INTRODUCTION

Top Abstract Introduction Methods Results Discussion References

THE OBJECTIVE of the present study was to test the hypothesis that a three-dimensional (3-D) gradient-echo magnetic resonance imaging (MRI) technique has the sensitivity to detect pulmonary edema without the use of contrast agents and can provide time-course data that can be used to characterize the transport of fluid across the lung microvascular barrier. To that end, an MRI scanning protocol was developed and tested, and the results were compared with other standard methods of measurement in a canine lung injury model. These methods included multiple-indicator dilution (MID) and wet-to-dry ratios. Although comparison with standard methods required integration of MRI signals over the whole lungs in this work, the 3-D nature of the MRI data would readily allow computation of parameters on a regional basis.

A dysfunction of the lung microvascular barrier leads to imbalances of fluid and solute exchange, interstitial pulmonary edema, and, ultimately, respiratory failure. This is theorized to be what occurs in acute respiratory distress syndrome (ARDS) (2, 20), a life-threatening clinical syndrome. Reported survival rates have a wide range, from 30 to 80% (2, 5, 26). The true incidence of ARDS is unknown, although it has been estimated to be from 1.5 to 71 patients per 100,000 population per year (2, 5, 37). The ability to quantify noninvasively the integrity of the lung microvascular barrier and the severity of pulmonary edema would add power to evaluating these patients and measuring response to treatment. Indeed, it has been shown that high microvascular permeability is correlated with severe ARDS (20).

In studying the microvascular barriers, investigators have devised many measurements to quantify the barrier integrity. By using gravimetric techniques (16), transvascular fluid filtration was measured, and the capillary filtration coefficient (K_fc) was estimated. Others have used optical techniques to make similar measurements (19). Gamma emitter scanning of materials that escape the capillaries provided a quantitative method for measuring the pulmonary transcapillary escape rate (PTCER) (18). Positron emission tomography (PET), based on positron-emitting radiolabeled macromolecules, has been used to obtain a similar measure for 3-D information (34). The MID technique is much more portable and less expensive than PET and offers a procedure for measuring tracer exchange, flow, and volumes (9). The MID technique involves a mixture of labeled tracers injected as a bolus into the inlet of an organ, with the subsequent concentrations of the tracers at the outlet examined as a function of time. The MID technique is used in this study to quantify and characterize injury. Numerous models have been used to analyze the MID concentration-time curve data and compute parameters (7, 10, 22). Models used in this research are described later and in the references as noted.

MRI is another method used to study pulmonary edema. Using many repetitions of a one-dimensional (1-D) spin-echo line-scan technique, Cutillo et al. (11-13) and Hayes et al. (23) demonstrated good correlation between MRI signal and lung water measured gravimetrically in rat models of pulmonary edema. Normalization of magnetic resonance (MR) lung signal to that in a water phantom provided good estimates of lung water from the MRI in these studies. Two-dimensional (2-D) spin-echo studies on sheep with hydrostatic edema (6) and rats with permeability edema (32) also demonstrated increasing MRI signal with increasing lung water content. Gravity-dependent signal gradients in 2-D spin-echo MRI have also been demonstrated in human lungs (6, 27, 28). Ventilatory conditions have also been shown to affect lung MRI signals (11, 28). In additional MRI and nuclear magnetic resonance spectroscopic studies in rats and dogs (in vitro and in vivo), researchers have demonstrated that, in addition to the obvious increase in spin density that occurs during edema, the relaxation time constants T1 (13, 32, 33, 35, 39) and T2 (32, 33, 35, 39) increase with increasing severity of edema. Just as edema forms nonuniformly, the increases in T1 and T2 occur nonuniformly in the lung volume. Finally, Berthezene et al. (4) demonstrated the ability to differentiate cardiogenic and noncardiogenic (i.e., leak) edema by using MRI with a macromolecular contrast agent.

Typical MRI of the lung is plagued with certain unavoidable limitations (17). Unique to the lung is the enormous number of air-tissue interfaces. Each interface contributes to the heterogeneity of magnetic susceptibility, which leads to signal loss, thus making parenchyma and small vessels difficult to visualize. Other significant limitations in pulmonary imaging are the low proton density of the lungs and the limited spatial resolution of the MRI when imaging whole lungs. At an interface between tissues having different magnetic susceptibilities, a local distortion of the magnetic field occurs. This distortion causes neighboring nuclei to spin or precess at different frequencies, thereby causing them to get out of phase with each other. This dephasing markedly reduces the measured T2* (a time constant that reflects the rate of dephasing caused by spin-spin and other interactions) in the inflated lung. The shortened T2* leads to very low image signals in gradient-echo scans and, if motion including diffusion occurs, in conventional spin-echo scans.

To overcome the effects of the shortened T2*, reduction of the time during which the unwanted dephasing occurs is beneficial (3). This dephasing occurs between the time of the initial radio-frequency tipping pulse and the signal echo [echo time (Te)]. Minimizing Te limits the time available for unwanted dephasing and can help increase signal.

This study presents the application, in a canine oleic acid model of ARDS, of a specially designed 3-D gradient-echo MR scan for noninvasively measuring regional pulmonary edema in the entire lung volume (8). With the use of a shortened Te and signal averaging, this gradient-echo scan is sensitive enough and fast enough to visualize regional pulmonary edema and measure its time course development without utilizing a contrast agent. In images created with this scan, the signal within a volume element (voxel) is a function of the water content within that voxel. Areas with more water generally have a higher signal or, correspondingly, appear brighter in an image.

	METHODS

Top Abstract Introduction Methods Results Discussion References

Animal model. An in vivo canine model of ARDS (1, 30, 40) was used for the experiments. After they were anesthetized with pentobarbital sodium (49-65 mg/kg initial dose, 6 mg · kg¹ · h¹ iv for maintenance), nine mongrel dogs (18.5 ± 2.5 kg) were intubated and respirated with an appropriate tidal volume (~750 ml) at 10 breaths/min by using a mechanical ventilator (Foregger, Smithtown, NY). Two catheters were placed into the right jugular vein: one for venous injections, and the other for the intravenous anesthesia drip. For arterial blood sampling and pressure measurements, a 1/8-in.-diameter catheter was placed in the right carotid artery. An introducer (Cordis, Miami, FL) was used in the other jugular vein to float a nonmetallic, MRI-specific balloon catheter (6.8-Fr, 70 cm; Cook, Bloomington, IN) into a pulmonary artery. Guidance, when needed, was provided via X-ray fluoroscopy. Before the dog was placed into the MRI scanner, MID curves were collected by using ³H₂O, ⁵¹Cr-erythrocytes, ¹²⁵I-albumin, and [¹⁴C]urea or [¹⁴C]butanediol as the tracers to measure baseline values of cardiac output (CO), extravascular lung water (EVLW), and permeability-surface area product (PS).

1

1

1

1

MRI methods. The specialized procedure for MRI of pulmonary edema has been presented in detail elsewhere (8). The images are acquired by using a syncopated gradient-echo sequence with repetition time (Tr) of 18 ms, flip angle of 12°, and two acquisitions. The echo time is 4.7 ms. A syncopated gradient-echo pulse sequence is one that is periodically interrupted to allow for more longitudinal relaxation than can occur between two excitations and, in time-of-flight MR angiography applications, for inflow of unsaturated spins. Image acquisition typically lasts 9 min and 20 s. During image reconstruction, the 256(x) × 180(y) × 64(z) raw data matrix is interpolated to 256 × 256 × 64. The imaging volume is oriented transversely, with a typical excited slab thickness (z-axis) of 192-215 mm and field of view (FOV; x- and y-axes) of 320-360 mm. In the presented studies, the slab thickness was chosen to be large enough to include the entire canine lung from base to apex. Furthermore, the FOV was chosen to be roughly twice the anterior-posterior dimension of the lung field to prevent coherent respiratory ghosts from overlapping the lungs.

(1)

(2)

(3)

	RESULTS

Top Abstract Introduction Methods Results Discussion References

The physiological data from the experiments on all nine animals are summarized in Table 1. Neither mean systemic nor pulmonary arterial blood pressure changed from baseline to injury; however, partial pressures of oxygen and carbon dioxide in the blood did change. On average, the animals became more acidotic. The CO, as measured by the indicator dilution method and indicated in Table 1, dropped significantly from 2,263 ± 765 ml/min at baseline to 1,512 ± 321 ml/min after injury (n = 9, P = 0.006). However, flow heterogeneity, reported as relative dispersion of microsphere-computed regional flow, did not change. Edema is indicated not only by WLW-to-DLW ratio but also by EVLW and changes in PS, both of which are derived from the MID data. As reported in related work (8), the WLW-to-DLW ratio was 5.30 ± 0.38, which is greater than the baseline value of 3.7 found previously (31), and EVLW increased significantly from a baseline value of 2.03 ± 1.11 to 3.00 ± 1.45 ml/g after injury (n = 9, P = 0.004). These indicators of lung injury are presented in Fig. 1.

View larger version (11K): [in this window] [in a new window]   Fig. 1.   Pulmonary injury is indicated by a number of parameters. This plot summarizes 4 ratios as indicators of lung injury. Permeability-surface area product (PS) of urea and butanediol (PSU and PSB, respectively) and effective diffusivity (D1/2S) ratios indicate injury to capillary boundary. Increase in extravascular lung water (EVLW) verifies presence of edema, as does the ratio of wet-to-dry lung weight (Wet:Dry). * Significant difference between baseline and injury values, P < 0.05.

The PS values in Table 1 and Fig. 1 were calculated by using the Crone model (10) and assuming homogeneous blood flow. The PS for the urea tracer (PS_U) did not change significantly from baseline to injury but trended down from 5.64 ± 4.40 to 3.96 ± 2.66 ml/s, respectively. PS for the butanediol tracer (PS_B) moved from 8.63 ± 4.88 to 4.18 ± 2.02 ml/s (P = 0.004). The ratio of PS_U to PS_B, an indicator of P which is insensitive to S (31), increased significantly, from 0.62 ± 0.17 to 0.97 ± 0.47 (P < 0.05). Taking into account the blood flow heterogeneity, we calculated D^1/2S by using the variable recruitment model. Results are shown in Table 1 and Figs. 1 and 2. Like the homogeneous PS values, heterogeneous D^1/2S for urea remained virtually unchanged and that for butanediol dropped with injury and reduced CO. D^1/2S_U changed from 0.148 ± 0.009 to 0.124 ± 0.006 ml · s^0.5 · g¹, and D^1/2S_B changed from 0.243 ± 0.012 to 0.140 ± 0.02 ml · s^0.5 · g¹ (P < 0.01). The ratio of D^1/2S_U to D^1/2S_B increased significantly, with the average increase from 0.583 ± 0.027 to 0.852 ± 0.154 (P < 0.05). The two ratios, heterogeneous D^1/2S and homogeneous PS, were well correlated at baseline (R = 0.96, P < 0.001) and after injury (R = 0.90, P < 0.001). Results from all animals (n = 9) were used to determine these PS and D^1/2S values.

View larger version (17K): [in this window] [in a new window]   Fig. 2.   D1/2S as measured by urea, butanediol, and their ratio. Top: D1/2Surea changes very little, but D1/2Sbutanediol decreases significantly (middle). Bottom: ratio, a surface area-independent indicator of permeability, increases significantly. DMO 3-11, 9 dogs.

The MRI showed the development of edema over time. At each time point, an imaging volume consisting of 64 imaging slices and covering the entire canine lung was acquired as an average over the 9-min duration of the scan. Figure 3 shows, for baseline (A) and 90 min postinjury (B), a slice from a caudal portion in the imaging volume for one of the subjects. Edema is evident as the brighter area within the dorsal region of the lung. By defining a small circular region of interest in this injured area for each of the images taken over time, the mean SNR can be calculated as a function of time within this injured region. By defining the region of interest to be the whole lung (and only the lung), whole lung SNR is found. Figure 4 depicts two images from a sample subject, one at baseline (A) and the other at 90 min postinjury (B), recreated from projections of the 3-D lung-image volume. Edema is evident in the dorsocaudal portion of the lungs in Fig. 4 and is similar to the pattern observed in all subjects. SNR increases in a dorsocaudal region of interest where injury was evident ranged from 161 to 629% (P < 0.01 for baseline vs. postinjury regional SNR). Because no obvious signal change occurred in large portions of the lungs, whole lung SNR increases ranged from only 1 to 106%.

View larger version (62K): [in this window] [in a new window]   Fig. 3.   Sample slice from 3-dimensional (3-D) lung edema magnetic resonance image (MRI) volume. Same imaging slice taken from a caudal portion of lung image volume at baseline (A) and at 90 min postinjury (B). Edema is clearly present as brighter areas in dorsal portion of lungs (arrow).

View larger version (50K): [in this window] [in a new window]   Fig. 4.   Maximum-intensity projection of lungs. Projection of lung-imaging volume from a sample subject. At baseline (A), little can be seen in the lungs except some major vessels. At 90 min postinjury (B), edema can be clearly visualized as the brighter areas in dorsocaudal region of lungs.

The total SNR within the lungs of all but the first subject is plotted as a function of time in Fig. 5. Whole lung SNR at baseline and injury correlated significantly with the PS ratio and the D^1/2S ratio as indicators of injury. Between SNR and PS ratio, the correlation coefficient was 0.64 (P = 0.004), and between SNR and D^1/2S, it was 0.58 (P = 0.012). Both correlations were computed for nine dogs, with baseline and injury values for each. Change in whole lung SNR also correlated significantly with the postinjury D^1/2S ratio (n = 9, R = 0.70, P = 0.04; Fig. 6).

View larger version (24K): [in this window] [in a new window]   Fig. 5.   Signal-to-noise ratio (SNR) in whole lung as function of time. By taking the whole lung as region of interest, one can see time-course development of edema. In this graph, 8 symbols, each representing a single dog, are plotted at time coordinate corresponding to beginning of MRI scan. Each MRI scan lasted just over 9 min. Oleic acid was injected at time 0. Baseline data are plotted at time 10 min for scale convenience; actual time of baseline scanning was >10 min before injury. The first dog was omitted from this plot because field of view and receiver gain were not held constant over the course of the experiment; this interfered with automated image segmentation and SNR comparisons.

View larger version (8K): [in this window] [in a new window]   Fig. 6.   Change in whole lung SNR vs. postinjury D1/2S ratio. Change in SNR (dimensionless) from baseline to last postinjury measurement correlates with postinjury D1/2S ratio (n = 9 dogs, R = 0.69, P = 0.04).

From the plot of whole lung SNR vs. time, the slope from the linear region is taken as the dS/dt (see Fig. 7 for an example). As described in Eq. 2, the product of dS/dt and  is the mismatch between _l and _f. The average _f  l mismatch was 0.76 ± 0.61 ml/min (0.015 ± 0.011 ml · min¹ · g¹ DLW, n = 8). Table 2 lists the  and interstitial fluid accumulation rate (_f  l) for each animal. To calculate K_fc, values must be assumed for the effective driving pressure difference and _l. Given the assumed values described in METHODS, the average K_fc divided by DLW was 0.025 ± 0.016 ml · min¹ · cmH₂O¹ · g DLW¹ and divided by WLW was 0.0040 ± 0.0026 ml · min¹ · cmH₂O¹ · g WLW¹ (n = 8). Individual results for K_fc are presented in Table 2. These estimates of K_fc correlated with the marker of injury, as indicated by heterogeneous D^1/2S ratio (R = 0.70, P = 0.05 for K_fc normalized to either grams DLW or grams WLW) as shown in Fig. 8, and tended also to increase with an increase in homogeneous PS ratio (R = 0.50 for K_fc/g DLW; R = 0.54 for K_fc/g WLW), but there was no significant correlation.

View larger version (13K): [in this window] [in a new window]   Fig. 7.   SNR in whole lung from 1 subject. Extracted from Fig. 5, the plot for this subject is used as an example for calculating capillary filtration coefficient (Kfc). dS/dt (where S is signal) is found by fitting a line to the points in the linear region after injury has begun (time 0). By dividing water volume at end of experiment by SNR at last time point,  (constant of proportionality between lung water volume and SNR) is found. Transcapillary filtration flow (f)  lymph flow (l) is calculated as  × dS/dt. Using a few assumptions (regarding l, P, and ), Kfc can be determined. See text for discussion.

View larger version (11K): [in this window] [in a new window]   Fig. 8.   Correlation between Kfc and D1/2S ratio. Although some assumptions are involved in calculating Kfc from MRI data, estimates of Kfc correlate with another indicator of permeability, namely, D1/2S (n = 8 dogs, R = 0.70, P = 0.05). DLW, dry lung weight.

To measure regional blood flow, we diced the lung, traced each slice, and recorded the location of each piece. The tracings of the outlines could be scanned into a computer, and each segment was assigned a gray scale value related to its flow, where brighter white refers to greater flow. Samples of this flow image for a slice at baseline and postinjury are shown in Fig. 9, A and B, respectively. The corresponding MRI slice can be recreated by reslicing the 3-D imaging volume. This is shown in Fig. 10. Notice that, for the sample slices in Fig. 9, the blood flow is generally reduced only in the animal's left lung, and specifically only in the caudomedial region. This corresponds to Fig. 10, where edema is mostly present in the same region of the left lung.

View larger version (66K): [in this window] [in a new window]   Fig. 9.   Microsphere-delineated regional blood flow. These images, representing a single slice through the lungs, are reconstructed from lung segments used to measure blood flow with labeled microspheres. A gray scale value is assigned to each piece corresponding to flow to that piece. Brighter value indicates greater flow. From baseline (A) to 90 min postinjury (B), there is a slight reduction in flow to left lung (right), especially so in the caudal portion.

View larger version (58K): [in this window] [in a new window]   Fig. 10.   Coronal images from 3-D MRI. A: baseline. By reslicing the MRI volume, these images were created to correspond roughly in position to microsphere flow images in Fig. 9. In same caudal region of left lung where flow was reduced postinjury (B), edema can be seen (arrow).

If we further examine Figs. 9 and 10, it is obvious that a direct registration of the two imaging volumes cannot be made. A rough approximation of registering the two data sets was accomplished by assuming each imaging pixel within the lung represented a constant mass. In this way, the 3-D imaging volume of just the lungs was resegmented into pieces roughly equivalent in relative mass and location within the lung. A piecewise comparison between the change in signal and the change in blood flow from baseline to injury is shown in Fig. 11 for a sample subject. For all cases, there was no significant correlation between the change in blood flow and the change in MR signal, although the tendency was for areas of greater signal increase to have a more greatly reduced flow.

View larger version (12K): [in this window] [in a new window]   Fig. 11.   Change in regional MRI SNR vs. change in microsphere flow. There is no positive correlation between change in SNR and change in flow. Indeed, the tendency is for pieces with a greater increase in signal to also have a more reduced flow (P = 0.002 for slope), although the relationship is not well correlated (R = 0.096). This supports the argument that signal increases seen in the lung postinjury are not caused by increase in regional blood flow.

	DISCUSSION

Top Abstract Introduction Methods Results Discussion References

Despite the poor imaging environment of the lungs, the MRI scan presented in this work not only visualizes edema but quantifies it and measures the time course of its development in vivo. The procedure provides a 3-D image set of an entire lung volume in <10 min without using contrast agents or ionizing radiation.

The PS and D^1/2S ratios, standard measures of lung injury, demonstrated that injury did indeed occur, and they were correlated with each other. We have generally found that PS and D^1/2S ratios correlate in lung studies. Heterogeneity has only a small effect on the numerical values of D^1/2S for urea and butanediol in the lung (7). Therefore, PS (which neglects flow heterogeneity) and D^1/2S in this study (which included flow heterogeneity) still correlate.

The extent of the oleic acid injury seen in these subjects is greatly reduced compared with that in other studies (40). This is likely because of the oleic acid dosages (0.021-0.048 ml/kg), which were lower in the present study than in other studies (36, 40) and were also varied slightly to give a range of injury. Despite the small extent of injury, the MRI scan was sufficiently sensitive to evaluate the edema. Indeed, SNR in a dorsocaudal region of interest increased from 161 to 629% in this study, despite the small extent of injury, whereas other regions showed no change in SNR at all. This might suggest that the 3-D MRI scan is sensitive to smaller changes in edema than are the more traditional non-MRI methods. Although it was necessary in this study to use the spatially integrated MR data for comparison with traditional methods for verifying physiological parameters (i.e., MID results), the 3-D MRI method presented here provides an examination of regional changes for all areas of the lungs.

The spatial resolution of the MRI volume is ~1.4 × 1.4 × 3.9 mm³. This resolution is not as fine as most nonpulmonary MR applications, but it exceeds the typical resolution of PET scans of the lung (34). The ~3.9-mm slice thickness, while allowing full coverage of the lung, adds to partial volume effects and other phenomena leading to signal loss. This causes any small vessels and capillaries to be obscured, with only the main vasculature being resolved (see Fig. 4). This is considered a virtue, because signals from blood in small vessels could confound the edema measurements.

Indeed, most MR scans are not sensitive enough to distinguish fluids in different compartments (i.e., plasma, interstitial, intracellular, and others). Signals from pooled blood would be difficult to distinguish from lung water. If gross vascular recruitment were to occur, the resultant increase in signals might be mistaken for edema. This is not the case in these experiments, because regional signal was not correlated to changes in regional blood flow (as illustrated in Fig. 11). In addition to blood distribution, anything that altered lung density would alter signal. The increased regional lung density in atelectasis, for example, could be interpreted as edema. For these oleic acid studies, however, the signal contribution caused by increased lung water is assumed to be significantly greater than any signal which may be contributed by atelectasis. Others (29), using X-ray computerized tomography of fiducial markers on the lung, have found that the dependent regions of the lung become flooded and expanded, not collapsed, during oleic acid injury.

In addition to the assumption that the signal change within the injured lungs comes from the accumulated fluid and not atelectasis, the MR signal is assumed, for purposes of K_fc calculation, to be linearly related to the water volume within an imaging voxel. Certainly, MR signal for any pulse sequence is always linearly related to spin density (), which equals the number of nuclei per unit volume (24). However, other nonlinear effects are superimposed and confound the linear relationship between MR signal and water content. As shown by nuclear magnetic resonance and MRI studies, both T1 and T2 increase as lung water content (and thus ) increases (13, 32, 33, 35, 39). The effects of these increases in T1 and T2 on MRI signal are dependent on the type of pulse sequence and parameters chosen. In a spin-echo pulse sequence, lengthening the repetition time minimizes signal dependence on T1 (13), thus better revealing the regional changes in  associated with pulmonary edema. Similarly, reducing excitation flip angle and/or lengthening repetition time minimizes T1 effects in gradient-echo pulse sequences. The flip angle and repetition time used in this work were chosen to balance the need to minimize T1 sensitivity while maintaining overall signal and minimizing scan time.

T1, which is dependent on field strength, was estimated by Cutillo et al. (13) to be ~800 ms for normal lung tissue at ~1 T. For a FLASH pulse sequence (14), on which the edema sequence was based, and for the repetition time (18 ms) and flip angle (12°) used, a 25% increase in T1 from 800 to 1,000 ms would result in an 11% decrease in signal. The periodic interruptions (syncopation) in the actual pulse sequence allow more longitudinal relaxation to occur, which would reduce the actual signal loss caused by such an increase in T1. Sensitivity of MRI signal to changes in T2 are minimized by reducing Te. For the short Te of 4.7 ms used in this work, a change in T2 from 50 to 90 ms, such as may occur with edema (35), results in only a 4% increase in MR signal for a FLASH sequence.

T1 and T2 changes associated with edema can lead to underestimation and overestimation, respectively, of the increase in . These effects add an element of nonlinearity to the relationship between MR signal and lung water, but they are minimized as much as is practical here, given the constraints of large volume coverage in a short period of time.

Heterogeneity of magnetic susceptibility, as is prevalent in the lung, leads to signal extinguishing local gradients in the magnetic field. Indeed, in most typical MR scans, the normal lung parenchyma is not visible because susceptibility artifact extinguishes all signal from any water present. At some water volume, however, this degrading effect is partially overcome, and an increase in water volume causes an increase in signal. However, as water volume within a voxel continues to increase, the water eventually replaces air, thus removing the air-water interface and reducing the extinguishing effect of the susceptibility artifact. Therefore, when interstitial pressures increase, such that airspace is reduced, or when alveolar flooding commences, signal will increase much more rapidly as water volume increases. In the present experiments, because the extent of injury was less than that seen by others (40) and there was no tracheal foaming present, the water volumes are assumed to be within the range where susceptibility effects are still present. Even if susceptibility effects are reduced, the strong linearity between MR signal and water content found by others (12, 23, 27, 31) indicates the effect is probably small. Whereas the gradient-echo technique used in this work is more sensitive to the effects of heterogeneous magnetic susceptibility than are the spin-echo methods used in the referenced work, the short Te (4.7 ms) minimizes the sensitivity.

This assumption of signal linearity is important in calculating filtration flows. Because there is no direct standard for comparison, the calculations of _f  l cannot be quantitatively verified in these studies. The average change in EVLW (as estimated by MID) was 0.011 ml · min¹ · g DLW¹. However, the MID technique underestimates EVLW when regional edema diverts blood flow from an area. The average fluid accumulation in the lungs, calculated as postmortem WLW minus a baseline WLW obtained from the previous studies (31) was 0.0162 ± 0.0092 ml · min¹ · g DLW¹. In their experiments, Frostell et al. (15) found an average maximum change in EVLW of 0.085 ml · min¹ · kg body wt¹ during the first 2 h after oleic acid injury. In studies of alloxan injury, Klaesner et al. (25) measured lung water accumulation to be 0.027 ml · min¹ · g DLW¹ at baseline and 0.12 ml · min¹ · g DLW¹ after injury. The mean MRI-derived estimate of _f  l in this paper, normalized to DLW (0.015 ml · min¹ · g DLW¹ or 0.039 ml · min¹ · kg body wt¹), compares favorably with these measurements.

Taking the next step and calculating K_fc from the MRI measurement of _f  l involves making assumptions about values which can vary greatly. _l from the lungs, for example, can easily exceed 10 times the assumed value of 0.0978 ml/min. Furthermore, neither capillary nor interstitial pressures are expected to remain constant after oleic acid injury, yet this measurement of _{f  l} spans at least two or three image acquisitions, or 19-28 min, compared with only a few minutes in the gravimetric technique of calculating K_fc. Variability notwithstanding, the average value for K_fc (divided by lung mass) estimated in these experiments (0.025 ml · min¹ · cmH₂O¹ · g DLW¹ or 0.0040 ml · min¹ · cmH₂O¹ · g WLW¹) is certainly within the realm of those values found for noninjured lungs by Harris et al. (19) (means ranging from 0.0016 to 0.0126 ml · min¹ · cmH₂O¹ · g DLW¹) and for oleic acid-injured lungs by Townsley et al. (38) (0.0055 ml · min¹ · cmH₂O¹ · g wet wt¹). Moreover, despite the assumptions for _l, P, and , these estimates of K_fc correlated with another standard indicator of lung injury, namely, the D^1/2S ratio.

Because the MR scan shows regionally specific information about edema formation, this calculation could be extrapolated to calculate a regional _f  l mismatch and K_fc. To do so, a better measure of regional pressures and regional lymph clearance would be necessary. Furthermore, this would involve the assumption that fluid accumulation remains generally within the region of injury.

We conclude that the 3-D gradient-echo method described for imaging pulmonary edema with a common clinical MR scanner is sensitive enough to measure relatively small changes in lung water and that the method has enough temporal resolution to resolve the time course of the edema formation. From these noninvasive, nondestructive scans, regional information regarding pulmonary edema formation can be obtained. Also, information from multiple scans over time has the potential for being used in quantifying microvascular barrier integrity through transcapillary filtration flow (_f  Q_l) and the K_fc. The quantitative importance of the assumption of (P  ) on the estimation of K_fc remains to be studied.