Chemical shift sodium imaging in a mouse model of thromboembolic stroke at 9.4 T




To estimate changes in the 23Na density and in the 23Na relaxation time T2* in the anatomically small murine brain after stroke.

Materials and Methods:

Three-dimensional acquisition weighted chemical shift imaging at a resolution of 0.6 × 0.6 × 1.2 mm3 was used for sodium imaging and relaxation parameter mapping. In vivo measurements of the mouse brain (n = 4) were performed 24 hours after stroke, induced by microinjection of purified murine thrombin into the right middle cerebral artery. The measurement time was 14 minutes in one mouse and 65 minutes in the other three. An exponential fit estimation of the free induction decay was calculated for each voxel enabling the reconstruction of locally resolved relaxation parameter maps.


The infarcted areas showed an increase in sodium density between 160% and 250%, while the T2* relaxation time increased by 5%–72% compared to unaffected contralateral brain tissue.


23Na chemical shift imaging at a resolution of 0.6 × 0.6 × 1.2 mm3 enabled sodium imaging of the anatomical small mouse brain and the acquired data allowed calculating relaxation parameter maps and hence a more exact evaluation of sodium signal changes after stroke. J. Magn. Reson. Imaging 2011;. © 2011 Wiley-Liss, Inc.

APART FROM CLASSICAL 1H magnetic resonance imaging (MRI), the use of 23Na imaging in animal stroke models is of growing interest (1–4) because of its availability for use as an intrinsic marker for brain integrity. The main interest thereby lies in the distinction between viable and nonviable tissue involved in stroke. The reversibility of the apparent diffusion coefficient (5) complicates the validity of diffusion-weighted MRI regarding this task. In addition, perfusion-weighted MRI cannot answer this question alone, since the loss of cell viability depends on both the degree of hypoperfusion and its duration, which is often unknown. Tissue sodium homeostasis is a fundamental property of viable cells (6) leading to an intracellular sodium concentration of about 10 mM/L and an extracellular concentration of 140 mM/L. An increase in tissue sodium concentration (TSC) from ≈40 mM/L up to more than 70 mM/L after loss of membrane integrity is an established marker for dead tissue (1) that makes the quantification of TSC (7) a promising approach for localized measurements of cell viability. Furthermore, it has been shown in animal models of stroke that the shapes of elevated 23Na-signal and of histological TTC-defined infarction are very similar and that an accumulation of 23Na is an unambiguous marker for dead tissue (3). Recently, 23Na MRI has been suggested as a complementary technique for ischemic stroke characterization (8).

In this study a novel mouse model of in situ thromboembolic stroke (9) was used in which the beneficial effects of recombinant tissue plasminogen activator (rt-Pa) induced thrombolysis were observed. This approach contains nonminor risks and therefore new treatment strategies in combination with adequate imaging techniques are necessary. The small size of the mouse brain complicates the application of many MRI techniques. In particular, sodium MRI with a 12,000 times lower signal-to-noise ratio (SNR) compared to proton imaging suffers from the required small voxel sizes for imaging of the mouse brain at adequate spatial resolution. Furthermore, the short relaxation times of the 23Na nucleus preclude standard MRI acquisition techniques. Thus, radial projection imaging techniques (10–12) are commonly used for sodium MRI. Although data acquisition thereby starts in the center of k-space, the echo time, TE, is still not negligible and an increase of the sodium signal may not only be explained by an increase in TSC, but also by prolonged T2*. There are approaches to measure local T2* values based on a gradient echo sequence (13, 14). These methods are applicable for sodium long component T2* mapping but the influence of the short T2* component (15) cannot be taken into account with the long echo times achieved with these techniques. While the short relaxation time T2* of sodium still remains a limiting factor, the short relaxation time T1 allows the use of very short repetition times TR enabling 3D sodium chemical shift imaging (23Na-CSI) which uses phase encoding in all spatial dimensions and therefore requires at least one repetition for each voxel to be measured. The 23Na-CSI sequence acquires the whole free induction decay (FID) for each phase encoding step, allowing for the reconstruction of relaxation parameter maps.

The aim of this work was to test the feasibility of this technique for estimating the sodium concentration increase and the change in the transversal relaxation time T2* of 23Na after stroke.


All experiments were performed on a 9.4 T Biospec 94/20 USR (Bruker, Germany) small animal system equipped with 740 mT/m gradients. A two-winding (12 and 20 mm i.d.) 105 MHz surface resonator element was developed for sodium imaging with variable tuning (0.5 to 6 pF, Voltronics NMQM6GE) in order to maximize SNR in the sodium channel. Inductive coupling was achieved via a longitudinally displaceable coupling loop (18 mm i.d.), which was mounted 10 mm above the surface coil. The 23Na transceiver surface coil was used in conjunction with a 1H linear birdcage resonator (72 mm i.d.) to acquire anatomical 1H images for slice positioning and shimming.

23Na-CSI measurements were performed with Hanning-weighted k-space acquisition (16) to reduce blurring and ringing and to increase SNR, instead of data acquisition with constant sampling density, which leads to a convolution of the measured data with a sinc function. The k-space data were acquired as presented by Pohmann and von Kienlin (17) leading to significantly increased SNR due to reduced negative side lobes at equal spatial resolution and measurement time compared to a CSI measurement with constant sampling density and the same number of repetitions. The matrix size was increased to keep the same resolution as in the non-weighted case. For all presented measurements, the acquired matrix size is given together with a matrix size corresponding to a nonweighted CSI experiment with equal spatial resolution.

Data analysis was performed with MatLab R2007b (MathWorks, Natick, MA).

Phantom Measurement

The phantom consisted of two vials filled with saline solution (0.9%). The left vial (compare Fig. 1) contained an additional 5% agarose to shorten T2*. A rectangular pulse of 50 μs duration was used. Together with the phase encoding duration, this leads to an acquisition delay of 365 μs for the weighted 3D-23Na-CSI measurement. A matrix size of 37 × 37 × 37 was used corresponding to a matrix size of 24 × 24 × 24 in a nonweighted acquisition experiment. Other sequence parameters were set to: TR = 60 msec, field of view (FOV) = 24 × 24 × 48 mm3, flip angle = 50°. A total of 65,536 repetitions led to a total scan time of 65 minutes. The FID of each phase encoding step was sampled with 580 spectroscopic data points during 58 msec after excitation. A fast Fourier transformation was applied to the k-space data of each of the 580 timepoints. Images were calculated by integrating the FIDs of each voxel over 5-msec time ranges, as shown in Figure 1. Additionally, the intensity time courses (i.e., all data points of the FID) in all reconstructed voxels were fitted to an exponential decay function of the form equation image estimated by the least squares fit to obtain parametric T2* and S0 maps, where S0 = S(t = 0) − const.

Figure 1.

23Na-CSI phantom measurements of a vial with pure saline solution (right) and with additional 5% agarose (left). The image received by integration of the first 5 msec of the FID mainly shows the intensity profile of the surface coil. Later integrations show a different contrast for the two vials caused by different T2* relaxation times.

Mouse Measurements

All experiments were approved by the local Ethical Committee in accordance with the animal protection guidelines. A thromboembolic stroke was induced in four male C5 black/6J mice (25 to 30 g, Charles River, Sulzfeld, Germany) by local injection of purified thrombin directly into the right MCA (9). The animals were anesthetized with 1.5%–2% Isoflurane (Abbott, Wiesbaden, Germany) in a mixture of oxygen/air (1:1) administered via a face mask and placed in a stereotactic frame. The skin between the right eye and the right ear was incised and the temporal muscle was retracted. To expose the MCA, a small craniotomy was performed and the dura mater was excised. A micropipette was introduced into the lumen of the MCA and 2 μL (1.5 U) of purified murine alpha-thrombin (Haematologic Technologies, Essex Junction, VT) were injected to induce the formation of a clot. Ten minutes after injection the clot has stabilized and the micropipette was removed. After surgery the incision wound was sutured. Twenty-four hours after MCA occlusion the animals were anesthetized with 1%–2% isoflurane and positioned into the magnet with a laser controlled system for the animal cradles. Movement of the head was restricted by a three-point fixation (ears, bite bar). Respiratory frequency and body temperature were monitored throughout the experiment and the latter was maintained with a water heating pad.

Anatomical proton images were acquired with a 1H-TX/RX surface cryoprobe (18) (Bruker) in a preliminary experiment to determine the position and size of the infarction. A T2-weighted RARE sequence was applied with the following parameters: TR = 2.5 sec, TE = 60 msec; echo train length = 4, 4 averages, matrix size = 384 × 384, FOV = 17 × 17 mm2, slice thickness = 0.4 mm, measurement time (12 slices) = 6 minutes 40 sec. Planimetric measurements of MR images (ImageJ software, National Institutes of Health, Bethesda, MD) were performed and used to calculate lesion volumes.

The 1H cryoprobe was removed after anatomical proton imaging and the 23Na surface resonator was inserted in conjunction with the 1H linear birdcage resonator. The position of the mouse could not be maintained during the change of the coil setup; therefore, the slices of proton and sodium scans do not exactly coincide.

Sodium images of the mice were acquired with two different setups. For one mouse a fast 23Na-CSI measurement was performed with 32,768 repetitions with a TR of 25 msec leading to a total measurement time of 13 minutes 39 seconds. All other mice were scanned with a longer protocol resulting in higher SNR and therefore more accurate S0 and T2* maps, achieved by doubling the number of repetitions to 65,536 (two averages) with a longer TR of 60 msec. Both factors lead to a total acquisition time of 65 minutes. One hundred fifty data points were acquired in 30 msec (100 points in 20 msec for the faster scan). The first 30 data points (6 msec) of each FID were integrated for the maximum SNR image due to the faster in vivo relaxation time compared to the phantom measurement. S0 and T2* maps were calculated as described in the phantom measurement section. A matrix size of 47 × 47 × 37 was used in the long scan and 39 × 39 × 33 in the short scan. Both acquisition schemes correspond to a 32 × 32 × 32 nonweighted acquisition matrix. The FOV was set to 19.2 × 19.2 × 38.4 mm3 leading to a spatial resolution of 0.6 × 0.6 × 1.2 mm3.

To estimate the changes in the sodium density and in the transverse relaxation time T2*, two regions of interest (ROIs) were manually defined in five maximum SNR slices, in which the stroke was clearly noticeable. One ROI contained the hyperintense area of stroke, the second ROI was put in the contralateral hemisphere, equally shaped and at the same distance from the surface coil. Figure 2 (left) exemplarily shows two used ROIs. For both ROIs the mean signal intensity was calculated at each sampled timepoint of the FID and the monoexponential fit estimation was performed for these values, as described in the phantom measurement part. The means of T2* and S0 in the infarcted area were calculated for both ROIs.

Figure 2.

ROIs and corresponding monexponential least square fits exemplarily shown in one slice of mouse 2 (compare Table 1). There is no evident discrepancy noticeable between the fit estimation and the acquired data. The resulting parameters of the least square fit are S0 = 2.26 ± 0.03 and T2* = 7.5 ± 0.3 msec in the lesion and accordingly S0 = 0.59 ± 0.03 and T2* = 6.0 ± 0.5 msec in the contralateral brain hemisphere. [Color figure can be viewed in the online issue, which is available at]


Phantom Measurement

No significant signal difference between the two vials is detectable in the image received by integration of the first 5 msec after excitation (Fig. 1). T2*-weighting increases with the acquisition delay time. In Figure 3a a maximum SNR image is presented, reconstructed by integration of the FID's first 20 msec. Figure 3b and c illustrates parameter maps of S0 and T2*, representing the fit results. The S0 map shows the intensity profile of the surface coil and the T2* parameter map demonstrates the difference of the transverse relaxation time caused by the agarose. T2* is 33 ± 1 msec in pure saline solution and 12 ± 1 msec with additional 5% agarose.

Figure 3.

a: Maximum SNR image of the phantom, reconstructed by integration of the first 20 msec of the FIDs. b: S0 map resulting from the least square exponential fit. Signal inhomogeneities represent the surface coil sensitivity profile. c: T2* parameter map resulting from the least square exponential decay fit. The T2* relaxation time in pure saline solution measured 33 ± 1 msec and 12 ± 1 msec with 5% agarose concentration.

Mouse Measurements

T2-weighted anatomical slices of the mouse brain acquired with the 1H cryoprobe are shown in Figure 4. The mouse, shown on the left, underwent fast 23Na-CSI scanning. The image on the right shows one mouse that underwent the long scan 23Na-CSI experiments (mouse 2 in Table 1). The infarcted region clearly appears hyperintense and is equally shaped as less stained area in hematoxylin-eosin stained coronal brain slices (not shown). Lesion sizes of the individual mice are given in Table 1. The relative error of the lesion volume was estimated to 1% by repeating the planimetric measurement of the lesion size for one mouse five times.

Figure 4.

In the T2-weighted proton images acquired with the cryoprobe 24 hours after MCA infarcted regions appear hyperintense. The left mouse was used for the short 23Na-CSI experiment, the right mouse for the longer 23Na-CSI experiment.

Table 1. Parameters Resulting from the Monoexponential Least Square Fit of the 23Na-CSI Data with Confidence Bounds (P = 0.05) and Infarct Size, Determined from the T2-Weighted Proton Scans
 ROI in infarcted tissueROI in contralateral tissue 
Measurement23Na-T2* [msec]23Na-S0 [a.u.]23Na-T2* [msec]23Na-S0 [a.u.]Infarct size [mm3]
Mouse 1, fast scan7.9 ± 0.50.95 ± 0.054.6 ± 0.30.27 ± 0.0226.9 ± 0,3
Mouse 2, long scan8.3 ± 0.42.28 ± 0.038.0 ± 10.67 ± 0.0655.0 ± 0,6
Mouse 3, long scan5.4 ± 0.31.31 ± 0.044.8 ± 0.30.37 ± 0.0117.2 ± 0.2
Mouse 4, long scan9.0 ± 0.71.08 ± 0.087.1 ± 0.50.41 ± 0.047.2 ± 0.1

In the maximum SNR image of the same two mice, reconstructed of the first 6 msec of the 23Na-CSI scan (Fig. 5), the infarcted area clearly appears hyperintense in both measurements. Although most relevant details are noticeable in the 14-minute scan, the 1-hour scan reveals better image contrast and better SNR. The central slice of Figure 5 of both mice was used for further data analysis to calculate S0 and T2* parameter maps. The S0 map (Fig. 6) reveals a strong signal increase in the infarcted region indicating an increase in TSC. The T2* maps in Figure 6 show that the sodium relaxation parameter T2* ranges from about 2 to 10 msec in the mouse brain. There is a slight T2*-increase in the infarcted hemisphere but the infarcted area does not significantly stand out against homotopic tissue. The ROI-based data analysis is more precise because the mean of all data points in the ROI was used. Figure 2 exemplarily shows the used ROIs and corresponding monexponential least square fits in one slice of mouse 2. There is no evident discrepancy noticeable between the fit estimation and the acquired data. The results of all mice are summarized in Table 1. The sodium concentration was significantly increased in the stroke area compared to the cortex in the contralateral brain hemisphere (tested with a paired Student's t-test, P < 0.005). The T2* in contralateral brain tissue varies from 4.6 to 8.0 msec and is increased in the stroke area of all mice. However, the increase in T2* of 5%–70% is much smaller than that in S0.

Figure 5.

Maximum SNR image, reconstructed by integration of the first 6 msec of the FIDs. (a) Result of the fast CSI scan, (b) of the longer scan.

Figure 6.

S0, and T2* parameter maps, reconstructed from the short 23Na-CSI data (upper row) and from the longer experiment (lower row).


The phantom measurements show that 23Na-CSI allows exponential fitting of an acquired time series of images within normal 23Na imaging times, while high SNR images can be calculated from the same dataset. Combination of the high field of 9.4 T, the inductive coupled surface coil, and the weighted 23Na-CSI sequence enable imaging at a spatial resolution of 0.6 × 0.6 × 1.2 mm3 in a measurement time of 13 minutes 40 seconds. For relaxation parameter mapping, however, a measurement time of 65 minutes is suggested to achieve more accurate values of S0 and T2*. To the best of our knowledge, these are the first sodium MRI measurements applied to a murine stroke model and the highest resolution in sodium imaging below 21 T reported so far. Furthermore, this is the first work describing changes of the 23Na transverse relaxation time T2* after stroke. The in vivo measurements of mice 1 to 3 show an ≈3.5-fold sodium density increase, which is the expected maximum possible increase from normal brain tissue concentration of 40 mM to the extracellular concentration of 140 mM (1–3). This indicates that the cell integrity is totally lost in the lesion. The slightly smaller 2.6-fold increase in mouse 4, which has a much smaller lesion volume, may be explained by partial volume effects. Lin et al. (3) found that T1 in infarcted brain tissue is unchanged, which makes an overestimation of the sodium density due to a decrease of the longitudinal relaxation time T1 unlikely. The same authors mentioned that changes in MR relaxation appear more likely to account for a decrease in the 23Na signal intensity. Our presented measurements indicate that T2* increases in infarcted tissue, which leads to a 23Na signal increase.

Bartha and Menon (14) measured sodium long component T2* values of 18–22 msec in the human brain at 4 T. The T2* values of the mouse brain measured in this work are, at 4.8–8 msec, much smaller. The whole acquired FID was sampled in 100 μs time steps starting 365 μs after excitation, thus the presented T2*-values depend on both short and long T2* components leading to more accurate values compared to techniques that only focus on one of the two components. Although T2 decay theoretically is biexponential, the case is more complicated for T2* relaxation, which is also influenced by field inhomogeneities, chemical shifts, and the presence of field gradients. However, the monoexponential fit function satisfactorily fits the measured data (compare Fig. 2), but for those reasons a comparison between human and murine T2* values, achieved with different measurement systems, is problematic.

The CSI sequence does not use a readout gradient during data acquisition. Thus, the presented technique is not prone to blurring and ringing artifacts. The more than 3-fold signal increase in stroke detected 24 hours after the occlusion of the right MCA makes sodium imaging very interesting as an in vivo marker for infarcted tissue. In future, more animals have to be measured at different timepoints after the occlusion, especially for the investigation of the change in the relaxation times T2* in stroke. Furthermore, the 23Na-CSI sequence allows the use of chemical shift reagents (19, 20) to distinguish between intra- and extracellular sodium signals, which are very important parameters and would benefit the application of sodium MRI in animal stroke models. The described technique is suitable for the use of shift reagents at the presented high resolution and should be able to strongly improve the investigation of sodium MRI in animal stroke models.

In conclusion, this work describes a first feasibility study of the CSI sequence to detect changes in the sodium MR signal after stroke. The achieved resolution of 0.6 × 0.6 × 1.2 mm3 allows for sodium MRI in the anatomically small mouse brain. Furthermore, this method permits estimation of the transverse relaxation time T2*, which is important for quantitative measurements of the sodium concentration. Sodium T2* measured in infarcted tissue was higher than in homotopic brain tissue. If not taken into account, T2*-relaxation changes after stroke more likely account for an overestimation of the tissue sodium concentration.


We thank Marén Neumann and Felix Hörner for support.