In vivo brain edema classification: New insight offered by large b-value diffusion-weighted MR imaging

Authors

  • Attila Schwarcz MD, PhD,

    Corresponding author
    1. Department of Neurosurgery, University of Pécs, Pécs, Hungary
    2. Pécs Diagnostic Institute, Pécs, Hungary
    • Department of Neurosurgery, University of Pécs, Rét utca 2, H-7624 Pécs, Hungary
    Search for more papers by this author
  • Zsuzsa Ursprung MD,

    1. Pécs Diagnostic Institute, Pécs, Hungary
    Search for more papers by this author
  • Zoltan Berente PhD,

    1. Department of Biochemistry, University of Pécs, Pécs, Hungary
    Search for more papers by this author
  • Peter Bogner MD, PhD,

    1. Institute of Diagnostic Imaging and Radiation Oncology, University of Kaposvár, Kaposvár, Hungary
    Search for more papers by this author
  • Gyula Kotek MSc,

    1. Institute of Diagnostic Imaging and Radiation Oncology, University of Kaposvár, Kaposvár, Hungary
    Search for more papers by this author
  • Philippe Meric PhD,

    1. Laboratoire de Resonance Magnetique Nucleaire (RMN) Biologique, Institut de Chimie des Substances Naturelles (ICSN), Centre national de la recherche scientifique (CNRS), Gif-sur-Yvette, France
    Search for more papers by this author
  • Brigitte Gillet PhD,

    1. Laboratoire de Resonance Magnetique Nucleaire (RMN) Biologique, Institut de Chimie des Substances Naturelles (ICSN), Centre national de la recherche scientifique (CNRS), Gif-sur-Yvette, France
    Search for more papers by this author
  • Jean-Claude Beloeil PhD,

    1. Laboratoire de Resonance Magnetique Nucleaire (RMN) Biologique, Institut de Chimie des Substances Naturelles (ICSN), Centre national de la recherche scientifique (CNRS), Gif-sur-Yvette, France
    Search for more papers by this author
  • Tamás Dóczi MD, DSc

    1. Department of Neurosurgery, University of Pécs, Pécs, Hungary
    2. Pécs Diagnostic Institute, Pécs, Hungary
    3. Neurosurgical Research Group of the Hungarian Academy of Sciences, Pécs, Hungary
    Search for more papers by this author

Abstract

Purpose

To assess the role of large b-value diffusion weighted imaging (DWI) in the characterization of the physicochemical properties of the water in brain edema under experimental and clinical conditions.

Materials and Methods

Vasogenic brain edema was induced in mice by means of cold injury. A total of 17 patients with extensive peritumoral brain edema were also investigated. The longitudinal relaxation time (T1) and apparent diffusion coefficient (D) were measured in the edematous area both in humans and in mice. D was calculated by using both mono- (Dmono) and biexponential (Dfast and Dslow) approaches in the low and overall range of b-values, respectively. The D values were correlated with the T1 values.

Results

A strong linear correlation was found between T1 and Dmono in vasogenic brain edema, both in humans and in mice. After breakdown of Dmono into fast and slow diffusing components, only Dfast exhibited a strong correlation with T1; Dslow was unchanged in vasogenic brain edema.

Conclusion

Large b-value DWI can furnish a detailed characterization of vasogenic brain edema, and may provide a quantitative approach for the differentiation of edema types on the basis of the physicochemical properties of the water molecules. Application of the DWI method may permit prediction and follow-up of the effects of antiedematous therapy. J. Magn. Reson. Imaging 2007. © 2006 Wiley-Liss, Inc.

NEW APPROACHES involving diffusion weighted imaging (DWI) have emerged in everyday clinical practice, with the aims of a better understanding and visualization of diseases of the central nervous system. For instance, q-space imaging has provided new insight into multiple sclerosis (1), and detection of the multicomponent diffusion of brain water may allow prediction of the effects of radiation therapy in brain tumors (2).

As concerns brain edema, while the water content of the edematous tissue can be determined by means of longitudinal relaxation time (T1) measurements in vivo (3–5), there are no noninvasive methods for measurement of the volumes of the intra- and extracellular spaces under clinical conditions. Further, in certain cases a mixed type of brain edema occurs, and thus the differentiation of extra- and intracellular water can be a difficult clinical task. Accordingly, the traditional classification of brain edema proposed by Klatzo (6) (i.e., extra- and intracellular water accumulation) is of only minor relevance in clinical practice. The classification of edema should be revisited, with new standards of noninvasive, in vivo measurements of the osmotic/bound/viscosity properties of the tissue water. Some physicochemical properties of brain water can be measured quantitatively by large b-value DWI (7, 8) and the results would afford a more realistic characterization of the properties of tissue water than would a description of water accumulation according to its location (i.e., in the extra- or the intracellular space).

The in vivo characterization of brain edema could furnish a prediction or even measurement, for instance, of the effect of the antiedematous therapy applied according to the edema type. However, a number of questions arise concerning the physicochemical properties of tissue water in relation to the biexponential signal decay seen in large b-value DWI, and it is not straightforward to decide what water populations lie behind the two components derived from the biexponential analysis of DWI data (9, 10).

The aim of this study was to examine perifocal/peritumoral edema by means of large b-value DWI in order to gain more information on brain edema water properties detectable in vivo. The parameters yielded by the biexponential analysis of the signal decay in DWI were correlated with the respective T1 values. Cold injury was induced in mice in order to elicit vasogenic brain edema, and the results were compared with those obtained in humans with peritumoral brain edema.

MATERIALS AND METHODS

Animal Study

The experiments were carried out on five male mice (25–30 g) of the C57BL/6 strain. The mice were anesthetized via a face mask with 1.2% isoflurane in a 70/30 mixture of N2O and O2. The rectal temperature of the animals was maintained between 36.5 and 37.5°C throughout the experiments by using a water-filled warm blanket.

The cold injury was induced with a copper rod (diameter = 6 mm) cooled in liquid nitrogen and applied to the denuded parietal bone of the skull for 10 seconds. MR scans were performed 18–24 hours after the cold impact.

All NMR measurements were made on a Varian UNITYINOVA 400 spectrometer operating at 9.4 T. The signal was excited and received by means of a homemade surface coil (inner diameter [i.d.] 10 mm) and the self-shielded gradient system with 2000 mT m−1 field gradients. The diffusion was measured in only one direction (dorsoventral) in the brain, since the anisotropy in the gray matter is not prominent (11).

The imaging parameters of the DWI spin-echo sequence were as follows: TR = 2000 msec, TE = 32 msec, in-plane spatial resolution = 78 × 156 μm2, slice thickness = 1 mm, diffusion gradient strength = from 0 up to 400 mT m−1, duration of diffusion gradients (δ) = 8 msec, time between leading edges of diffusion gradients (Δ) = 1 msec, and number of averages = 1.

Diffusion signal analysis was carried out on 13 images recorded at different diffusion gradient strengths: b values = 0, 24, 299, 880, 1564, 2205, 2956, 3818, 4789, 6255, 7483, 8820, and 10,775 mm−2second.

The T1 values in mouse brain were determined in the same slice as chosen in DWI by using inversion prepared Turbo–fast low-angle shot (Turbo-FLASH) imaging (12) (TR = 2.05 msec, TE = 1.1 msec, 12 inversion times ranging from 66 to 10,000 msec, and flip angle = 3°).

Human Study

The investigation protocol in human subjects with extensive peritumoral brain edema of metastases of lung cancer (11 cases) or breast cancer (six cases) was approved by the local ethical committee. A total of 17 patients were involved in the study (seven females, 10 males; age 52 ± 4.2 years). Experiments were performed with a whole-body MR scanner (Siemens Magnetom Harmony) operating at 1 T. A Siemens Helmholtz head coil was used throughout the experiments. After localization, a brain slice containing extensive peritumoral edema without apparent tumor tissue was positioned parallel to the intercommissural plane. An inversion-prepared Turbo-FLASH sequence and a diffusion trace–weighted echo-planar imaging (EPI) sequence were applied to measure T1 and apparent diffusion coefficient (D) values, respectively. The parameters and a detailed description of the T1 measurement have been published elsewhere (4). The diffusion trace–weighted EPI parameters were as follows: TR = 3000 msec, TE = 135 msec, in-plane resolution = 1.9 × 1.9 mm2, slice thickness = 10 mm, b-values = 0, 500, 1000, 1500, 1800, 2100, 2400, 2700, 3000, and 3300 mm−2second, and number of averages = 4. The spatial parameters (field of view, matrix size, and slice thickness) of the Turbo-FLASH and EPI sequences were identical.

Data Analysis

Freehand regions of interest (ROIs) were used in the determination of the signal intensities. In mice, the whole edematous region apparent on diffusion-weighted images (Fig. 1a) with a decreased signal was used as the ROI. In humans, the geometrical distortion of the EPI images was taken into account subjectively by drawing the ROI carefully in comparison with FLASH images. ROIs were always drawn at the peripheral part of the edema.

Figure 1.

Figure shows diffusion weighted image with a b-value of 1712 mm−2second (a), Dfast (b), Dslow (c), and T1 (d) maps of the same slice obtained from cold injured mouse brain. The penumbra region (which appears dark on the diffusion weighted image) can be easily differentiated from the cold injured area (which appears white on the diffusion weighted image). The ROIs were placed in the penumbra and in the contralateral gray matter. The penumbra in the Dfast and T1 maps is lighter than the contralateral gray matter showing increased Dfast and T1 values, respectively. The penumbra in the Dslow map appears slightly lighter, but it is grossly identical to the contralateral gray matter (see Table 1).

D and volume fractions were calculated by fitting the signal intensities to the equation

equation image(1)

where I is the signal in the presence of diffusion sensitization, I0 is the signal in the absence of diffusion sensitization, Dfast and Dslow are apparent diffusion coefficient values, and ffast and fslow are the contributions to the signal of the fast and slow diffusing water compartments. To visualize the spatial D changes, D maps were generated. To compare Dfast and Dslow images to the standard D maps, signal intensities were also fitted in the low b-value range (up to 1500 mm−2second), using the monoexponential method to yield Dmono maps.

T1 maps of mouse and human brains were generated in the same slices as those in which DWI was performed. The water contents of the mouse and human brains were calculated from the T1 values according to the field strength and pathology-specific equations (3, 5). The values are expressed as means ± SD. Differences between data were examined with the two-tailed Student's t-test, with a P value < 0.05 considered to indicate a significant difference.

RESULTS

Animal Study

The perifocal edema was well differentiated from the cold injured area in the diffusion-weighted images obtained from the cold injured mouse brain (Fig. 1). From the images with large b-values, the diffusion appeared to be slowed down in the necrotic core, while larger diffusion constants were measured in the perifocal edema (Fig. 1a). Examples of the respective diffusion maps and the T1 map generated from an examined brain slice are also shown in Fig. 1. In the perifocal edema, both Dmono and T1 were increased significantly and a strong linear correlation was found between them, with a correlation coefficient of r = 0.79 (Fig. 2). When the b-value range was extended and the diffusion constant was split into Dfast and Dslow, a significant increase in Dfast could be observed, whereas Dslow did not display any significant alteration (Table 1). Dfast exhibited a strong linear correlation with T1 (r = 0.69). Similarly to Dslow, the volume fraction of the fast-diffusing component did not correlate with T1. The volume fraction of the fast component demonstrated a tendency to increase; however, the change did not reach the level of statistical significance (Table 1). Table 1 also shows the water content values calculated from T1 to indicate the change in water concentration in perifocal edema in the mouse brain (78.1–79.8%).

Figure 2.

Plot of Dmono vs. T1 values determined in the mouse (triangles) or the human study (squares). A strong linear correlation can be observed in each data set.

Table 1. Means and SD of Apparent Diffusion Coefficients (D), Volume Fractions (pfast) of Fast Diffusing Component, Longitudinal Relaxation Time (T1), and Water Content (W) Measured in Humans and Mice
 WT1 (msec)Dmono (×10−4 mm2second−1)Dfast (×10−4 mm2second−1)Dslow (×10−4 mm2second−1)pfast (%)
  • Correlation coefficients (r) are also displayed: T1 vs. Dmono, Dfast, Dslow, or pfast.

  • W in % of dry weight/wet weight.

  • *

    P < 0.05 (unpaired t-test vs. control values).

  • **

    P < 0.005 (unpaired t-test vs. control values).

Human      
 Intact white matter68.9 ± 0.9550 ± 217.12 ± 0.7512.7 ± 2.182.42 ± 0.6763 ± 12
 Peritumoral edema82.3 ± 3.8**1038 ± 221**12.9 ± 2.54**16.1 ± 3.5*2.3 ± 0.8588.8 ± 5**
   (r = 0.91)(r = 0.71)(r = 0.02)(r = 0.7)
Mouse      
 Intact cortex78.1 ± 0.561477 ± 655.9 ± 0.228.1 ± 0.352.03 ± 0.2275 ± 4
 Perifocal edema79.8 ± 0.8**1689 ± 118**7.74 ± 0.5**10.7 ± 1.6**2.02 ± 0.779 ± 7
   (r = 0.79)(r = 0.69)(r = 0.2)(r = 0.3)

Human Study

The data on six of the 17 examined patients were omitted due to the low signal-to-noise ratio (SNR) (i.e., SNR < 3) obtained in images with large b-values. Figure 3 shows the respective T1, Dmono, Dfast, and Dslow maps of an examined patient. The Dmono maps closely resembled the Dfast maps, but distinct differences could be observed at the edge of the edematous area. In the peritumoral edema, the Dmono and T1 values were increased relative to those in the contralateral normal white matter. Dfast and the volume fraction of the fast component were also increased in the edematous region, whereas Dslow was not changed significantly (Table 1). Table 1 also presents the mean water content values calculated from T1 to indicate the mean water concentration change in the edematous area (68.9–82.3%). A strong linear correlation was found between T1 and Dmono, with a correlation coefficient of r = 0.91 (Fig. 2). When the diffusion measurement was extended to larger b-values, T1 displayed strong linear correlations with Dfast (r = 0.71) and with the volume fraction of the fast component (r = 0.7). However, similar to the mouse study, no correlation was observed between T1 and Dslow. The measured mean values and the results of the t-tests are displayed in Table 1.

Figure 3.

Dmono with a b-value up to 1500 mm−2second (a), Dfast (b), Dslow (c), and T1 (d) maps of the same slice. Images were obtained from a patient with extensive peritumoral edema. The huge mass effect of edema resulted in the disappearance of the extracerebral liquor spaces on the affected side. The regions of interests were placed in the edematous region and in the contralateral white matter. The edema in the Dmono, Dfast, and T1 maps appears lighter than the contralateral white matter, showing increased Dmono, Dfast, and T1 values, respectively. However, no difference can be observed between the edematous region and the contralateral white matter in the Dslow map.

DISCUSSION

The noninvasive character of MRI allows in vivo investigations of brain pathologies such as brain edema that can lead to brain herniation and secondary complications in severe cases. The method of large b-value DWI attracted huge attention and evoked major excitement in clinical neuroscience when it emerged that the separation of extra- and intracellular compartments had become feasible in cell cultures (13, 14). However, various in vivo and in vitro studies later demonstrated that the biexponential signal decay seen in large b-value DWI did not correspond to the extra- and intracellular compartments (8, 15–17), but rather reflected water molecules in different physicochemical states (7, 8). The main physicochemical factors that can play significant roles in the biexponential nature of the signal decay seen in DWI would be macromolecular water binding, molecular crowding, microviscosity, and compartmentation, including microcompartments within the cells.

It may be hypothesized that measurement of the physicochemical characteristics of edema water is of more clinical significance than estimation of the ratio of the extra- and intracellular spaces, and it may be possible to test new antiedematous drugs such as aggressive HyperHaes solutions (18) or aquaporine antagonists (19). In our study we examined the situation in which D is broken down into Dfast and Dslow with the use of large b-value DWI.

The basic MR observations on the cold injury model in mice were published earlier (8, 20, 21). However, in the present experiments it seemed obvious that, besides the well-characterized cold-injured area (necrotic core), there was also a penumbra to which the brain edema fluid propagated. In contrast with the cold-injured area with membranes disintegrated by freezing (8), the penumbra consists of gray matter with the extracellular space flooded by edema fluid. In this penumbra region, T1 displayed a strong correlation with Dmono, similar to a study on vasogenic brain edema elicited in cats (22). After the breakdown of D into Dfast and Dslow, a linear correlation was observed between T1 and Dfast, whereas no correlation was detected between T1 and Dslow. This can be explained in that the edema fluid affected only the diffusion of the fast component, and not that of the slow component. The volume fraction of the fast component in brain edema exhibited an increased level as compared with the normal gray matter, though the change was not statistically significant (Table 1).

It can be concluded from the mouse experiments that the vasogenic edema elicited by the cold injury affected only Dfast, which demonstrated a strong correlation with T1.

In the human study, the production of the biexponential signal decay in DWI was rather challenging since the usual clinical scanners have a relatively low magnetic field with inherent low SNR, in contrast with MR instruments for experimental studies. However, it appears feasible to also obtain the biexponential signal decay in DWI at low magnetic field (2). It should be noted that the data from six patients were omitted since no biexponential decay was observed in DWI due to the low SNR. However, despite the low SNR, our trace-weighted D data correspond well with those obtained at higher magnetic fields with other methods (23–25). The trace weighting was crucial to free our data from the anisotropy effect occurring in the white matter (11). Despite the considerable partial volume effect (i.e., thick slices and freehand ROIs) Dmono correlated strongly with T1 in the peritumoral edema and obviously with the water content calculated from T1. The measured T1 and water content values in the intact white matter are in good agreement with the data in the literature (3).

It is intriguing that the correlation coefficient (r = 0.92) and correlation equation for Dmono vs. the water content, y = 43.93x – 2318, were basically identical to those calculated in Ref.22 in a cat study, where vasogenic edema was elicited by cold injury in the white matter. This suggests that the correlation between Dmono and the tissue water content may be fairly constant in this type of pathology. A strong correlation between Dmono and T1 was also observed in the brain maturation of cats (26).

The D values obtained with high diffusion weighting were broken down into Dfast and Dslow by means of the biexponential approach. Similarly as in the mouse study, Dfast correlated with T1, whereas no correlation was observed between Dslow and T1. The T1 values for the two water populations with Dfast and Dslow were found, using a hybrid, diffusion-weighted, inversion-recovery sequence, to be identical (27). We presume that the increase in T1 takes place in the water population with Dfast during vasogenic edema, though this point remains to be elucidated.

As regards the volume fractions, the fast component also correlated strongly with T1 in the human study. Such a correlation could not be confirmed in the mouse study, probably because of the smaller increase in water content: calculations in the mouse study (5) indicated a water content increase from 78.1% to 79.8%. This is a significantly smaller change than that detected in the peritumoral edema in human subjects (68.9–82.3%). The smaller water-content increase in the gray matter could also be a reason for the weaker correlation between Dmono and T1 in the mouse study. It appears that the small water content elevation in the penumbra is an inherent limitation of the cold injury model in mice since the peak of edema occurs 24 hours after the cold impact (28). The limitation of the cold injury model is one of the explanations for the relatively small number of animals used in this study. There was no reason to sacrifice more animals, since no further water content or T1 elevation can be achieved with this edema model.

Our study confirms that, in brain edema, the water can be broken down into two distinct populations that differ in their physicochemical properties. The neuropathological (histomorphological) differentiation of brain edema, i.e., extra- and intracellular edema, is of limited practical value in clinical work, because of the lack of in vivo measuring methods. Moreover, in some pathological cases a mixed type of brain edema can occur, even in one MR voxel. Thus, the differentiation of vasogenic and cellular edema in these cases can not be based on Dmono values as has been suggested (29).

The biexponential approach in DWI might predict, for instance, the effectiveness of osmotic therapy by measuring the water populations with different physicochemical properties. It was pointed out in an erythrocyte study that the state of binding of the water to proteins determines whether it can be removed from the cell through the application of mannitol dehydration (30).

In conclusion, our study suggests that vasogenic edema affects Dfast and the volume fraction of the fast component, and Dslow is preserved. In contrast, cellular cytotoxic edema (i.e., ischemic tissue) is characterized by an increased Dslow, while Dfast is unchanged (24).

Acknowledgements

We thank Dr. Jean-Loup Correze and to Dr. Joël Mispelter from Institut Nationale de la Santé et de la Recherche Médicale (INSERM) (U350) for skillful assistance in the construction of the microimaging probe used in the animal study.

Ancillary