Magnetic resonance imaging provides sensitive in vivo assessment of experimental ventilator-induced lung injury

Abstract

Animal models play a critical role in the study of acute respiratory distress syndrome (ARDS) and ventilator-induced lung injury (VILI). One limitation has been the lack of a suitable method for serial assessment of acute lung injury (ALI) in vivo. In this study, we demonstrate the sensitivity of magnetic resonance imaging (MRI) to assess ALI in real time in rat models of VILI. Sprague-Dawley rats were untreated or treated with intratracheal lipopolysaccharide or PBS. After 48 h, animals were mechanically ventilated for up to 15 h to induce VILI. Free induction decay (FID)-projection images were made hourly. Image data were collected continuously for 30 min and divided into 13 phases of the ventilatory cycle to make cinematic images. Interleaved measurements of respiratory mechanics were performed using a flexiVent ventilator. The degree of lung infiltration was quantified in serial images throughout the progression or resolution of VILI. MRI detected VILI significantly earlier (3.8 ± 1.6 h) than it was detected by altered lung mechanics (9.5 ± 3.9 h, P = 0.0156). Animals with VILI had a significant increase in the Index of Infiltration (P = 0.0027), and early regional lung infiltrates detected by MRI correlated with edema and inflammatory lung injury on histopathology. We were also able to visualize and quantify regression of VILI in real time upon institution of protective mechanical ventilation. Magnetic resonance lung imaging can be utilized to investigate mechanisms underlying the development and propagation of ALI, and to test the therapeutic effects of new treatments and ventilator strategies on the resolution of ALI.

acute respiratory distress syndrome (ARDS) is characterized by severe inflammatory lung injury leading to diffuse pulmonary infiltrates, severe hypoxemia, and acute respiratory failure (16, 35). Mechanical ventilation can exacerbate existing lung injury through several mechanisms including volutrauma, atelectrauma, and biotrauma (45) leading to ventilator-induced lung injury (VILI). This can lead to a systemic inflammatory response and the development of multiple organ dysfunction syndrome (10, 24, 38, 41, 46, 49, 50, 56) with unacceptably high morbidity and mortality. Early recognition of lung injury in critically ill patients is essential to implement lung protective ventilator strategies and target future therapeutics (20, 39). One of the barriers to developing ARDS therapies and preventing VILI has been the lack of an accurate method for early recognition of lung injury. Imaging may provide a method for early detection of lung injury and may be useful for following the progression or regression of VILI in animal models. Our study describes magnetic resonance imaging (MRI) as a sensitive method for early detection of lung injury in rat models of VILI. We present an MRI system for serial imaging that functions in tandem with a ventilator that makes interleaved measurements of lung mechanics.

MRI of lung parenchyma is technically challenging for several reasons, most notably because the signal is short lived (4, 9, 32), and one must use methods that also successfully image long-lived signals (28). Nonetheless, pulmonary MRI is a very active field for technique development. Pertinent to acute lung injury (ALI), hyperpolarized gas MRI has recently been used to evaluate alveolar recruitment and atelectasis-induced overdistension during mechanical ventilation (7, 8). Gases with fast spin-rotation relaxation at thermal equilibrium polarization have been used to image ventilation-perfusion ratios (1, 30) and time constants for fast and slow filling lung compartments (22). Magnetic resonance (MR) lung tissue imaging has also been used to image the elastic properties of lungs with VILI (37).

X-ray computed tomography (CT) is progressing as a technology in laboratory animal imaging (2, 21, 26, 27, 42, 43, 51). Recent developments in micro-CT include the use of cinematic images to analyze lung mechanics (27) and imaging the development of infiltrations next to experimentally induced atelectasis (43). However, presently, radiation doses for cinematic images are high for small laboratory animals and hourly, serial, cinematic imaging is not available.

Free induction decay (FID)-projection MR imaging overcomes the problem of imaging a rapidly decaying signal from hydrogen in the water and other molecules in lung parenchyma while also imaging long-lived signals, and is gaining popularity for both human and laboratory animal imaging (6, 12–14, 47, 48, 52, 55). Notably, Weiger et al. (55) achieved 81-s data acquisitions with 80,888 FIDs, which means a 9-min scan time if one signal is averaged twice and used 30% of the ventilator cycle for an inhaled volume and/or exhaled volume image. They further shortened the scan time by holding the lungs at one volume for the majority of the cycle so the other volumes were unavailable. Although the resolution was only 0.31 mm for mouse lungs, the parenchyma was well resolved at that level by using an appropriate bandwidth.

We have developed FID-based methods for serial imaging of anesthetized rats that develop VILI over many hours. These methods permit half-hour scan times and allow the lung tissue to be fully resolved at 0.4 mm in cinematic images (13 phases of the respiratory cycle) or 0.3 mm for imaging only the inhaled and exhaled volume. Three-dimensional images covering the entire lung (28) can be made hourly and interleaved with measurements of lung mechanics. When lung parenchyma is successfully imaged with MR, the lung pixels are brighter than the airway lumens. When the images are of sufficient resolution, the fissures are visible. In short, the images resemble X-ray CT of the lung and are similar to MRI of contiguous soft tissues, with which radiologists are familiar. When the parenchyma is resolved, one can develop quantitative scales of tissue density, measure deformations, and perform other image analyses similar to those used in CT (2, 21, 26, 27, 42, 43). Additionally, one is sure that minor pathology is not overlooked and that the anatomical information is present.

In this study, we used FID-projection imaging for serial in vivo assessment of experimental VILI interleaved with serial ventilator measurements of lung mechanics. We demonstrate that MRI has the sensitivity to detect lung injury significantly earlier than can be detected with altered lung mechanics and show that we can quantify the degree of lung infiltration throughout the progression and regression of experimental VILI. Real-time detection and longitudinal assessment of lung injury using MRI may aid in unraveling disease mechanisms and in optimizing lung protective ventilator strategies in rat models of ALI.

MATERIALS AND METHODS

Animal Models of VILI and LPS + VILI

All animal experiments were performed in compliance with the relevant laws and were approved by the Lovelace Respiratory Research Institute Animal Care and Use Committee. Nine adult male Sprague-Dawley rats (Charles River, Wilmington, MA) weighing approximately 300–400 g were studied. Three rats were untreated, three were intratracheally instilled with 0.75 ml of sterile PBS, and three were instilled with Pseudomonas lipopolysaccharide (LPS; Sigma, St. Louis, MO) at 1 or 2 mg/ml dissolved in 0.75 ml of PBS. For the instillation procedure, rats were anesthetized with 3–5% isoflurane in an induction chamber. Animals were imaged 2 days later to allow a full inflammatory response to LPS (3, 17). For imaging, animals were anesthetized with pentobarbital (50–70 mg/kg ip; Akorn, Lake Forest, IL), tracheotomized, and ventilated with a flexiVent ventilator (SCIREQ, Montreal, QC, Canada) with FiO₂ of 0.21 and respiratory rate of 60 breaths per minute. Heart rate and rectal temperature were monitored throughout the experiments. Additional doses (¼-½ of the original dose) of pentobarbital were administered every 1–2 h as needed to maintain a continuous plane of anesthesia and suppress respiratory effort. Rats were ventilated for up to 15 h with continuous monitoring.

Ventilator settings.

Settings for injurious ventilation were tidal volume 25–38 ml/kg, positive end-expiratory pressure (PEEP) 0 cmH₂O, 0.14 s inspiratory time, 0.31 s end inspiratory pause, and 0.55 s expiratory time. Settings for protective ventilation were tidal volume 11–19 ml/kg, PEEP 2 cmH₂O, 0.24 s inspiratory time, 0.31 s end inspiratory pause, and 0.45 s expiratory time, with semisinusoidal ramps.

The magnetic flux density at the ventilator was less than 18 mT, and 2 m of vinyl tubing [4.7 mm inner diameter (ID), 8 mm outer diameter] passing through a wave guide below cutoff in the Faraday cage reached the rat in the bore of the magnet. Although the calibration procedure nominally takes into account the compliance of the gas in the tubing, we found that the tidal volume measured from the images was 0.75 of the stroke volume and that the ventilator software's calculation of tidal volume was erratic. The absolute values of the mechanical parameters include some of the gas compliance, but the changes in serially measured parameters are unaffected.

The rats were supine. Images and measurements of respiratory mechanics were taken hourly. Two total lung capacity maneuvers (3 s at an airway pressure of 30 cmH₂O) were performed to standardize lung volume history prior to measurements of mechanics. Dynamic compliance, elastance, and resistance were measured using a forced oscillatory technique with a flexiVent ventilator (3, 17, 25). Quasistatic compliance was determined via generation of pressure-volume (P-V) curves. Lung injury detection by lung mechanics was defined as a sustained decrease in compliance of at least 33% compared with baseline.

Histology

Lungs were fixed at 20 cmH₂O pressure at room temperature in 10% neutral buffered formalin solution (Sigma) and embedded in paraffin. To section infiltrated regions, the terminal MRIs were reviewed and the locations of infiltration were determined relative to anatomic landmarks identifiable on the fixed lungs to choose the plane for trimming and sectioning. Lungs were stained with hematoxylin and eosin (H&E) and assessed for lung injury (36).

MRI System and Procedures

System.

An Oxford 1.89 T, 30-cm bore horizontal superconducting magnet was equipped with a Resonance Research 120-mm-ID shielded magnetic field gradient coil set, a 54-mm ID Morris Instruments proton-free birdcage RF coil, and a Tecmag Redstone console. FID-projection images of lungs were made using previously described methods (28), but with improvements to lower scan times to 30 min, improve the signal-to-noise ratio (SNR), and accommodate many large data sets.

The low-level mixer was bypassed by directly digitizing the nuclear magnetic resonance (NMR) radiofrequency (RF) signal with the 40 MHz digitizer. This allows the preamplified signal to be as strong as 3 V peak to peak to avoid noise from imperfect receiver grounding. ARR tuned preamplifiers were used to filter noise outside the 40 MHz bandwidth. The Redstone sends data directly to a Dell Precision T3600 computer with 3.7 GHz clock speed and 64 GB of RAM, which can handle a nearly continuous stream of complex data at 1 MHz and data sets over 10 GB. A separate Macintosh Pro computer with six dual core Xeon ES processors at 3.5 GHz, 64 GB of RAM, and 1 TB of solid-state storage was used to process the images.

Pulse sequence.

MRI of lung tissue is particularly challenging because water is diamagnetic, gas is not, and the many water-air interfaces make the magnetic field inhomogeneous (4, 9, 32). The frequency spectrum is inhomogeneously broadened to a line width of ∼8 parts per million, and the line shape is closer to Gaussian than Lorenzian. We used FID-projection imaging, Lauterbur's original pulse sequence (33), which is a method of making three-dimensional images of samples that have rapid signal decay. It falls under two primary acronyms presently, UTE (ultra-short echo time) (4, 5, 19, 40) and ZTE (zero-time echo) (23, 28, 29, 33, 44, 54). The timing difference between the two is that with UTE, the RF spin excitation occurs before the imaging gradient is ramped, and one collects complete data. With ZTE, the RF spin excitation occurs during the gradient, and one misses data during the equipment dead time, which must be compensated for (31). We used ZTE as previously described (28).

Imaging parameters.

The repetition time is 1.2 ms, and data are collected for 0.7 ms and 5 μs after a 4-μs RF pulse. Some data are discarded at the end. There are 120,960 gradient directions and 60,480 positive-negative gradient pairs, which provide sufficiently small solid angles to grid data for 256³ pixel images. The data grid is 512³, but we confine the field of view (FOV) to the 256³ inner pixels. This avoids the shading common in projection images. For cinematic images the gradient strength is 44 mT/m; the resolution is 0.4 mm; the FOV is an 88-mm sphere; the data acquisition time, including 2 FIDs for opposite gradient directions and the missing early data is 1.335 ms; the image grid is 220³ pixels; and the scan time is 32 min. For higher-resolution images of just the inhaled and exhaled volume, the corresponding figures are 56 mT/m, 0.3 mm, 88 mm, 1.399 ms, 300³, and 25 min, respectively, with threefold signal averaging. The cinematic images typically contain two or three images at inhaled volume and at exhaled volume, and we add these to improve the SNR.

Imaging Processing

Most image processing was programmed in IDL, which used C subprograms for computer-intensive, threaded calculations. The Chest Imaging Platform (www.chestimagingplatform.org) and 3D Slicer (15) were used to denoise, segment, and visualize the lungs. Tidal volumes are measured by segmenting the lungs and airways from images of undamaged lungs at inhaled volume and exhaled volume and subtracting the two volumes.

Data Analysis and Statistics

Data from the images are reported with 95% confidence intervals (CI) calculated according to the construction of the statistic. Unless otherwise specified, we analyzed images at exhaled volume. This is the most reproducible degree of inflation and essentially represents the functional residual capacity. The lungs are segmented from the image and comprise ∼100,000 pixels (n_pix,lung). Noise in the image is a function of location, so we calculate the noise in the lung region by subtracting two consecutive images of the segmented lung. Unlike the segmentation used to compute tidal volumes, this segmentation does not include the airways or major pulmonary vasculature, but does include minor airways and vasculature. For one image, the root-mean-square noise (RMSN) is the root mean square of pixels in the difference image divided by $\sqrt{2}$. The noise in our images is normally distributed because we phase the image and use the real part (they are not magnitude images). The RMSN for the average of n_img images, or standard deviation (SD) for lung pixels (s_pix,lung), is the RMSN for one image divided by $\sqrt{n_{img}}$. The pixel value for air $(\overline{p{\nu }_{air}})$ is calculated as the average of pixel values in the large airway lumens (∼1,000 pixels) and has sample variance (s_air²), the variance of the average $(S_{air}^2=\text{Var}(\overline{p{\nu }_{air}})),$ not the variance for individual pixels. The pixel value for contiguous tissue $(\overline{p{\nu }_{contig}})$ is the average of ∼2,000 muscle pixels and has sample variance of s_contig². Muscle has similar NMR relaxation as that of lung tissue and infiltrates. The scaled pixel values for lung are $p{\nu }_{lung,scaled}=r_{scale}(p{\nu }_{lung}-\overline{p{\nu }_{air}})$, where $r_{scale}=1/(\overline{p{\nu }_{contig}}-\overline{p{\nu }_{air}}).$ They range from 0 (also 0 g/ml) to 1 (also 1 g/ml) with individual SD s_{pix,lung,scaled} = r_scale s_pix,lung.

Lung mass, calculated from an image (m_lung) is

$$V_{voxel}\underset{}{\overset{}{\mathrm{\sum }_{lung}^{}{pv}_{lung,scaled}}},$$

where V_voxel is the voxel volume, and the sum has variance n_pix,lung s_{pix,lung,scaled}² by virtue of image noise. The variance of the estimates of location parameters must be combined to achieve the sample SD used to compute the CI:

$$s_{m_{lung}}=V_{voxel}\sqrt{n_{pix,lung}[s_{pix,lung,scaled}^2+r_{scale}^2(s_{air}^2+s_{contig}^2)],}$$

and the 95% CI of lung mass is m_lung ± 1.96 s_{lung_mass}. Lung density is

$${\rho }_{lung}\underset{}{\overset{}{=\mathrm{\sum }_{lung}^{}{pv}_{lung,scaled}/n_{pix,lung},}}$$

and its SD is

$$s_{\rho lung}=\sqrt{s_{pix,lung,scaled}^2+r_{scale}^2(s_{air}^2+s_{contig}^2)}/\sqrt{n_{pix,lung},}$$

so the 95% CI is ρ_lung ± 1.96 s_ρlung.

The Index of Infiltration is defined as I_infil = (f − f_{exp_light})/f_{exp_dark}, where f is the fraction of light pixels in the lung, f_{exp_light} is the fraction expected light (0.25), f_{exp_dark} is the fraction expected dark (0.75), and the boundary between light and dark for the scaled lung pixels is 0.5 (these values were chosen empirically to yield I_infil ≈ 0 for healthy lungs). Alternatively, it is the sum of ones and zeros assigned to the pixels divided by the number of pixels (n_pix,lung) and thus has a binomial distribution with sample variance f(1 − f)/n_pix,lung. The combined sample SD is

$$s_{infil}=\sqrt{[f(1-f)+s_{pix,lung,scaled}^2]n_{pix,lung}+r_{scale}^2(s_{air}^2+s_{contig}^2),}$$

and the 95% CI is I_infil ± 1.96 s_infil.

Lung volume is V_voxeln_{lung_pix}. Its measurement error comes from the success of the segmentation. We may include or exclude ∼500 pixels out of 100,000 or ±0.005 of the volume. Lung density and infiltration index should be minimally affected by the success of segmentation because the pixels included or excluded should have similar pixel values.

Indices of Infiltration and lung density data were not normally distributed and were therefore log₁₀-transformed to normalize the data; this allowed us to then perform standard parametric tests on the transformed data. After log₁₀ transformation, data were analyzed by one-way ANOVA followed by a Tukey's posttest for pairwise comparisons. Time to detection of VILI by lung mechanics and by MRI were compared using a Kaplan-Meier curve and significance was determined by the log-rank test. Statistical analysis was performed with Prism (GraphPad Software, La Jolla, CA). All P values were two-tailed, and statistical significance was accepted at P < 0.05.

RESULTS

MRI for in vivo assessment of rat ALI.

Figure 1 demonstrates an example of one of the 0.3-mm resolution images at inhaled volume for an uninjured rat. The fissures are visible, the lungs are brighter than the airway lumens, and lung tissues are fully resolved. These features verify that the imaging system is functioning optimally and is properly triggered by the ventilator. Figure 2 and Supplemental Videos S1 and S2 show examples of cinematic MR images of rat VILI. They are resolved at 0.4 mm but contain images from the entire respiratory cycle. The lungs are darker at inhaled volume than exhaled volume, and one can see whether infiltrated regions change brightness to indicate they are ventilated and to visualize the regional mechanics of ventilation. Figure 3 and Supplemental Video S3 illustrate an example of measuring the lung volumes throughout the respiratory cycle and determining the delivered tidal volume from the computed lung volumes at inhalation and exhalation.

Fig. 1.

High-resolution magnetic resonance (MR) images of a rat lung. Coronal (A), sagittal (B), and axial (C) slices through a three-dimensional (3D) image of a healthy rat with 0.3-mm resolution at end-inhalation. The arrows point to the fissures. The tick marks on the side indicate the location of the other two views. The gray scale of lighter tissues is compressed to see the lungs better. The airways are darker than the lung parenchyma.

Fig. 2.

Cinematic MR images of rat ventilator-induced lung injury (VILI). Still images are shown from Supplemental Video S1 (A) and Supplemental Video S2 (B). Each breath of the videos is 13 images made from a 30-min scan, and the breaths represent states of the lung approximately 1 h apart. In Video S1, the rat (animal 41006) was untreated and the ventilation was injurious for 8 h. Infiltrations (arrows) first appear after 3 h in the lateral aspects of the middle left lung and the apical/middle lobe boundary of the right lung and increase for the duration. In Video S2, the rat (animal 41010) was treated with saline and the ventilation was injurious for 3.9 h, and then protective for 6.9 h. A focal infiltration (arrow) developed in the right lung with injurious ventilation and regressed during protective ventilation.

Fig. 3.

MR image quantification of lung volumes across the respiratory cycle. A: lungs were segmented using the 3D Slicer software to measure lung volumes during the respiratory cycle. Four still images are shown from Supplemental Video S3, which shows a three-dimensional rendering of the lung volume for one cycle with a superimposed maximum volume outline and paging through various slices. B: pixels were counted for inhaled and exhaled volumes to calculate the tidal volume.

Untreated animals had variable responses to high tidal volume ventilation. One animal developed mild VILI detectable by MRI (Fig. 4, A–C) but had minimal histologic damage (Fig. 4D) and unchanged lung mechanics (Fig. 5, A and B). However, two untreated animals developed severe VILI characterized by mild alveolar inflammation and early hyaline membrane formation (Fig. 4H), which was detected by MRI early (Fig. 4, E–G), prior to late changes in lung mechanics (Fig. 5, C and D).

Fig. 4.

MR images and histological images of rat acute lung injury (ALI). Coronal, sagittal, and axial views of MR images of a minimally damaged untreated rat (A–C), an untreated rat with VILI (E–G), a PBS-treated rat with VILI (I–K), and a lipopolysacharride (LPS)-treated rat with VILI (M–O). Hematoxylin and eosin (H&E) staining of lungs from an untreated rat with minimal injury (D), an untreated rat with VILI (H), a PBS-treated rat with VILI (L), and an LPS-treated rat with VILI (P).

Fig. 5.

Lung mechanics in rat ALI models. Dynamic compliance, elastance, and resistance measured hourly in an untreated rat with minimal injury (A), an untreated rat with VILI (C), a PBS-treated rat with VILI (E), and an LPS-treated rat with VILI (G) using a flexiVent ventilator (B, D, F, and H). Graphs show pressure-volume (P-V) curves for animals at the beginning and termination of ventilation.

In general, PBS-treated animals (Fig. 4, I–K) were more susceptible to VILI than untreated animals. PBS-treated rats that developed extensive infiltrates were found to have widespread lung injury upon histopathological assessment, with areas of infiltrate on MRI characterized by mild multifocal edema and early hyaline membrane formation (Fig. 4L). Over time, rats with early radiologic detection of VILI developed significant alterations in lung mechanics with decreases in compliance, increases in elastance and resistance, and a rightward shift in the P-V curve (Fig. 5, E and F).

Rats treated with LPS had extensive infiltrations from the start of imaging (Fig. 4, M–O). Histopathology revealed that LPS/VILI rats had diffuse injury with marked alveolar neutrophilic and histiocytic infiltration, and scattered hyaline membrane deposition (Fig. 4P) in areas that correlated with infiltration on MRI. Lung mechanics showed substantially decreased compliance, increased elastance and resistance, and a profound rightward shift in the P-V curve that worsened with time (Fig. 5, G and H). LPS-treated animals either died during the last image acquisition or were euthanized because they developed bradycardia, indicating they were moribund.

For each animal, we were able to quantify the degree of lung injury throughout and at the termination of each experiment (Table 1). Untreated and PBS-treated animals developed a similar degree of injury with a significant increase in the Index of Infiltration compared with the initial images of untreated rats (P = 0.0027 by one-way ANOVA; P < 0.01 for untreated VILI vs. untreated controls, P < 0.05 for PBS-treated VILI vs. untreated controls) (Fig. 6). LPS-treated animals developed the most severe injury with a significantly higher Index of Infiltration compared with the initial images of untreated rats (P < 0.01). LPS-treated animals also had increased lung density compared with other VILI animals, however, it did not reach statistical significance. There were different patterns of injury in untreated and PBS-treated animals with VILI compared with LPS/VILI animals. Untreated and PBS-treated rats developed injury that originated in lateral regions of the lung ventral to the trachea and main bronchi. In contrast, LPS-treated animals developed infiltrates that surrounded the major airways and occupied the dorsal regions of the lung.

Table 1.

Quantification of lung injury by MRI

Animal No.	Treatment	Weight, g	Lung Mass, g	Lung Density, g/ml	Index of Infiltration
40903*	Untreated	436	3.726 ± 0.003	0.4444 ± 0.0004	0.0241 ± 0.007
40929*	Untreated	293	2.732 ± 0.003	0.4205 ± 0.0004	0.0036 ± 0.010
41006*	Untreated	310	2.953 ± 0.004	0.4024 ± 0.0005	0.0020 ± 0.007
40903	VILI	436	3.750 ± 0.004	0.4886 ± 0.0005	0.2116 ± 0.008
40929	VILI	293	3.346 ± 0.002	0.4960 ± 0.0003	0.2870 ± 0.006
41006	VILI	310	4.627 ± 0.004	0.6073 ± 0.0006	0.5380 ± 0.008
41002	PBS/VILI	314	4.055 ± 0.003	0.6470 ± 0.0004	0.678 ± 0.007
41008	PBS/VILI	320	3.404 ± 0.005	0.4087 ± 0.0006	0.038 ± 0.008
41010	PBS/VILI	398	3.793 ± 0.003	0.4863 ± 0.0003	0.250 ± 0.008
40904	LPS/VILI	382	4.690 ± 0.003	0.6264 ± 0.0004	0.532 ± 0.006
40910	LPS/VILI	412	4.942 ± 0.003	0.8901 ± 0.0005	0.886 ± 0.009
50715	LPS/VILI	377	4.660 ± 0.003	0.5656 ± 0.0004	0.300 ± 0.004

^*Results were obtained from the initial images of untreated rats prior to the induction of VILI. All other quantitative results were from segmenting the terminal images from each rat. Values represent ± 95% CI. LPS, lipopolysaccharide; MRI, magnetic resonance imaging; PBS, phosphate-buffered saline; VILI, ventilator-induced lung injury.

Fig. 6.

Increased Index of Infiltration in rat ALI. Coronal slices from untreated rats at baseline prior to VILI (A–C), coronal slices from the terminal images of untreated rats with VILI (D–F), PBS-treated rats with VILI (G–I), and LPS-treated rats with VILI (J–L). Images correspond to animals listed in Table 1. Quantification of the Index of Infiltration (M) and lung density (N). P = 0.0027 by one-way ANOVA; *P < 0.05, **P < 0.01 by Tukey's posttest.

MRI detects lung injury before changes in lung mechanics.

Serial imaging allowed us to track the development of infiltrates in untreated and PBS-treated animals with VILI and visualize their progression while lung mechanics remained essentially unaltered until substantial injury was visualized via MRI. In one rat with VILI (Fig. 7), we visualized a region of infiltration after 6.5 h (Fig. 7A, left) and another infiltrate after 11.2 h (Fig. 7A, middle) of injurious high tidal volume ventilation. Both regions progressed on imaging over time, but there was no detectable change in lung mechanics until 12.9 h (Fig. 7, B and C). After this time, there was a dramatic deterioration in lung mechanics (14 h), by which point the infiltrates had become extensive (Figs. 7A, right). Animals with VILI that had abnormal lung mechanics prior to necropsy had evidence of severe lung injury with diffuse damage and hyaline membrane formation on histopathology (Fig. 7D). In addition to the visual progression on imaging, we were also able to quantify the degree of lung injury within each infiltrated region over time (Fig. 8). The appearance of infiltrates and their progression in MRIs preceded alterations in lung mechanics in all of the animals that developed VILI except for the LPS-treated rats, which had altered lung mechanics and extensive infiltrations from the beginning (Table 2). The mean time to detection of VILI by MRI (3.8 ± 1.6 h) was significantly earlier than that detected by altered lung mechanics (9.5 ± 3.9 h) (Fig. 9) (P = 0.0156 by the log-rank test).

Fig. 7.

MRI detection of early VILI precedes measurable changes in lung mechanics. A: three coronal MR image slices from three different times in a representative rat with VILI (animal 40929). Infiltrations (arrows) first appear early (6.5 h), then late (11.2 h), and at the end of the experiment (14.5 h). B: compliance, elastance, and resistance remain essentially unchanged until hour 14. C: P-V curves show a rightward shift in the P-V curve at 15 h. D: H&E staining of lungs harvested at 15 h shows diffuse damage with early hyaline membrane formation.

Fig. 8.

Index of Infiltration for infiltrates that develop during VILI. The Index of Infiltration was quantified over time for regional infiltrates that developed in a representative rat with VILI (animal 40929). The Index of Infiltration was calculated in spherical regions of interest shown in single coronal slices as circles (A) with nearby spheres of muscle defining the pixel value for contiguous tissue. B: the Index of Infiltration is plotted for a region that developed infiltration early (6.5 h) and one region that developed an infiltrate later (11.2 h). The index is the fraction of air space that has been filled in and ranges from 0 to 1. The bar labeled SD shows the height of one standard deviation for all points on the graph.

Table 2.

Detection of VILI by MRI and lung mechanics

Animal No.	Treatment	MRI Detection of VILI, h	Lung Mechanics Detection of VILI, h
40903	VILI	3.6	11*
40929	VILI	6.5	12.9
41006	VILI	3.9	5.9
41002	PBS/VILI	2.8	3.5
41008	PBS/VILI	4.6	12.7*
41010	PBS/VILI	1.8	10.7*
40904	LPS/VILI	0	0
40910	LPS/VILI	0	0
50715	LPS/VILI	0	0

^*Did not meet criteria for detection of VILI by lung mechanics (defined as a sustained decrease in compliance of at least 33%) at the termination of the experiment.

Fig. 9.

MRI detection of early VILI. Time to detection of VILI by MR imaging vs. lung mechanics (P = 0.0156). Detection of VILI by lung mechanics was defined as a decrease in compliance of at least 33%. Statistical significance was determined by the log-rank test.

MRI detects development and regression of early VILI.

In one animal with VILI, we were able to detect and reverse early lung injury by instituting protective low tidal volume ventilation after observing a single, limited infiltration before there was any change in lung mechanics (Fig. 10). In this animal, we identified a small region of infiltration in the right middle lobe as early as 1.8 h after injurious ventilation. We were able to follow this region of interest on serial imaging and visualize progression of the infiltrate over 3.6 h of injurious ventilation (Fig. 10, A–D), during which time there were no measurable changes in lung mechanics (Fig. 10E). Because we were able to detect this area of early injury by imaging, we were able to determine whether it would regress with protective ventilation. We know of no other method that could detect an injury this minor in vivo. With institution of low tidal volume protective ventilation, we were able to observe regression of this infiltrate over 7 h (Fig. 10, A–D). This particular focal area had mild perivascular and alveolar inflammatory infiltrates, and hyaline membranes along alveolar ducts, while the surrounding lung appeared normal in H&E-stained slides (Fig. 11). We also quantified the degree of injury in this region over time, which demonstrated an increase in the Index of Infiltration with injurious ventilation and a decrease with protective ventilation (Fig. 12, A and C). An unexpected finding was that two apparently unaffected control regions showed a small increase in the Index of Infiltration with a similar trajectory as the affected region, suggesting that early injury was emerging, but quickly resolved with protective ventilation (Fig. 12, B and C).

Fig. 10.

MR imaging detects development and regression of early VILI. Coronal and sagittal images at end-inhalation (A and B) and end-exhalation (C and D) demonstrate development, progression, and regression of a right middle lobe infiltrate without changes in lung mechanics (E) in a representative rat with VILI (animal 41010). The image showing the greatest extent of infiltration was taken during protective ventilation, shortly after the maximal damage from the injurious ventilation. Arrows indicate the infiltrate.

Fig. 11.

Histopathology confirms detection of small region of ALI by MRI. A: an oblique section through the terminal image of a representative rat with VILI (animal 41010) whose orientation is shown in B. The outline in A corresponds to the H&E-stained section (C) shown at ×8 magnification. The outlined area in C corresponds to the infiltration in A. The rectangular inset in C corresponds to the ×200 magnified image (D), which shows mild to moderate edema, mild perivascular and alveolar inflammation, and early hyaline membranes along alveolar ducts.

Fig. 12.

Index of Infiltration for VILI that regresses with protective ventilation. The Index of Infiltration is graphed for an early infiltrate that developed in a representative rat (animal 41010) with injurious ventilation (A) and two control locations (B). Symbols on the graph (C) are paired with images (A and B) indicating regions of lung quantified and nearby regions of muscle. The index is highest for the infiltrate in an image taken during protective ventilation, after maximum damage. The control regions show an unexpected rise in the index, indicating there is a global problem that escapes simple visual detection and that resolves quickly with protective ventilation.

DISCUSSION

Early recognition of lung injury is critical for implementing lung protective ventilator strategies (20, 39) and may provide a therapeutic window through which to target future therapies. Although studies are ongoing to identify improved clinical predictors and develop biomarkers for the development of ARDS in at-risk patients (18, 34, 53), current diagnostics have limited sensitivity and specificity early in the disease. Recent studies suggest that positron emission tomography imaging may detect early lung injury and aid mechanistic studies in animal models of ALI (11).

In this study, we investigated the ability of MRI to detect early lung injury and assess progression of VILI in rat models of ALI. We demonstrate that MRI is a sensitive method for detecting small regions of lung injury before alterations in lung mechanics. We also show that regions of infiltration can be quantified, and trajectories of individual regions can be followed over time in vivo with adjustment of the ventilator to either induce or reverse lung injury. These regions were confirmed as demonstrating the presence of lung injury histologically and, in the future, they might be studied at the molecular level in preclinical models to help elucidate biological pathways underlying the progression and/or resolution of lung injury.

MRI of lung tissue is canonically difficult and technically demanding, but we have demonstrated techniques that can generate images with relatively inexpensive equipment. To our knowledge, we are not aware of other systems that can provide hourly, cinematic images of the entire lung that are capable of detecting and quantifying small infiltrations. X-ray CT is the current standard for lung imaging and produces high-quality laboratory animal images (2, 27). However, at present, the radiation dose is often high, limiting serial measurements and raising concern for both pulmonary and extrapulmonary radiation exposure. Comparable cinematic images of rat lungs (27) require 3 Gy for one respiratory cycle. MRI also has the advantage of a variety of contrast mechanisms that may allow us to distinguish cellular inflammation, atelectasis, edema, etc. At present, however, we describe the development of tissue-density images, allowing comparable resolution to CT without the radiation exposure.

Although postmortem histological and biological assays are the current gold standards for assessing lung injury in animal studies, they do not allow for in vivo intervention and serial assessment over time within the same animal. Lung mechanics can be used to longitudinally assess a response to therapeutics within the same animal, but they are limited by poor sensitivity. In our study, alterations in lung mechanics emerged only after there was extensive injury on imaging. With MRI we show that it is now possible to detect lung injury early in its development and track its time course and severity. With the ability to accurately image the lungs and quantify regions of injury, each animal has a documented history, and we can choose at what stage to euthanize the animal for histologic and molecular assays. This provides many advantages over current postmortem assays and opens new opportunities to test protective ventilator strategies and pharmacologic interventions in real-time for ALI. Taken together, our study shows that MRI is a sensitive method for in vivo assessment of lung injury in animal models of ALI. Although our study addresses MR lung imaging only in rodent models of ALI, it raises the possibility that ongoing advancements in MR imaging and technology might one day make it feasible to use MRI for monitoring lung findings in human ARDS.

GRANTS

This work was supported by the Lovelace Respiratory Research Institute (LRRI) as part of the Brigham and Women's Hospital/LRRI Lung Research Consortium; and by National Heart, Lung, and Blood Institute Grants R01 HL-116931 to R.S.J. Estépar, R01 HL-091957 to R.M. Baron, and R01 HL-114839 to L.E. Fredenburgh.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

D.O.K., G.R.W., R.M.B., and L.E.F. conception and design of research; D.O.K., P.T.F., and J.M.H. performed experiments; D.O.K., A.P.G., R.S.J.E., and L.E.F. analyzed data; D.O.K., G.R.W., and L.E.F. interpreted results of experiments; D.O.K. and L.E.F. prepared figures; D.O.K. and L.E.F. drafted manuscript; D.O.K., G.R.W., R.M.B., and L.E.F. edited and revised manuscript; D.O.K., P.T.F., J.M.H., A.P.G., R.S.J.E., G.R.W., R.M.B., and L.E.F. approved final version of manuscript.