3D ³He diffusion MRI as a local in vivo morphometric tool to evaluate emphysematous rat lungs

Abstract

In this work, we investigate ³He magnetic resonance imaging as a noninvasive morphometric tool to assess emphysematous disease state on a local level. Emphysema was induced intratracheally in rats with 25 U/100 g body wt of porcine pancreatic elastase dissolved in 200 μl saline. Rats were then paired with saline-dosed controls. Nine three-dimensional (3D) ³He diffusion-weighted images were acquired at 1, 2, or 3 wk postdose, after which the lungs were harvested and prepared for histological analysis. Recently introduced indexes sensitive to the heterogeneity of the air space size distribution were calculated. These indexes, D₁ and D₂, were derived from the moments of the mean equivalent airway diameters. Averaged over the entire lung, it is shown that the average ³He diffusivity (D_ave) correlates well with histology (R = 0.85, P < 0.0001). By matching small (0.046 cm²) regions in ³He images with corresponding regions in histological slices, D_ave correlates significantly with both D₁ and D₂ (R = 0.88 and R = 0.90, respectively, with P < 0.0001). It is concluded that ³He MRI is a viable noninvasive morphometric tool for localized in vivo emphysema assessment.

emphysema is a chronic obstructive pulmonary disease (COPD) typified by tissue destruction and airway expansion (40). Emphysema, particularly the early stages, may be characterized by a heterogeneous distribution of air space sizes (13). Indeed, McLaughlin and Tueller (23) showed that in apparently normal lungs of smokers harvested at autopsy (none of the deaths were attributed to COPD), numerous localized regions of tissue destruction were surrounded by normal tissue. Because early detection of COPD can play a significant role in managing the disease and improving patient outcomes (2, 20), a locally specific and sensitive method of detection may prove beneficial. Hyperpolarized (HP) ³He and ¹²⁹Xe gas have been exploited in clinical and preclinical magnetic resonance imaging (MRI) studies to evaluate lung tissue structure (5, 6, 8, 18, 22, 33), although perfluorinated gases have also been used (14). In several studies, ³He MRI has been demonstrated to be sensitive to mild or early emphysema (10, 28, 34, 38). Gas-phase MRI is well suited for detecting emphysema because the diffusivity (Brownian motion) of gas is strongly affected by restrictive barriers, such as alveolar walls. Additionally, the gas contained in larger air spaces contributes proportionally more to the MRI signal in each voxel, making the signal within each voxel a volume-weighted average sensitive to the presence of enlarged air spaces or blebs. Therefore, MRI provides noninvasive, localized information about lung structure while allowing for frequent assessment of disease state without repeated exposure to ionizing radiation.

Despite its potential utility, validation of ³He diffusion MRI with histology at the local, or near-pixel, level is still needed before the technique can be generally accepted as a reliable localized measure of tissue destruction. Only a few studies demonstrating a correlation between histology and MRI exist (22, 28, 41), and they generally exhibit a correlation between MRI and regions of the lung of a much larger scale than the image resolution. Woods et al. (41) studied cylindrical cork-bore samples (1.3-cm diameter × 2-cm thick) from explanted emphysematous human lungs. These lungs were either healthy but rejected for transplant, or severely emphysematous and removed at transplant. Their results found a fair correlation between morphometry and ³He diffusivity (R = 0.611) using the lung samples and corresponding regions from the diffusion images, although their whole lung results correlated much more strongly (R = 0.964). Unfortunately, no mildly emphysematous lungs were available for their study. Peces-Barba et al. (28) studied mild elastase-induced emphysema in rats using a dose three times larger than used herein: 75 U/100 g body wt vs. 25 U/100 g body wt. Forty-five days after the dose, ³He diffusion MRI was performed and was compared with morphometry on a lobar level (with R = 0.71). Mata et al. (22) studied mild elastase-induced emphysema in rabbits and found that ³He diffusivity correlated well with regional morphometry that was performed by dividing lungs into four parts (R = 0.78).

Previous work has established that ³He MRI is sensitive to tissue destruction and that measured gas diffusion correlates well with regional histology. To correlate MRI results with tissue characteristics, previous studies used various methods of histological analysis. For example, Mata et al. (22) measured an average chord length within the air spaces of rabbit lungs, Woods et al. (41) measured the mean linear intercept (L_m) in excised human lungs using small cork-bore samples of lung sections, and Peces-Barba et al. (28) measured the number average of alveolar air space area and L_m in rats. Although L_m measurements have become generally accepted as the standard stereological measure of air space size, it is actually a measure of volume-to-surface area ratio. Using point-counting methods, L_m is essentially the ratio of the number of test points within air spaces to the number of grid intercepts with air space walls (39). It is therefore expected to be rather insensitive to a few enlarged air spaces in the presence of many small ones. Indeed, it has been shown that L_m is not a sensitive method for identifying early or mild emphysema in humans (36) nor in a rat elastase model, even after 5 wk (9). Furthermore, L_m is not intended to provide localized information in a heterogeneous structure since its rigorous calculation requires exhaustive, unbiased sampling of the entire lung or large lung regions. Consequently, a weighted index sensitive to distributions of air space size is argued to be more appropriate for this study, in which the aim is to compare MRI and local histology.

As part of previous efforts to characterize local airway structure from histological sections, Parameswaran et al. (27) demonstrated that variability in air space sizes may not be well reflected by L_m measurements. The alternatives they proposed were weighted indexes of air space size distributions D₁ and D₂. Mathematically, these describe the first (D₁) and second (D₂) moments of the distribution of airway sizes determined from histological samples. To implement this approach, airway sizes are first quantified in terms of an equivalent diameter (D_eq = ) that is calculated from measurements of individual air space areas (A). The mean μ, the variance σ², and the skewness γ of D_eq are then determined to calculate D₁ and D₂:

(1a)

(2b)

It is seen that a normal distribution, but with a large variance, will weight D₁ to higher values. A nonnormal, skewed distribution of air space size will further weight D₂. Hence, the presence of a few large air spaces causes the indexes to increase rapidly, unlike with L_m. Nevertheless, it must be emphasized that D₁ and D₂ provide information about the air space size distribution, not actual air space sizes. Therefore they may be used to describe a disease state but cannot strictly be used to quantify air space geometry parameters.

In this paper we examine an emphysematous disease state in an animal model on a local level. This is done by applying a model of air space geometry introduced by Yablonskiy et al. (42) to diffusion-weighted, three-dimensional (3D) ³He MRI (15). The results are then compared with weighted indexes of the alveolar size distribution from histological samples. In brief, the Yablonskiy model depicts the lung parenchyma as a network of cylindrical bronchioles sleeved by alveoli. Diffusion longitudinal to a bronchiole, D_L, and diffusion transverse to a bronchiole, D_T, are separated out as key parameters that can be used to describe lung microstructure. The Yablonskiy model then allows for direct measurement of the diffusion anisotropy D_an = D_L − D_T, and the voxel-average diffusivity D_ave = (D_L + 2D_T)/3 through acquisition of multiple diffusion-weighted images. We note that D_ave is equivalent to what is commonly referred to as the apparent diffusion coefficient (ADC) (42).

METHODS

Animal dosing.

Humane animal handling procedures approved by the Institutional Animal Care and Use Committee at Pacific Northwest National Laboratory were followed. A total of 18 male Sprague-Dawley rats (Charles River Laboratories, Wilmington, MA) were used in this study; half were dosed intratracheally with porcine pancreatic elastase (Calbiochem, La Jolla, CA), and the other half with sterile saline. The average body weight of the rats at the time of dosing was 209 ± 9 g. About 10–15 min before dosing, the rats were administered 0.02 ml/kg body wt glycopyrrolate subcutaneously to inhibit oral secretions and facilitate insertion of an endotracheal tube. The animals were then anesthetized with 2.5–3.5% isoflurane in oxygen and placed supine on a tilting intubation table. Approximately 0.02 ml of lidocaine was applied directly to the larynx; then a 14-gauge catheter tube was orally inserted into the trachea. The tube was precut to a length to ensure that it would pass through the larynx but not come near the carina. A high-pressure aerosolizing syringe (PennCentury, Philadelphia, PA) was sterilized with ethanol, then thoroughly rinsed with sterile saline before dosing. The elastase solution (or saline) was drawn into the syringe, the end of the aerosolizer was inserted through the length of the trachea tube, and the dose was administered. Rats were dosed intratracheally with either 25 U/100 g body wt of elastase in 200 μl of sterile physiological saline or with 200 μl saline only as a control, with nine rats in each group. After the dose was administered, rats were then tilted upward for several breaths after which the trachea tube was removed. When a rat began to revive from the anesthetic, it was returned to its cage and provided food and water ad libitum. The dosing was conducted over three consecutive days with six weight-matched animals dosed per day (3 each of elastase and saline). At each 1-, 2-, and 3-wk time point, elastase-dosed animals were imaged with a control. The different time delays were intended to generate a range of responses to the elastase dose. Animal mortality was zero.

HP ³He generation and delivery.

HP ³He was generated via hybrid spin-exchange optical pumping (1, 37) using a home-built, spectrally-narrowed (3), dual-laser apparatus. Two ≈110-ml Pyrex ³He vessels, each filled to ≈7 atm (for a total of ≈800 ml of HP gas per vessel), were used to generate a ³He polarization of ≈50% overnight (16). For imaging, a bolus of ³He gas was released into a 2-liter Tedlar bag (Jensen Inert Products, Coral Springs, FL) that resided in the fringe field of the imaging magnet. The HP ³He was mixed with ≈0.4 liter of N₂ to ensure an adequate volume of gas for an entire imaging experiment. The gas mixture was then delivered to the rat via a positive-pressure mechanical ventilator, as described below.

Imaging.

All of the animal preparation procedures, specific ventilation parameters, and details of imaging parameters used are described in detail elsewhere (15) and will be explained only briefly here. Animals were prepared for imaging by anesthetizing with isoflurane, following which an endotracheal tube was inserted, as was done for the dosing. With the tube inserted, the anesthetized rat was secured supine to a wooden tray and attached to a home-built, computer-controlled mechanical ventilator based on those described by others (8, 12). A rectal temperature probe was then inserted and ECG leads attached to the forepaws for physiological monitoring. A breathing rate of 50 breaths/min was used to maintain the animals under gas anesthesia while allowing sufficient time for breathing maneuvers and imaging. Each breathing cycle consisted of ≈1 ml of the HP ³He-N₂ gas mixture followed by ≈4 ml of air to a maximum pressure of ≈15 cmH₂O (an oxygen-enhanced mixture of 30% O₂-70% N₂ was used in place of normal breathing air). Following a passive exhalation period, imaging data were acquired during a breath hold. Based on ³He ventilation imaging studies by others (4) and on 3D ³He diffusion imaging on healthy rats in our laboratory using this same gas delivery technique (15), we expect that this stacked method of ventilation results in a homogeneous distribution of ³He gas in the lungs. The air pressure at the trachea after exhale, or end-expiratory pressure (EEP), was measured by a solid-state pressure transducer and was recorded for each rat. Although EEP was not rigorously controlled, its value was documented so that diffusion data could be corrected for rat-to-rat differences in lung inflation at the same point in each rat's breathing cycle.

The wooden tray with the animal and ECG/temperature-monitoring unit (SA Instruments, Stony Brook, NY, model 1025) slid into the MRI coil on a mating rail to enable animal positioning. Warm air, typically ≈50°C, was circulated within the magnet bore to maintain body temperature. The rectal temperature was maintained at 35 ± 2°C, and the heart rate was generally 250–350 beats/min, depending on body weight and level of anesthesia.

All imaging was performed in a 2.0-T horizontal-bore magnet (Oxford, UK) with 150 mT/m gradients (Resonance Research, Billerica, MA) and a Varian UnityPlus spectrometer (Palo Alto, CA). A home-built, birdcage-style MRI coil tuned to both ³He and ¹H frequencies (64.4 MHz and 84.9 MHz, respectively) was used. A trigger signal from the ventilator ensured that image data were acquired at the same point of each breathing cycle; cardiac motion was ignored. A standard 2D multislice spin-echo sequence was used for proton imaging for animal positioning. A hybrid 3D imaging sequence combining radial acquisition (in the xy-plane) with conventional phase encoding (along the z-axis, parallel to the magnetic field) was used for 3D ³He diffusion imaging. This pulse sequence is referred to as ZIPR in Ref. 29 and is described in detail there and in Ref. 15. Diffusion encoding was applied along the phase-encode axis with a gradient pulse duration of 600 μs and a gradient separation of 1.8 ms. The 9 b-values used were 0, 1, 3, 6, 10, 15, 20, 30, and 40 s/cm². Each 3D ³He data set consisted of nine interleaved images, each with a different diffusion sensitization, and required 990 breathing cycles to complete, consuming ∼1 liter of the ³He-N₂ mixture. Images were reconstructed to 3D Cartesian grids, which were then divided into 2-mm-thick transverse slices for analysis. Final reconstructed ³He image planar resolution was 0.5 mm × 0.5 mm. We refer the reader to Ref. 15 for additional details on image acquisition and reconstruction.

After reconstruction, images were thresholded to eliminate background noise from the diffusion maps. A model-based approach of acquiring quantitative information about air space size was then implemented; we refer the reader to Yablonskiy et al. (42) for full details of the model. The voxel-average diffusivity, D_ave, diffusion anisotropy D_an, and apparent airway radius R_mean were calculated (using equations 5 and 8 in Ref. 42) on a pixel-by-pixel basis using the Non LinearRegress function of Mathematica (Wolfgram Research, Champaign, IL). Pixels with a standard error (i.e., fit uncertainty) exceeding 50% were discarded and assigned a value of zero. Global median values for the whole lung diffusivities were then calculated for each rat. Global means were not used because the distributions of diffusivities were nonnormal (typical skewness = 7.1). In such cases, medians are not distorted by extreme values, such as high diffusivities in the trachea and major airways, but generally provide a useful measure of the typical value (24). The ³He free diffusivity was also calculated for each animal by measuring the apparent diffusion coefficient from the first two b-values in several slices that included only the trachea, which was aligned generally parallel with the z-axis of the magnet.

After calculation of the diffusion coefficients, 8–10 regions of interest (ROI), each ∼5 mm in diameter and generally excluding major airways, were selected from the 2D slices of each 3D image set. These regions were then used to guide histological sampling of the lung tissue with the intention of facilitating localized correlation between MRI and histology.

Morphometric analysis.

Following imaging, animals were humanely killed using CO₂ asphyxiation. The lungs were exsanguinated, harvested, trimmed, and filled with 10% buffered formalin to a pressure of 25 cmH₂O (19) using a Mariotte bottle system. After filling, the trachea was tied off and the lung placed in a formalin bath for >48 h. We estimate tissue shrinkage due to fixation to be approximately 10–15%. The lungs were then sectioned into 2-mm-thick transverse slices using the MR images as a guide. Lung sections that corresponded to the image slices that had a circled ROI were retained, resulting in a total of 101 sections. After imbedding in paraffin, a representative 5-μm-thick slice of each section was placed on a microscope slide for histological analysis.

For a global (i.e., averaged over the whole lung) comparison between MRI and histology, slides were examined at 100× magnification (10× objective, 10× ocular), and 1,600 × 1,200 pixel digital images (covering ≈1.0 mm × 0.7 mm) were acquired for analysis. Global characterization was performed by taking 8–10 digital images of tissue sections for each rat; images were taken from all histological sections to span the entire lung. Generally, large airways and blood vessels were avoided. By design, histological sampling was guided by the ROIs initially identified in ³He images. In many cases, however, histological sections bore little resemblance to the ³He images they were intended to represent. Primarily, this discrepancy occurred because lung shape was generally distorted during sectioning. This is illustrated in Fig. 1, where the histological slide (Fig. 1A) and the MRI slice (Fig. 1B) it was supposed to represent are shown.

Fig. 1.

Histological slice (A) and ³He image slice (B) that were intended to represent the same transverse lung section. The distortion of the lung tissue in A due to sectioning illustrates the difficulties encountered in attempting localized correlations between histology and MRI.

Because of difficulty matching ³He ROIs to histological sections, a different approach was adopted for performing local correlations. To prevent bias, this was done without knowledge of the dose history, animal identity, or the ³He diffusivity results within each slice. Each histology slide and a printout of the corresponding ³He spin-density image (such as that shown in Fig. 1B) were examined together and evaluated qualitatively for regions that could be matched with a high degree of confidence, as judged by the blinded reviewer. Of the 101 slides from all the rats, a total of only 12 regions were deemed to match sufficiently well for analysis. Once a region was selected, its location was marked on the ³He image, and six slightly overlapping 1,600 × 1,200 pixel digital microscope images were acquired within the corresponding region on the slide covering an approximately rectangular region of ≈2.0 mm × ≈2.3 mm.

Analysis of all digital histology images was done using ImageJ (32). Image sets for local correlation (those consisting of 6 images) were first assembled into a mosaic using MosiacJ (35), an ImageJ plugin. All images were filtered with a 2.0-pixel radius Gaussian blur to remove speckle and a 200-pixel radius rolling ball background subtract filter to remove background intensity variations. Next, the optical images were threshholded and examined. Particles were removed, major airways, and blood vessels were manually filled in, and unthresholded tissue regions were repaired. Figure 2 shows an example of the image processing sequence for an image mosaic. Finally, the areas of the individual air spaces were measured by selecting the air spaces that were entirely bounded by tissue, measuring the areas, and reporting all results in a spreadsheet. The equivalent diameter D_eq of each air space was then calculated and converted to micrometers using a microscope calibration. The average equivalent diameter μ for the entire image mosaic was found, and D₁ and D₂ were then calculated to represent the air space size distribution of each mosaic. We note that partial airways truncated by the edge of the image mosaics were excluded (see Fig. 2). For the smaller single images used in the global comparison, all air spaces were included in the analysis. Airways bordering the edge were therefore defined by the edge of optical data. In the emphysematous samples most of the largest air spaces were truncated by at least one edge of the 1,600 × 1,200 image frame, thereby skewing the results toward somewhat smaller values. Although this method is imperfect, excluding the edge-bordering airways would serve to further skew the results toward lower values by eliminating the largest air spaces altogether. Several of the images were randomly selected and reanalyzed with all edge-bordering airways excluded; this test confirmed the hypothesis. The exclusion of edge air spaces in the saline controls had little effect.

Fig. 2.

An example of the sequence for processing of image mosaics. A: raw image of the assembled mosaic. B: after filtering and background subtraction. C: after threshholding. D: after manual repair. Scale bar represents ∼200 μm.

To calculate the diffusion results in each ³He image region corresponding to a mosaic, the ³He spin-density image and D_ave, D_an, and R_mean maps were opened in ImageJ. An image stack was created with the spin-density image on top. Looking only at the spin-density image, a 2.0 × 2.5 mm rectangular region (i.e., 4 pixels × 5 pixels) was selected at the location chosen by the blinded reviewer. Without moving the selected region, the average values of D_ave, D_an, and R_mean were then measured. Pixels within the region with a value of zero were excluded from the average. Figure 3 shows an example of this process. Figure 3A shows a digital photograph of the histological slide with the targeted region outlined in black. The blinded observer deemed this region to correlate with the region in the ³He image (Fig. 3B) with high confidence. Figure 3, C and D, shows the D_ave and D_an maps, respectively, with the same region outlined (the R_mean map is not shown). Hence, the histology results from the mosaic images were correlated with the average MRI results from the corresponding region. We note that the mosaic in Fig. 2 is from the selected region in Fig. 3A.

Fig. 3.

Illustration of one of the 12 regions, this from rat E9, used for the localized correlation between histology and MRI. A: photograph of the histology slide (only the right lung is shown) with a selected region outlined in black. B: the corresponding ³He spin-density image with the region outlined. C: the average diffusivity (D_ave) map. D: the diffusion anisotropy (D_an) map. The bottom color scale for both diffusion images ranges from 0.0 (gray, left) to ≥0.8 cm²/s (purple, right).

RESULTS

It was observed that each animal exhibited a different EEP, despite each being ventilated in exactly the same way. The signal-to-noise ratio (SNR) for ³He images also varied from rat to rat, presumably, due to differences in gas polarization and/or mixing with N₂ in the Tedlar bag. Figure 4 shows that the global median value of D_ave for the saline controls varied linearly with the product of EEP and SNR over the range of measured diffusion coefficients (R = 0.90, P < 0.005). In principle, this is not surprising since differences in EEP indicate that diffusion data were likely collected at slightly different levels of lung inflation, and because measured diffusion is affected by ³He SNR (25). To facilitate comparisons between different animals and treatment groups, an empirical, linear correction was made to all D_ave values to EEP = 0 cmH₂0. Linear variations with the product of EEP and SNR were also found for D_an (R = 0.85, P < 0.01) and R_mean (R = 0.91, P < 0.002), with respective corrections made to those data. We note that D_ave error bars are not shown in Fig. 4 due to the lack of a conventional measure of uncertainty of a median value. However, the average fractional error determined from the numerical fitting for D_ave ranged from 0.10 to 0.27 for the highest and lowest SNR saline rats, respectively.

Fig. 4.

Plot showing the empirical relationship between the global median of D_ave vs. the product of ³He signal-to-noise ratio (SNR) and end-expiratory pressure (EEP) for the saline-treated control rats. Data for all rats were then corrected to the y-intercept (EEP = 0 cmH₂O). Similar relationships were found for, and corrections made to, D_an and mean airway radius (R_mean).

Table 1 shows physiological and MRI data collected from each rat. The rat label refers to the type of dose given (S for saline, E for elastase) and the rat number. Saline and elastase rats of the same number were imaged on the same day. Rats S1 and E1 were eliminated from the analysis due to technical problems encountered during imaging.

Table 1.

Data from 16 of the 18 rats in the study

Rat	BW, g	LW, g	LV, ml	EEP, cmH2O	3He S/N	D0, cm2/s	Dave*, cm2/s	Dan*, cm2/s	Rmean*, mm
S2	271	1.700	5.148	1.49	12.46	1.08	0.0696	0.219	0.148
S3	240	1.308	4.918	2.98	16.95	0.94	0.0762	0.222	0.157
S4	323	1.631	6.770	2.24	13.06	1.03	0.0749	0.218	0.150
S5	309	1.443	4.827	3.59	21.92	0.97	0.0770	0.228	0.153
S6	261	1.347	3.540	2.73	21.52	0.99	0.0747	0.222	0.147
S7	344	1.616	5.854	4.33	16.10	0.97	0.0781	0.235	0.153
S8	347	1.759	7.596	2.24	18.24	1.03	0.0850	0.247	0.158
S9	337	1.542	6.973	1.37	15.25	1.02	0.0846	0.243	0.157
E2	292	1.881	6.857	2.24	14.15	0.95	0.0728	0.221	0.145
E3	240	1.877	7.188	1.74	15.90	0.97	0.0977	0.269	0.174
E4	267	1.461	7.405	3.22	14.08	1.07	0.1005	0.282	0.173
E5	321	1.482	3.511	2.11	21.65	1.01	0.0858	0.246	0.159
E6	298	1.920	10.040	3.59	19.24	†	0.1097	0.308	0.170
E7	381	1.823	5.796	2.36	18.33	1.01	0.0967	0.272	0.166
E8	386	2.017	9.323	2.48	18.84	1.02	0.0870	0.255	0.160
E9	356	1.908	8.089	3.47	11.77	1.01	0.0892	0.277	0.161

S rats were given saline dose; E rats were given elastase dose. Two rats, S1 and E1, were excluded due to technical problems with the imaging data. BW, body weight at time of imaging; LW, lung weight at time of necropsy; LV, lung volume after fixation determined by fluid displacement; EEP, end- expiratory pressure; ³He S/N, average global ³He signal-to-noise ratio for the b = 0 s/cm² image, excluding major airways; D₀, ³He free diffusivity measured in the trachea.

^*Data for average diffusivity (D_ave), diffusion anisotropy (D_an), and mean airway radius (R_mean) are corrected global medians from MRI.

^†Insufficient data.

Figure 5 shows an example of a 3D data set from rat E4 for which the D_ave values have been calculated for each pixel using the Yablonskiy model (42). Each slice is 2-mm thick with a planar resolution of 0.5 mm × 0.5 mm. The color scale (in units of cm²/s; see Fig. 5 legend) shows the range of values of D_ave; major airways are generally visible as regions of the highest D_ave values.

Fig. 5.

An example of a full 3D data set, taken from rat E4, showing D_ave calculated on a pixel-by-pixel basis from ³He MRI data and corrected to EEP = 0 cmH₂O. The left lung is on the observer's left. Planar resolution is 0.5 × 0.5 mm, and the slice thickness is 2 mm. The bottom color scale ranges from 0.0 (gray, left) to ≥2.5 cm²/s (purple, right).

Figure 6 shows a comparison between two rats, rats S6 and E6, both imaged on the same day 2 wk after dosing. Figure 6 demonstrates differences in ³He diffusivity between a saline-dosed and an elastase-dosed animal. Three approximately corresponding representative slices are shown for each animal; the color scale is the same for each. The saline-dosed animal (Fig. 6, column A) has significantly lower D_ave (P < 0.0001) and less apparent heterogeneity than the elastase-dosed animal (Fig. 6, column B).

Fig. 6.

Comparison between calculated D_ave values in approximately corresponding slices from a saline-dosed rat (A) and an elastase-dosed rat (B), rats S6 and E6, respectively. The bottom color scale ranges from 0.0 (gray, left) to ≥2.5 cm²/s (purple, right). Data were corrected to EEP = 0 cmH₂O.

For the global histological analysis, the overall mean D_eq (i.e., μ), its variance (σ²), and skewness (γ) were calculated by combining the results from the 8–10 microscope images from each rat. D₁ and D₂ were then calculated using Eqs. 1a and 1b (see Table 2). The fact that D₂ > D₁ is largely due to the skewness of the D_eq distributions. For the saline-dosed rats, the average value of μ was 35.5 ± 3.4 μm and the average value of D₂ was 102 ± 13 μm. These uncertainties are comparable to those reported for L_m in the literature (for example, Ref. 9 reported an L_m of 37 ± 10 μm in 7 control rats, and Ref. 28 reported an L_m of 85 ± 14 μm in 6 control rats). Therefore, we assume that any consequences of tissue processing did not have an unusual effect on our data analysis.

Table 2.

Global histological results

Rat	μ, μm	σ2	γ	D1, μm	D2, μm
S2	30.77	648.03	2.2856	51.835	79.416
S3	40.73	1324.78	2.1908	73.256	112.30
S4	35.55	996.07	2.8170	63.568	106.08
S5	35.23	1175.83	3.0967	68.606	121.17
S6	30.86	739.37	2.5834	54.818	88.537
S7	36.60	1033.43	2.4723	64.835	103.09
S8	36.01	1082.87	2.7781	66.081	110.39
S9	38.01	981.16	2.4394	63.822	99.661
E2	37.73	1270.06	2.5269	71.392	115.77
E3	52.55	2315.44	2.1770	96.611	148.26
E4	53.15	2594.00	2.7272	101.95	170.53
E5*	30.34	881.67	3.4295	59.400	109.84
E6	56.52	3854.04	2.6469	124.71	208.18
E7	38.07	1272.48	2.6047	71.495	117.11
E8	43.53	1509.03	2.6031	78.197	126.95
E9	45.80	1933.69	3.1648	88.020	156.49

μ, σ², and γ are the mean, variance, and skewness of equivalent diameter D_eq; D₁ and D₂ are weighted indexes of air space size distribution.

^*All sections of lung E5 appeared underinflated on the histological slides.

Figure 7 is a plot of the MRI parameters D_ave and R_mean vs. D₂; correlations between the global ³He diffusivity results (Table 1) and D₂ (Table 2) for each rat are shown in Table 3. We note that D_ave and D_an correlate strongly with each other in both the global and local measurements (R = 0.99 and R = 0.97, respectively), implying that uniform scaling of the airway structures is occurring. Thus the plot of D_an vs. D₂ is not shown. The dashed lines in Fig. 7 represent the 95% confidence intervals from the mean of the saline control group, as calculated from the Student's t-distribution. That is, points lying to the right or above the dashed lines have high confidence of representing emphysematous rats according to either MRI or histology, respectively. For D_ave, two saline controls were outside the confidence interval, while three were outside for R_mean. Only a single elastase-dosed animal, rat E2, was within the confidence interval for D_ave and R_mean. For D₂, two saline controls were outside the confidence interval while one elastase animal, rat E5, was within. We note that rat E5 appeared quite underinflated on the histological slides, thus resulting in low histological measurements. Because the elastase-treated rats were imaged at various time points and therefore presented a range of disease response, we examined the confidence intervals based on the mean of the saline group, as opposed to attempting direct comparisons between the elastase and saline groups.

Fig. 7.

Correlations between global results of weighted index of air space size distribution D₂, and D_ave and R_mean. The dashed lines indicate the 95% confidence interval from the mean values of the saline controls. Fit quality parameters are shown in Table 3.

Table 3.

Correlation parameters for global and local measurements of histology and MRI

	Dave	Dan	Rmean
Global
μ	R = 0.86	R = 0.87	R = 0.80
	P < 0.0001	P < 0.0001	P < 0.0005
D1	R = 0.85	R = 0.86	R = 0.79
	P < 0.0001	P < 0.0001	P < 0.0005
D2	R = 0.85	R = 0.88	R = 0.77
	P < 0.0001	P < 0.0001	P < 0.001
Local
μ	R = 0.81	R = 0.85	R = 0.42
	P < 0.002	P < 0.0002	P = 0.175
D1	R = 0.88	R = 0.93	R = 0.50
	P < 0.0002	P < 0.0001	P = 0.196
D2	R = 0.90	R = 0.93	R = 0.56
	P < 0.0001	P < 0.0001	P = 0.0593

The local correlations between the MRI results and the mosaic histological results are shown in Fig. 8. The quantities μ, D₁, and D₂ are plotted against D_ave. Significant correlations between the ³He diffusivities and histological measurements were found, as shown in Table 3. Data for the mean airway radius R_mean were not found to correlate significantly.

Fig. 8.

Plots showing the correlations between the local MRI ³He diffusion, D_ave, and the local histology measurements, μ, D₁, and D₂. Fit quality parameters are shown in Table 3.

DISCUSSION

We found a significant correlation between ³He diffusion measurements and histology on both a global (i.e., over the whole lung) and localized basis (see Table 3). Figure 7 shows good distinction between the saline controls and elastase-dosed animals for both the global MRI results and histological measurements, with most of the elastase-dosed animals lying outside the saline control 95% confidence interval. Therefore, we conclude that both MRI and the histological measurements are sensitive to air space enlargement in the elastase animal model.

On a local level, a highly significant correlation between the localized size-weighted indexes D₁ and D₂, and D_ave was found, as shown in Fig. 8. Table 3 also shows that the correlation between the MRI results and the histological measurements improves as higher-order moments are considered, implying that the macroscopic MRI measurements contain information about the microscopic air space size distribution, more so than they do about the mean diameters (μ). We speculate that this is because the higher moments are weighted by air space size, and MRI is essentially a volume-weighted imaging technique; proportionally more signal within each voxel comes from the larger air spaces. These results provide the most compelling evidence yet that ³He MRI diffusivity results are accurate on a local basis, suggesting that in vivo ³He MRI can be used as a noninvasive histological tool to assess disease. The results also suggest that D₁ and D₂ are more sensitive to localized tissue destruction than a numerical average of air space size. This supports the assertion of Parameswaran et al. (27), who showed that the presence of a few large blebs, which may be indicative of early emphysema, surrounded by many small air spaces will be masked by a numerical average or by traditional point-counting methods. We note that, although R_mean correlated somewhat better with D₂ than with μ, the local correlation between R_mean and histology was not significant. We attribute this to the low SNR in many of the image sets. Since R_mean is the result of fitting two complicated functions, fit uncertainties are expected to be higher as SNR decreases. Because R_mean was calculated on a pixel-by-pixel basis, random variations in R_mean are likely to be averaged out globally while they remain pronounced locally. With improved SNR, one could better test the assertion of Fichele et al. (11), whose computer simulations indicated that R_mean may not be a good indicator of mild emphysema.

The Yablonskiy model (42) is a two-parameter model that assumes a locally homogeneous distribution of air space sizes. We have previously shown that the model describes the MRI diffusion data exceedingly well in healthy rats (15), where the assumption is more rigorously satisfied than in the current study. However, we have continued to use this model to enable consistent comparisons with previous work. Although fit parameters in emphysematous rats are not readily interpreted in terms of Yablonskiy's original model, the strong correlation between histology and MRI data shown herein suggests that the Yablonskiy model is able to describe the signal decay for a broader range of conditions than the specific case used for derivation.

Localized correlations in this study were successful, although only 12 regions from the 101 histology slides were deemed suitable for comparison. Generally, incomplete use of histological samples was dictated by the difficulty in confidently matching specific locations with those in corresponding ³He images. In practice, this difficulty arose from two problems. First, the in situ shape of the lung was not well preserved, so the shapes of the lung sections did not well match those of the ³He images. Lungs that have been removed from the chest can change significantly from the in situ shape defined by the thoracic cavity (14, 30). Second, the lungs were not rigid during sectioning. The nonrigidity of the fixed lungs allowed the lobes to mislocate from their correct relative in situ positions (cf. Fig. 1). More rigid lung fixation, done preferably in situ and perhaps accomplished with long-term (i.e., >3 wk) soaking in a formalin bath, may facilitate future local correlation studies. Freezing the lungs for sectioning, as performed by Woods et al. (41), may also be a viable option.

In this study the lung inflation levels were different for imaging and histology. However, a valid correlation between ³He diffusivity and histology is possible because no attempt was made to quantitatively calibrate the ³He diffusion coefficient with an absolute airway size. Furthermore, the lung inflation level does not affect the intrinsic tissue structure, and hence does not affect air space size distribution; only the average airway size is affected. Once the MRI data were corrected for variations in EEP, the data were assumed to fairly represent each rat at some constant level of inflation for which EEP = 0 cmH₂O. With the lungs all inflated to the same pressure for fixation, a qualitative comparison between the histology and MRI could then be made.

The strong dependence of the ³He diffusivity on the small differences in EEP observed in this study was unanticipated, although previous observations by the authors and those reported by others (5) have indicated that large differences in breath hold level significantly affect measured ³He diffusivity. Although the cause of the variations in EEP was not clear, we propose that variations in lung compliance may have played a role. It has been reported that rat-to-rat variability in pulmonary compliance and functional residual capacity (FRC, the lung volume after full, passive exhalation) exists, even within healthy groups (17, 26). If it is assumed that FRC is measured at EEP = 0 cmH₂O, then an EEP > 0, such as reported herein, can only be obtained when there is incomplete exhalation. This may be the case when insufficient time is allowed for full exhalation, and it would be particularly pronounced in rats with relatively high lung compliance (or low elasticity). In this work, all rats were allowed to exhale for the same fixed duration. Hence, we speculate that natural variations in compliance may have contributed to the differences in EEP. In this case, a simple, empirical correction based on the results of the saline controls was implemented to allow fair assessment of the data. A proposed experimental solution to this problem may be to acquire ³He data at a controlled breath hold level above EEP. However, a correction to the data may still be necessary due to lung compliance differences. This is because the degree of tissue distension or alveolar recruitment—both of which affect air space size and ³He diffusivity—at any given inflation pressure or volume may be affected by the lung compliance, particularly in a severe disease state (21). How such a correction would be made remains unclear and remains to be the subject of future study. Therefore, acquiring data for all animals at the same point in the breathing cycle (i.e., EEP) may currently be the best method of ensuring experimental consistency.

Localized disease assessment may allow for improved early diagnosis for COPD patients. It may also be a useful research tool in preclinical pulmonary studies requiring screening of large populations of animals. In addition, accurate, localized detection of disease distribution may facilitate treatment with innovative new therapies, such as magnetically targeted delivery of aerosolized pharmaceuticals (7, 31).

GRANTS

This work was funded by Battelle and by National Heart, Lung, and Blood Institute Grant RO1-HL-073598.