coronal	Quiet tidal breathing followed by maximum inspiration and expiration	Tumour volume, lung volume
Plathow et al44	1.5T (Siemens)	TrueFISP	Lung motion—coronal; tumour motion—sagittal, coronal and axial	Quiet tidal breathing followed by maximum inspiration and expiration	Lung and tumour volume
Blackall et al45	1.5T (Phillips)	SSFP	Coronal	Breath-hold 15 s at tidal inhalation and exhalation	Lung and tumour motion
		FFE		Free breathing
Koch et al46	1.5T (GE)	fGRE—modified	Axial, sagittal and coronal	Free breathing	Pulmonary vessels

3D, three-dimensional; FFE, fast field echo; fGRE, fast gradient echo; FSE, fast spin echo; GRE, gradient echo; NA, not applicable; SSFP, steady-state free precession; TrueFISP, true fast imaging with steady-state precession.

GE; Milwaukee, WI; Phillips, Amsterdam, Netherlands; Siemens, Erlangen, Germany.

Table 4.

MRI-based functional imaging

Author	Scanner	MRI protocol	b-value (s mm−2 )	ADC parameter	Acquisition plane	Breathing manoeuvre	Image reference	Outcome
Lichy et al47	1.5T (Siemens)	DWI (EPI with SE), respiratory gated	0, 400, 1000	NS	Transverse	Free breathing	FDG-PET (uptake value, NS)	Metastatic disease DWI at b=1000 s mm−2 corresponded to that on FDG-PET images; ADC maps did not improve nodal detection
Mori et al48	1.5T (Phillips)	DWI (EPI with SE)	1000	ADCmin	Transverse	Shallow breathing	FDG-PET (SUVmax, SUV-CR)	Inverse correlation between ADCmin and SUV-CR; DWI had higher sensitivity in the presence of inflammation
Pauls et al49	1.5 T (Siemens)	DWI (EPI with SS)	0, 400, 800	NA	Transverse	NS	FDG-PET (uptake—NS)	T1 DCE images were equivalent to DWI in nodal detection when compared with PET; DCE and DWI understaged nodal disease when compared with PET
		DCE (T1 2D GE+FLASH)	NA	NA	Transverse	NS
Abdel Razek et al50	1.5T (Siemens)	DWI (EPI with SE)	0, 300, 600	ADCmean	Transverse	NS	Histology	Mean ADC values for malignant disease significantly lower than benign nodes
Xu et al51	1.5T (Phillips)	DWI (EPI with SE)	0, 1000	ADCmean	Transverse	Free breathing	Histology	Visual detection of malignant nodes was higher than that of benign nodes; subsequently, ADC values were significantly lower for malignant nodes when compared with benign nodes
Wang et al52	1.5T (Siemens)	IVIM—DWI (EPI with SS) respiratory gated	0, 5, 10, 15, 20, 25, 50, 80, 150, 300, 500, 800	NA	Transverse	Free breathing	FDG-PET histology	IVIM parameters were lower and DCE parameters were higher in the presence of disease when compared with consolidation; however, there was poor correlation between IVIM and DCE parameters
		T1 3D VIBE	NA	NA	Transverse	Breath-hold
Yang et al53	1.5T (GE)	DWI (EPI with SS SE + ASSET)	500	ADCmean	Transverse	Breath-hold	FDG-PET	DWI images were equivalent to PET in the differentiation between tumour and atelectasis; T2 images allowed differentiation between tumour and atelectasis then T1; SCLC had higher ADC values than SCC and AC
		T1 SE; respiratory and cardiac gated	NA	NA	Transverse	Free breathing
		T2 FRFSE respiratory and cardiac gated	NA	NA	Transverse	Free breathing
Ohno et al54	3T (GE)	DWI (EPI with STIR)	1, 1000	ADCmean	Transverse	Free breathing	FDG-PET (SUVmax)	DWI (ADC) may have better potential than FDG-PET (SUVmax) for the prediction of tumour response during chemoradiotherapy for NSCLC
Chang et al55	3T (GE)	DWI (EPI with ASSET)	600	ADCmean	Transverse	Breath-hold	NA	Patients who responded to treatment demonstrated increased ADC values; those who had no treatment response had a slight decrease in ADC
Yabuuchi et al56	1.5T (Phillips)	DWI (EPI) respiratory and pulse-gated	0, 1000	ADCmean	Transverse	Free breathing	NA	Correlation between early ADC change and post-treatment tumour size; better progression-free survival in patients with increased ADC
		DCE with SE				Breath-hold
Iizuka et al57	1.5T (Siemens)	DWI (EPI) respiratory gated	0, 500, 1000	ADCmedian	Transverse	Free breathing	FDG-PET (SUVmax)	No correlation noted between SUVmax and ADCmedian; however, patients with low ADC value and high SUVmax demonstrated disease progression
Chen et al58	3T (Siemens)	DWI (EPI with SE)	50, 1000	ADCmean, ADCmedian and ADCmin	Transverse	Free breathing	Histology	Significant inverse relationship between tumour cellularity and ADC mean and minimum values
Hunter et al59	1.5T (GE)	T1 DCE with EPI T1 SE	NA	NA	Transverse	Quiet breathing	FDG-PET (MRGIc)	Tumour vascular physiology as measured on DCE correlated with glucose metabolism; changes in vascular physiology noted during and after treatment
Regier et al60	1.5T (Phillips)	EPI with SE; respiratory triggered	0, 500	ADCmin and ADCmean	Transverse	Free breathing	FDG-PET (SUVmax and SUVmean)	Inverse correlation between ADCmin and SUVmax
		T1 3D GRE	NA	NA		Breath-hold (<20 s)
Pauls et al61	1.5T (Siemens)	T1 sGRE	NA	NA	Transverse	Maximum end inspiration breath-hold	Histology	DCE parameters allowed differentiation between tumour subtypes
Iwasawa et al62	1.5T (GE)	DCE with SE	NA	NA	Not specified	Breath-hold		DCE MRI with perfusion data allows evaluation of pulmonary perfusion
Plathow et al63	1.5T (Siemens)	TrueFISP	NA	NA	Coronal, sagittal and transverse	Free breathing followed by maximum inspiration and expiration	Spirometry	MRI dimensions and spirometry data demonstrated significant correlation
Shiabata et al64	1.5T (Philips)	bFFE	NA	NA	Mid-sagittal plane	Forced deep breathing	Spirometry	Correlation between spirometry and respiratory motion

2D, two-dimensional; 3D, three-dimensional; AC, adenocarcinoma; ADC, apparent diffusion coefficient; ADC_mean, mean apparent diffusion coefficient; ADC_median, median apparent diffusion coefficient; ADC_min, minimum apparent diffusion coefficient; ASSET, array spatial sensitivity encoding technique; bFFE, balanced fast-field echo; DCE, dynamic contrast-enhanced; DWI, diffusion-weighted imaging; EPI, echoplanar imaging; FLASH, fast low-angle shot; FDG, fludeoxyglucose; FRFSE, fast-recovery fast spin echo; GE, gradient echo; GRE, gradient echo; IVIM, intravoxel incoherent motion; MRGIc, glucose metabolic rate; NA, not applicable; NS, not specified; NSCLC, non-small-cell lung cancer; PET, positron emission tomography; SCC, squamous cell carcinoma; SCLC, small-cell lung cancer; SE, spin echo; sGRE, spoiled gradient echo; SS, single shot; STIR, short-tau inversion recovery; SUV, standardized uptake value; SUV-CR, contrast ratio for standardized uptake value; SUV_max, maximum standardized uptake value; SUV_mean, mean maximum standardized uptake value; TrueFISP, true fast imaging with steady-state free precision; VIBE, volumetric interpolated breath-hold.

GE, Milwaukee, WI; Phillips, Amsterdam, Netherlands; Siemens, Erlangen, Germany.

Table 5.

MRI sequence adaption for lung imaging

Sequence description	Type	Challenges for lung	Further improvements
T2 weighted anatomy	Fast spin echo with gating (e.g. TSE/FSE or HASTE)	Long scan time requires gating or multiple breath-holds	Self-navigation using amplitude or phase
T1 weighted anatomy	3D volume gradient echo (e.g. VIBE, LAVA) without gating	Breath-hold duration	Parallel imaging, partial k-space to reduce time of scan
Real-time motion	Steady state (e.g. TrueFISP, bSSFP, FIESTA)	Off-resonance and cardiac artefacts	Select appropriate FOV and use specific cardiac shim
Diffusion	Echoplanar imaging	EPI artefacts; low spatial resolution	Shaped excitation to reduce volume of the tissue
Perfusion	Dynamic fast gradient echo (e.g. FLASH, FSPGR) with contrast	Has to be run in one or multiple breath-holds depending on requirements	Radial k-space to reduce motion artefacts with/without motion correction

3D, three-dimensional; EPI, echoplanar imaging; FIESTA, fast imaging employing steady-state acquisition; FLASH, fast low-angle shot; FSE, fast spin echo; FSPGR, fast spoiled gradient recall echo; FOV, field of view; HASTE, half-Fourier acquisition single-shot turbo spin echo; LAVA, liver acquisition with volume acceleration; TrueFISP, true fast imaging with steady-state precession; TSE, turbo spin echo; VIBE, volumetric interpolated breath-hold.

MRI-based anatomical imaging with the potential for tumour delineation

No studies specifically looked at lung tumour delineation based on MRI; however, three articles investigated the utility of MRI in detecting pulmonary nodules.^35–37 A T₁ weighted three-dimensional (3D) spoiled GRE sequence with volumetric interpolated breath-hold (VIBE) demonstrated a pulmonary nodule-detection rate comparable with CT. However, with participants who were not compliant with the breath-hold manoeuvre, blurring and ghosting artefacts impacted image quality.³⁵ Two studies evaluated breath-hold T₁ and T₂ weighted turbo spin-echo sequences to detect pulmonary nodules in the axial plane.^36,37 MRI was able to detect nodules between 5 and 10 mm.³⁷

The main challenges for imaging in the lung are susceptibility and motion artefacts. The sequences described for pulmonary nodule detection have different applications and limitations in lung imaging. VIBE^35,36 is a 3D fast T₁ weighted spoiled GRE sequence, which is particularly robust in the presence of cardiac pulsation. With its high spatial resolution, it allows the detection of relatively small pulmonary nodules in 3D. The inherent speed of this sequence means it can be acquired in a single breath-hold.⁶⁵ It is therefore ideal in pulmonary imaging, where the entire thorax can be acquired in one breath-hold. It is, however, very sensitive to respiratory motion, which can result in significant motion-related artefacts. Being a GRE sequence, it is also prone to susceptibility artefacts especially at 3 T. T₂ weighted sequences suffer from longer acquisition times (Table 5) and therefore require gating or multiple breath-holds to be employed. To help further reduce the scan time of fast spin echoes, half-Fourier acquisition single-shot turbo spin echo has been used.³⁶ These sequences combine the speed-up advantages of acquiring half of the k-space data together with a single repetition time (TR) or “shot”.^65. However, this makes it prone to blurring, which may be an issue when defining the tumour volume for RT. Inversion-recovery (IR) sequences³⁶ with black blood contrast have also been investigated. IR protocols allow robust differentiation between the fat and water owing to the added inversion time to null the signal from the fat rather than using radiofrequency-suppression pulses. This makes it ideal for identifying the tumour and nodal volume close to the mediastinum as subcutaneous fat and bone marrow of the ribs will be saturated, allowing for greater visibility of the area of oedema and inflammation. However, it has “longer acquisition times caused by longer TRs, low signal-to-noise and tissues with similar T₁ values will all be suppressed”.⁶⁶

Miller et al³² also highlighted the use of ultrashort echo-time (UTE) sequences to improve structural imaging. The development of UTE offers two specific advantages in lung imaging; acquiring signal at tens of microseconds minimizes susceptibility artefacts and allows image acquisition at extremely short relaxation times in lung. The resultant image provides similar contrast to that of CT. To date, this sequence has not been studied for the definition of tumour or pulmonary nodules.

MRI-based tumour motion

To capture respiratory motion during free breathing, real-time imaging is required. The majority of the studies assessing tumour motion utilized a variation of GRE sequences.^38–44 While there is evidence to suggest that MRI data are prone to geometric distortion in the presence of motion,³⁰ two studies demonstrated results contrary to this.^41,45 Imaging in the sagittal and coronal planes demonstrated minimal error when compared with the axial plane using a thoracic phantom and a fast gradient-echo (fGRE) sequence.⁴¹ The integrity of the structures was maintained on free-breathing real-time MRI scans, while a large error was noted for intracycle tidal volume reproducibility on breath-hold scans.⁴⁵

fGRE sequences were shown to be feasible in assessing lung motion; however, pulmonary vessels rather than lung tumour were used for assessment.^42,46 Respiratory mechanics of the lung and tumour can also be assessed with a true fast imaging with steady-state precession (TrueFISP) sequence.^43,44 Plathow et al⁴⁴ demonstrated a variation in motion between tumour- and non-tumour-bearing hemithorax with the motion magnitude varying according to the tumour location. Variation in the tumour motion before and after RT was also assessed and showed no change, although there was a reduction in the craniocaudal motion of the tumour-bearing hemithorax.⁴³ The change in the craniocaudal motion of the tumour-bearing hemithorax was not reflected in spirometry results.

Two studies compared the motion measured on real-time MRI with that seen on 4DCT with conflicting results.^38,39 In a phantom study using GRE sequences, Bieder et al³⁸ showed that the lesion diameter was larger but the lesion displacement smaller on MRI than that on 4DCT. Cai et al³⁹ investigated the internal target volume error between real-time MRI and simulated 4DCT data. The results indicated that owing to the nature of 4DCT acquisition, the excursion of tumour motion may not be accurately depicted on a 4DCT scan. The magnitude of internal target volume error correlated with the variability in participants' breathing. Acquiring tumour motion data over a prolonged period was shown to be more accurate than limited breathing cycles.⁴⁰

Steady-state-free precession sequences are ultrafast GREs designed around very short TRs and have demonstrated the required temporal and spatial resolution to acquire multiple images during free breathing to allow quantification of tumour motion.⁶⁵ These sequences are highly dependent on field homogeneity; it is therefore essential to perform shimming of the heart prior to acquisition to help minimize off-resonance artefacts.⁶⁵ Figure 2 demonstrates a TrueFISP sequence with and without the cardiac shim, which helps minimize artefacts. Trade-offs in temporal resolution between two-dimensional (100 ms) and 3D imaging (1 s) can lead to blurring, unless a slow-breathing manoeuvre is performed.⁶⁷

Figure 2.

3-T coronal true fast imaging with steady-state precession (TrueFISP) images without (a) and with (b) dedicated cardiac shim to minimize off-resonance artefacts.

For tumour motion assessment, an imaging protocol with ultrafast GRE sequences and parallel imaging is appropriate (Table 5). These usually consist of steady-state sequences (such as TrueFISP, steady-state free precession etc.) for optimum temporal resolution.

MRI-based functional imaging

Currently, functional imaging in lung RT satisfies two purposes, identification of nodal disease and differentiation between the tumour and surrounding consolidation. This, in most cases, is achieved with FDG-PET imaging. The most commonly employed functional MRI techniques are diffusion-weighted imaging (DWI) and dynamic contrast-enhanced (DCE) imaging. Studies looking at functional imaging included both the tumour and healthy lung tissue.

DWI is based on sensitizing the sequence to the motion of water molecules at a microscopic level (described by the b-value of the image). This motion may be quantified by generating parametric maps using at least two different b-value images and calculating the apparent diffusion coefficient (ADC). Qualitative interpretation of DWI is based on the visual assessment of signal intensity on a high b-value image set; a region of high signal intensity depicts restricted diffusion in the extracellular space. Areas of restricted diffusion will translate to areas of low values on the resulting ADC map.

A number of studies utilized DWI to assess the presence of malignant lymph nodes^47–51 and detection of tumour in the presence of consolidation.^52,53 Most studies compared DWI with FDG-PET imaging. DWI was not able to improve the detection of metastatic mediastinal nodal disease for lung cancer compared with PET;^47,48 however, it had higher specificity in the presence of inflammation.⁴⁸ Pauls et al⁴⁹ showed that DWI had 80% agreement with PET for nodal stage with 15% of the cases understaged and 5% of the cases overstaged. In those cases where MRI overstaged the nodal disease, restricted diffusion was noted in both mediastinal and supraclavicular lymph nodes (4–7 mm), with no evidence of elevated glucose metabolism. Nodal disease was adjacent to the primary tumour volume, in cases where MRI understaged the disease.⁴⁹ It should be noted that neither study had pathological correlation of the imaging results; both were only assessing the agreement between the two imaging modalities.^48,49 Therefore, there is potential for PET to be false negative in the case of small tumours.

DWI can also differentiate between malignant and benign mediastinal lymph nodes.⁵⁰ Malignant node detection on pre-operative DWI was compared with histologically confirmed malignant lymph node status post operatively.⁵¹ A whole-body version of DWI termed DWIBS (diffusion-weighted imaging with background signal suppression) has been used to produce images that are visually similar to FDG-PET,⁵¹ and the visual detection of malignant nodes on the resultant images was significantly higher for both enlarged and normal-size lymph nodes. ADC values also correlated with the visual detection rate.⁵¹

Two studies demonstrated the potential of DWI to differentiate lung cancer from consolidation. Yang et al⁵³ compared DWI with FDG-PET. DWI was able to detect the difference between tumour and consolidation in all patient cases, based on the hyperintensity of the tumour. The ADC map also demonstrated lower values in the presence of the tumour. An intravoxel incoherent motion (IVIM) sequence was also able to differentiate between tumour and consolidation as compared with both DCE and FDG-PET.⁵² IVIM is a modified DWI technique in which images are acquired with lower than conventional b-values that are sensitive to blood microcirculation. Both DCE and IVIM were able to distinguish between cancer and consolidation; however, there was a poor correlation between IVIM and DCE parameters.

There is potential for the ADC map to detect early treatment-related changes better than FDG-PET.^54,55 Increase in the ADC value in the early phase of treatment correlated with final tumour size reduction, indicating potential use in detecting early treatment response.⁵⁶ Median progression-free survival in patients with increased ADC change was shown to be 12.0 months compared with 6.7 months for those patients where ADC remained stable or decreased.⁵⁶ To predict disease progression following stereotactic RT for stage I non-small-cell lung cancer, Iizuka et al⁵⁷ performed pre-treatment DWI and FDG-PET. Patients with low ADC value and higher SUV_max had greater disease progression, but results were not statistically significant. From a slightly different perspective, Chen et al⁵⁸ demonstrated an inverse relationship between minimum and mean ADC values and tumour cellularity.

The main application of DWI has primarily been in imaging neurological disorders. However, there is increasing evidence to utilize it in imaging for cancer detection and treatment monitoring.⁶⁸ Echoplanar imaging (EPI) sequences are commonly used for DWI but are prone to susceptibility distortion and ghosting artefacts.⁶⁵ While a breath-hold scan can be performed with EPI to eliminate motion artefact using only a single b-value, it is more common practice to acquire two or more b-values for the quantification of tumour diffusion. There is potential to improve EPI for lung DWI (Table 5) in the lung, by reducing the volume of the excited tissue and limiting artefacts from tissue outside this field of view. Figure 3 illustrates an example of DWI using EPI combined with reduced excitation.

Figure 3.

Axial diffusion-weighted imaging (echoplanar imaging with ZOOMit) with b-value 0 (a), 250 (b), 500 (c) and 750 (d) of a 69-year-old male patient with lung cancer diagnosed with large-cell carcinoma of the left upper lobe. There is a marked area of hyperintensity in the left upper lobe characteristic of restricted diffusion. Artefacts are present, but the image quality is sufficient for tumour edge definition (d) and apparent diffusion coefficient measurement (e).

DCE involves the acquisition of images before, during and after administration of a suitable contrast agent. Data may be evaluated and quantified in a number of ways, from simple measurements to complex pharmacokinetic modelling.⁶⁹ Highly perfused regions demonstrate a high and rapid uptake and washout of gadolinium-based contrast. DCE for perfusion assessment of lung cancer requires high temporal resolution in order to adequately assess the enhancement of the tumour volume. fGRE sequences are generally utilized with partial or shared k-space approaches to optimize temporal resolution and run with multiple short breath-hold manoeuvres over the required time course. It has recently become feasible to acquire this data during free breathing by using a radial stack of stars (Table 5), sampling scheme to compensate or even correct for motion.⁷⁰ Free-breathing perfusion data have been shown to be as reproducible as breath-hold and also better tolerated. However, further investigation is required for adaption in lung imaging.⁷¹

Two studies highlighted a possible relationship between metabolic activity and cellularity.^59,60 Tumour vascularity⁵⁹ on DCE and restricted diffusion⁶⁰ as measured on ADC were found to be correlated to increased FDG uptake or SUV_max on PET scans. Hunter et al⁵⁹ also demonstrated changes in vascular physiology which were apparent during and after treatment, highlighting the potential role in clinical management. DCE perfusion parameters can also allow identification of histological subtypes for lung cancer.⁶¹ A number of contrast-uptake parameters were used to investigate correlation with tumour subtypes, and these played a significant role in differentiating non-small-cell lung cancer (NSCLC) from small-cell lung cancer (SCLC). Time-dependent kinetic parameters were more relevant in differentiating adenocarcinoma from squamous-cell carcinoma.

Functional MRI data can also be used to assess healthy lung function prior to the course of treatment. Iwasawa et al⁶² evaluated whether functional MRI could predict post-operative lung function. A correlation was seen between the perfusion ratio on MRI and radionuclide study (scintigraphy) and also between the predicted FEV1 on MRI and the tested FEV1 post operatively. MRI can also be a useful tool in evaluating respiratory mechanics and volumetry. Using the diaphragm and chest wall motion as measured on MRI imaging, Plathow et al⁶³ created a volumetric model to calculate vital capacity and compared this with spirometry-based vital capacity measurement as lung function tests. They were able to show a strong correlation between the vital capacity measurement on MRI and that measured on spirometry. Similarly, Shibata et al⁶⁴ investigated the difference in respiratory motion between healthy individuals and patients with chronic obstructive pulmonary disease, tracking pulmonary vessels using an in-house developed algorithm. MRI-based respiratory motion correlated strongly with spirometry data.

While not yet in routine clinical use, MRI using hyperpolarized gases or oxygen enhancement can allow for ventilation studies of the lung.³⁴ For hyperpolarized imaging, two noble gases are predominantly used, 3-helium (³He) and 129-xenon (¹²⁹Xe). ³He has been used to asses ventilation in post-treatment assessment^72,73 or identifying well-ventilated lung for avoidance of functional lung at the time of treatment planning.^{74,75 129}Xe has the added advantage in that it can be imaged following ventilation perfusion and it has been shown to highlight gas exchange impairment.⁷⁶ Oxygen-enhanced MRI utilizes pure oxygen as the contrast agent to study ventilation abnormalities.⁷⁷

DISCUSSION

Improvements in MRI technology continue to enable the imaging of low-proton-density lung tissue in the presence of respiratory and cardiac motion. The main challenges of lung RT are accurate identification of the GTV and nodal volume, particularly in the presence of pathological changes in the surrounding lung tissue. Accurate measurement of tumour motion is also necessary to ensure that the RT field encompasses appropriate margins. MRI has the potential to help overcome both these challenges. Table 6 highlights the potential application of MRI in the RT process of lung cancer.

Table 6.

Image modality comparison table

Radiotherapy end point	Imaging modality
	CT	PET	MRI
Tumour volume delineation	✓	×	✓35–37
Tumour motion assessment	✓16–21	×	✓38–46
Functional information	×	✓4	✓47–63

PET, positron emission tomography.

However, the universal application of MRI in lung RT is currently limited by the ability to generate treatment plans on MRI data sets and create reference images for treatment verification. Currently, MRI is used for specific lung cancer cases (e.g. Pancoast tumours) to aid in anatomical delineation; however, the CT data set is still used for treatment planning. The image registration process between CT and MRI is prone to error owing to changes in the position and shape of the organ and tumour, even if the scans are contemporaneous. A MRI-only RT workflow is being considered for many RT sites and will be necessary for MRI-guided RT systems such as the MRI linear accelerator. RT treatment-planning systems still require CT or CT-equivalent data for dose calculation. A number of methods have been proposed to derive CT equivalent data from MR images.⁷⁸ Assigning bulk density to the entire patient anatomy is the simplest solution, and has been shown to give acceptable dosimetric accuracy.⁷⁹ The previously mentioned UTE sequences may also have a role in improving bone and tissue classification and direct CT conversion approaches, which makes their application in the lung even more interesting.⁷⁹

In addition, for lung GTV delineation, detection of a pulmonary nodule or mass is just as important as being able to define the boundary of the nodule or mass. To allow volume definition, tumour infiltration into the chest wall and or mediastinum is required along with detection of involved mediastinal lymph nodes. A combination of sequences is required to facilitate GTV delineation. T₂ weighted images such as half-Fourier acquisition single-shot turbo spin echo (HASTE) or short-tau inversion recovery (STIR) can demonstrate tumour infiltration. T₁ weighted images are ideal for identifying mediastinal lymph nodes. A limitation of these sequences for anatomical detection is that they are either breath-hold or respiratory-gated images. The exhale phase in respiratory gating with bellows or navigation is generally used, as this is the longest period of relaxation during the respiratory cycle. Knowledge regarding the deformation of tumour volume during the respiratory cycle is important. Hence, a more dynamic approach to imaging is required or at the very least anatomical scans at inspiration and expiration to capture the position and shape of the tumour at the extremes of the respiratory cycle. Imaging requirements for RT are different from those of diagnostic imaging. Imaging for RT serves the purpose of tumour and associated nodal volume delineation rather than detection and staging via diagnostic imaging. As such, these anatomical image sequences need to be further assessed in a radiation oncology setting.

There is evidence supporting the value of MRI in quantifying lung tumour motion. However, the data reported are predominantly based on sagittal and coronal planes, imaging planes that in most cases are not compatible with the majority of RT-planning systems. Furthermore, image acquisition is generally in two dimensions with either a single plane or the given number of planes through the region of interest. This can potentially neglect any out-of-plane tumour motion during respiration and introduce geometric errors in the planning process. Further study into 3D registration and incorporation of non-axial data into treatment-planning systems is needed. While 4DCT remains the gold standard for lung tumour motion, it is limited by the number of breathing cycles acquired and any irregularity in breathing. Real-time MRI can be acquired over a greater number of respiratory cycles to better understand and capture motion over time and is not influenced by irregular breathing patterns as 4DCT currently is.

In terms of functional imaging, FDG-PET remains the gold standard in defining metabolically active disease, particularly in mediastinal nodes and in the presence of consolidation. There is evidence to support DWI with ADC mapping and DCE MRI as an alternative to FDG-PET for assessing functional tumour activity. However, studies to date have all been from a diagnostic perspective, where the end points are disease presence or absence. For RT, the definition of the malignant target volume is necessary. Further research is needed to determine whether MRI will be a complementary or competing technology for PET imaging in the lung. The availability of hybrid PET-MR systems also offers promise for lung imaging.⁸⁰

While the use of ADC and DWI imaging has been reported on extensively for lung cancer, it should be noted that reproducibility of quoted ADC parameters such as the minimum, maximum and mean is a consideration. Kivrak et al⁸¹ demonstrated this in their study of ADC values across six different MRI scanners with a phantom. However, rather than looking at ADC values, the analysis of histogram distributions⁸¹ shows potential and may be better for comparing data acquired on different scanners.⁸² Conversely, DWI is reproducible between scanners and has shown good interobserver and intraobserver agreement for lung cancer for tumour sizes >2 cm.⁸³ The ability to differentiate histological subtypes is potentially useful for patients in whom biopsy confirmation of lung cancer is not possible, usually owing to underlying lung disease. Histological subtype is important in determining treatment, and MRI parameters could have a potential role in this.⁶¹

However, improvement in the standard EPI technique is required to overcome some of the current limitations of susceptibility and motion artefacts.

Fundamentally, breath-hold MRI scans are challenging for patients with lung cancer owing to their already limited respiratory function. Further improvements in image technology and navigation for gating may make breath-hold scans obsolete in these patients.

Lung function tests prior to the start of any treatment to some extent dictate whether a patient is able to receive radical or palliative RT. This is usually based on spirometry to assess lung function. However, two studies^63,64 highlighted the potential of MRI to predict lung function in patients, which could be useful in patients who cannot undergo spirometry. Hyperpolarized gas- and oxygen-enhanced MRI can potentially allow for the analysis of lung microstructure and quantify ventilation and perfusion of the lung. MRI scans can thus provide an anatomical and functional representation of pulmonary function, which could potentially be used in RT planning by avoiding areas of a well-functioning lung and assessing treatment response. However, both techniques are currently restricted to research settings with limited clinical use.

MRI shows potential for monitoring of early response during a course of treatment, information which is currently not utilized in RT. There is evidence in the literature to suggest a link between early changes as seen on functional MRI and progression-free survival.⁵⁶ This information can be used to adapt treatment to an individual patient's tumour response.

CONCLUSION

Using a combination of free breathing, breath-hold and gated scans with parallel imaging techniques, the quality of lung imaging has improved, with minimal artefacts from respiratory and cardiac motion. There are still challenges in adopting MRI for RT imaging but nevertheless based on the evidence available in the literature, a potential lung RT-imaging protocol can include for GTV delineation a T₁ and T₂ weighted gradient and spin-echo sequence either as breath-hold or respiratory gated. Motion assessment is feasible; however, the incorporation of the motion data to RT planning needs further investigation. DWI can prove to be an ideal non-invasive imaging technique to assess functional information and assist in GTV delineation. A DCE scan has the potential to provide additional information on the vascular nature of the tumour volume and healthy lung perfusion. However, imaging sequences need to be further assessed in the radiation oncology setting to evaluate and further develop RT-specific requirements.