Reduction of Diffusion-Weighted Imaging Contrast of Acute Ischemic Stroke at Short Diffusion Times
Background and Purpose—Diffusion-weighted imaging (DWI) of tissue water is a sensitive and specific indicator of acute brain ischemia, where reductions of the diffusion of tissue water are observed acutely in the stroke lesion core. Although these diffusion changes have been long attributed to cell swelling, the precise nature of the biophysical mechanisms remains uncertain.
Methods—The potential cause of diffusion reductions after stroke was investigated using an advanced DWI technique, oscillating gradient spin-echo DWI, that enables much shorter diffusion times and can improve specificity for alterations of structure at the micron level.
Results—Diffusion measurements in the white matter lesions of patients with acute ischemic stroke were reduced by only 8% using oscillating gradient spin-echo DWI, in contrast to a 37% decrease using standard DWI. Neurite beading has recently been proposed as a mechanism for the diffusion changes after ischemic stroke with some ex vivo evidence. To explore whether beading could cause such differential results, simulations of beaded cylinders and axonal swelling were performed, yielding good agreement with experiment.
Conclusions—Short diffusion times result in dramatically reduced diffusion contrast of human stroke. Simulations implicate a combination of neuronal beading and axonal swelling as the key structural changes leading to the reduced apparent diffusion coefficient after stroke.
Diffusion-weighted magnetic resonance imaging (MRI; DWI) is the gold standard for the sensitive detection and diagnosis of acute ischemic stroke. The hyperintensity identifying the lesion on DWI is because of the reduction of the mean apparent diffusion coefficient (also known as mean diffusivity, MD) of water within minutes of ischemia.1 Although the MD reductions after stroke were discovered 25 years ago and DWI is used daily for diagnosis of stroke, the underlying biophysical mechanisms have remained elusive. Upon ischemia, glucose and oxygen deprivation results in the failure of Na+/K+ ATPase ion pumps in cell membranes, leading to an osmotic shift of water and swelling (ie, cytotoxic edema). Accordingly, the long-standing hypothesis is that this shift of water from extracellular space to a more restrictive intracellular space is the underlying mechanism for the MD changes. However, the models based on swelling alone cannot account for the magnitude of the change (≈50%) nor do they explain that the MD of intracellular-only metabolites decrease after ischemia, as determined from diffusion-weighted MR spectroscopy.2,3 Recent simulations and preclinical models implicate that cytotoxic edema causes the formation of enlargements and constrictions in neuronal membranes (ie, beading), which introduces barriers along neural fibers that restrict water mobility and may account for the MD decreases observed in acute stroke (Figure 1).4 Beading is a generalized property of axons and dendrites that occurs after injury, such as depolarization or stress.5,6 This beading hypothesis has yet to be evaluated in human acute ischemic stroke given the technical limitations of routine diffusion MRI.
The beading in white matter (WM) occurs with length scales on the order of the axon diameters, and adjusting the length scale sensitivity of the diffusion MRI experiment may shed light on changes that occur on stroke. This has only recently become possible with the implementation of oscillating gradient spin-echo (OGSE) diffusion MRI on clinical MRI scanners,7,8 although it was initially proposed many years earlier on animal MRI scanners.9 This OGSE method allows different microstructure length scales to be probed by varying the diffusion time, Δeff (Figure 1). If most molecules do not travel far enough to interact with any obstacles during the diffusion time, the MD is the intrinsic diffusion coefficient of cellular water.10 For increasing diffusion times, the molecules interact with more barriers and the MD decreases toward an asymptotic value.11,12 In vivo human diffusion MRI traditionally uses a pulsed gradient spin-echo (PGSE) technique13 that is in the asymptotic long diffusion time regime (free water diffusion distances ≈20 μm versus cell size ≈1 μm). However, the OGSE diffusion MRI technique has enabled diffusion times low enough to break out of the asymptotic regime and probe short-length scales near cellular dimensions in rodents (diffusion distances ≈2 μm).9 In a rat model of global ischemia, the MD reduction was much less (about half) for shorter diffusion times of 0.5 ms (versus 9.8 ms), suggesting that the MD decreases during ischemia are because of structural changes and not permeability or viscosity changes.14,15 More recently, a model of unilateral ischemia in the cortices and hippocampi of mice showed decreased diffusion contrast of ischemia for OGSE frequencies larger than 100 Hz (diffusion time <2.5 ms).16 However, OGSE has not yet been applied to human stroke nor any other clinical disorder. This work evaluates the diffusion parameters in lesions of 11 patients with acute stroke using short (4 ms with OGSE) and typical (40 ms with PGSE) diffusion times for diffusion tensor MRI (DTI), and tests whether these changes are consistent with cellular beading using Monte-Carlo simulations of diffusion in beaded cylinders that would reflect the hypothesized biophysical alterations in ischemic axons.
Participants included 11 male patients diagnosed with ischemic stroke, who were recruited with informed consent (approved by the Health Research Ethics Board) from the University of Alberta Hospital stroke unit (Table). Other recruitment criteria were that ischemic stroke onset was within 5 days and there were no contraindications to high field MRI at 4.7 T. Lesions in WM, deep gray matter (GM), and cortical GM were outlined with separate regions of interest, leading to the analysis of a total of 20 lesions (9 WM; 6 cortical GM, 5 deep GM).
MRI acquisition of the brain was performed in 1 scan session on a Varian Unity Inova 4.7 T. A b=1000 s/mm2 (b1000) DWI protocol with 3 orthogonal diffusion encoding directions was used to locate the lesion(s). The single-shot echo planar imaging parameters for this b1000 protocol were: repetition time=10 s; echo time=60 ms; field-of-view=24×19 cm2; 1.5×1.5 mm acquired in-plane resolution (zero-filled to 0.75 mm×0.75 mm resolution); 80 slices, thickness 1.5 mm (no gap); 4 averages; R=2 GRAPPA; scan time 3.5 minutes. Two DTI protocols were acquired using Δeff=4.1 ms (OGSE, 50 Hz) and 40 ms (PGSE), with diffusion gradients, sequence timings, and data quality the same as our recent investigations in healthy subjects.8 Both protocols acquired a 5-cm slab centered on the lesion and used b=300 s/mm2 with 6 diffusion-encoding directions and a maximum gradient amplitude of 57.5 mT/m per channel. Although diffusion imaging with this low b-value is not desirable for general clinical application for the identification of lesions, the hypothesis of this work is that it will shed light on the causes of MD changes after stroke. Other single-shot echo planar imaging parameters for the b300 protocols were: repetition time=12.5 s; echo time=110 ms; field-of-view=24×24 cm2; 2 mm×2 mm acquired in-plane resolution (zero-filled to 1 mm×1 mm resolution); 20 slices, thickness 2.5 mm (no gap); 6 averages; R=2 GRAPPA; scan time 5 minutes per protocol. Confounding effects from eddy currents were mitigated using a combination of gradient precompensation and postacquisition phase correction, as in previous work.8
For each individual, motion between the OGSE and PGSE scans was corrected using rigid body translations and rotations17 with an autocorrelation cost function. The b1000 DWI scan was resampled to be the same resolution as the OGSE and PGSE scans, and aligned to them using the same rigid body motion correction algorithm. The lesions were outlined on the b1000 DWI on all slices with the aid of a semiautomated tool that defines the boundary of the regions of interest where the DWI signal intensity is halfway between a value manually selected in the lesion and a value in nearby healthy tissue (developed in-house in MATLAB). Regions of interests were also manually drawn in the contralateral healthy tissue. Portions of the lesions or contralateral healthy tissue within GM or WM were differentiated with the aid of the b1000 DWI, which had strong gray–white matter contrast, and PGSE fractional anisotropy maps. Although initially outlined separately, deep and cortical GM were considered together in summary statistics.
DTI eigenvalues were computed for the PGSE and OGSE scans using ExploreDTI v4.8.3. The regions identified using the DWI scan were applied to the PGSE and OGSE DTI parameter maps to measure parallel (λ‖) and perpendicular (λ┴) eigenvalues and MD (MD=(λ‖+2λ┴)/3) in the WM lesions (n=9) and MD in the GM lesions (n=11). The fractional changes of eigenvalues and MD in the lesion compared with the contralateral healthy tissue were calculated. Statistical significance was evaluated using a 2-way ANOVA for changes with OGSE relative to PGSE and in lesions relative to contralateral tissue (P<0.05). To evaluate potential dependencies on the delay between stroke onset and imaging, linear regression was performed between the relative b=0 image intensity and MD of lesion and contralateral healthy tissue (for both OGSE and PGSE) versus MRI time poststroke onset. In a water phantom where no boundaries are present, consistent MD was measured for OGSE (2.17±0.02×10–3 mm2/s) and PGSE (2.18±0.03×10–3 mm2/s) in 4 separate scanning sessions.
Monte-Carlo Diffusion Simulations
Simulations of the DWI experiment for various geometric surfaces4 were performed using the Camino diffusion toolkit.18 Three-dimensional (3D) mesh surfaces consisting of cylinders arranged in a hexagonal lattice were generated (Matlab) with a uniform initial radius and intracellular volume fraction. Beaded axons were modeled as axisymmetric 3D unduloids with a degree of undulation (A) scaling from 0 (cylinder) to 1.4
The dynamics of the simulation consisted of a random walk of 20 000 magnetic spins. The initial position was random but consistent with the intracellular and extracellular volume fractions of the geometric surfaces. At each of 1000 time points, spins moved a distance equivalent to a free diffusivity of 1.7×10–3 mm2/s, which was chosen from the current study using λ‖ measured in healthy WM tissue using OGSE. Spins that intersected a boundary were elastically reflected. The simulated diffusion gradient waveforms were identical to those of the in vivo experiments. The spin phase was updated at each time step with the total signal equal to the phase-sensitive average of all spins. The effects of T2 relaxation were ignored.
Simulations were performed using all possible combinations of physiologically plausible initial axon diameters of (1, 2, 3, 4, 5, 6, 7, 8) μm, axonal volume fractions of (0.4, 0.5, 0.6, 0.7, 0.79), and beading amplitudes of (0, 0.2, 0.4, 0.6, 0.73). The volume fraction of 0.79 and beading amplitude of 0.73 were the maximum possible while enforcing the unduloid shape. The signal values obtained from the simulations for the various axon diameters were combined in a weighted mean using an axon distribution similar to ex vivo findings in the human corpus callosum by Aboitiz et al.19 Aboitiz et al estimated that the axons shrank by ≈65% after fixing and embedding the neural tissue, which was incorporated into the estimated axon distribution. The simulation parameters that provided the closest match to in vivo WM eigenvalue differences between the lesions and healthy tissue were determined using a least squares analysis of a cubic-spline interpolated (factor of 20) simulation space, and agreement with experiment was assessed using a coefficient of determination (R2) adjusted for the number of fitting parameters.
The raw DWI images show hyperintense stroke lesions, which correspond to hypointensities on quantitative MD maps for standard PGSE with a typically long diffusion time of 40 ms; however, there was little change of MD for the shorter 4 ms diffusion time OGSE scans (Figure 2). Over the 9 subjects with WM lesions, the OGSE yielded an MD decrease of only 8% on average in WM lesions compared with the typically observed decrease of ≈40% for PGSE (Figure 3A). This trend was consistently observed in the ischemic WM for each of the 9 subjects. Diffusion measured either parallel or perpendicular to the WM fibers (ie, axons) revealed further insight into the microstructure of ischemic axons. Both the parallel (λ‖) and perpendicular (λ┴) diffusion were reduced in the lesions relative to healthy tissue for PGSE (Figure 3C and 3D), but notably λ‖ did not reduce nearly as much for OGSE, whereas λ┴ was almost unchanged in the lesion compared with healthy tissue for OGSE. In the GM where tissue structure is macroscopically isotropic (ie, λ‖ and λ┴ are similar), a similar MD decrease in the lesion was observed for both OGSE and PGSE (Figure 3B). No statistically significant correlations were observed between the b=0 signal ratio or relative MD between lesion and contralateral tissue versus MRI time poststroke onset.
In Monte-Carlo simulations using an axon diameter distribution similar to histological results in the human corpus callosum (Figure 4A and 4B), both beading and swelling strongly affected the diffusion parameters (Figure 4C and 4D). Beading with an amplitude of 0.53 in combination with cellular swelling from an axonal volume fraction of 0.66 to 0.79 closely approximated the changes in the in vivo DTI parameters of WM (Figure 5). Notably, simulations with only swelling or only beading did not recapitulate the overall experimental findings nearly as well (adjusted R2=0.67 for swelling only, 0.72 for beading only, 0.88 for combination in Figure 5). Both swelling and beading are required for accurate modeling because the decreases in the parallel diffusion (for PGSE and less so for OGSE) that were observed in vivo were only mirrored in the simulations when beading occurred, and the decrease in perpendicular diffusion for PGSE observed in vivo was only mirrored in the simulations with an increase in volume fraction (Figures 3 and 4).
The key finding is a dramatic decrease in diffusion contrast of stroke in WM for OGSE (−8% change in MD) relative to PGSE (−37% change in MD) DTI. The pseudo normalization of MD in ischemic WM in humans for decreased diffusion times shows similar trends with work by Does et al14 and Wu et al16 in rodent GM that is ischemic. The results also agree with a PGSE study in rats where the estimated diffusion coefficients between ischemic and healthy tissue were more similar at diffusion times near 5 ms compared with 12 ms (based on Figure 2 in Norris et al20). However, the rodent studies did not report WM measurements, and our human study showed no significant difference of the degree of MD reductions in the GM portions of the stroke lesions between PGSE and OGSE over all 11 patients. The differential GM findings in rodents and human may be because the OGSE diffusion time achievable on a clinical MRI scanner may not be short enough to be sensitive to changes that occur at the smaller size of dendrites in GM relative to the larger axons in WM (eg, our minimum diffusion time, limited by human MRI hardware constraints, was 4 ms which is a factor of 4 to 10× larger than the rodent studies). In addition, MD measurements in the GM is more challenging given the shape and thinness of cortical GM which is more prone to error from partial volume effects and lower signal-to-noise ratio in the deep GM because of greater T2 decay. The lack of diffusion time–related differences in ischemic GM likely also reflects unique microstructural aspects given the large volume of GM occupied by cell bodies and dendrites instead of ordered axons in WM (ie, fewer neurites). The OGSE/PGSE results could provide the biophysical mechanism behind the general trend of greater decreases of MD in ischemic lesions in WM relative to GM lesions measured with PGSE21 (eg, Figure 2, 41-hour case).
The good agreement of simulation and in vivo human experiment in patients with acute stroke indicates that beading and cytotoxic edema are essential features for microstructural alterations in ischemic WM tissue and argues against the long-held belief that swelling alone is responsible for MD decreases in stroke.1 Notably, dendrite beading is a general phenomenon that applies to both axons and dendrites5 and can explain ischemia-induced MD changes in both WM and GM. Although the results from this work do not directly suggest any mechanism in the GM, observations at shorter diffusion times in rodent models14,16 are consistent with beading. This confirms the prior hypothesis that beading is sufficient to decrease the MD based on simulations and ex vivo results in a stretch model of rat sciatic peripheral nerve beading.4 This insight into the fundamental microstructural changes that occur post ischemia were enabled by the ability to probe much shorter diffusion length scales than previously in human studies (4 versus 40 ms) using the OGSE diffusion method. The experimental findings also argue against changes in viscosity or cytoplasmic streaming as the cause of diffusion reductions because in that case the measurements would not depend on the diffusion time, as they clearly do here. Notably, beading can also explain the reduction of diffusion of intracellular-only metabolites post stroke, such as N-acetyl aspartate, where there is no compartmental shift of the molecules from extracellular to intracellular space.2,3 It is known that the MD of water remains uniformly reduced for the first several days after stroke.22,23 In line with this, there was no significant correlation of the attenuated MD reduction with OGSE with imaging delay across the wide stroke onset times of 13 to 106 hours, but future work will need to examine earlier hyperacute onset times.
The general in vivo trends of a smaller parallel diffusion decrease for OGSE compared with PGSE and a decrease in perpendicular diffusivity for PGSE compared with little change for OGSE were mirrored in the simulations for almost any increase in volume fraction and beading amplitudes over the range 0.3 to 0.7. Notably, the simulation parameters that resulted in the best match with the in vivo results correspond well to values estimated using other methods. The beading amplitude of 0.53 is consistent with ex vivo measurements of beading induced by stretch4 and osmolarity changes,24 and a volume fraction increase from 0.66 to 0.79 is consistent with iontophoresis measurements in rodent models.25
A marked diffusivity reduction of water in rodent brain has been shown to be induced by silencing aquaporin 4 channels.26 Such an explanation can account for diffusivity decreases without requiring any cell swelling; however, edema is known to occur acutely after stroke and beading is associated with edema. Our results suggest that a microstructural change is a significant component responsible for diffusivity reductions after stroke, but does not preclude additional participation from aquaporin 4 or other alterations in membrane permeability. However, the differential responses of the diffusivity related to direction parallel or perpendicular to the WM tracts in our experimental results suggest that isotropic changes from aquaporin 4 that is only in astrocytes27 may not play a major role. Moreover, it has been shown in vivo that the effects of permeability in stroke lesions are observed only at high b-values (>3000 s/mm2), which is considerably greater than those used here (300 s/mm2).28
WM consists of a considerable volume of glial cells (mostly astrocytes), which were not modeled in the simulations. However, the anisotropic findings in WM here and observations of fractional anisotropy changes after stroke29 suggest that the axons (and dendrites, by extension) play a key role. Nevertheless, the specific role of glial cells needs to be determined, although incorporation of glial cells in simulations is not trivial given that astrocytes have many processes with size similar to dendrites and axons (ie, spheres or cylinders are not adequate).
Beading is associated with a variety of phenomena, including axon stretch,4 osmolarity imbalances,24 spreading depression,30 and seizures.31 Most importantly, recent studies using in vivo microscopy have demonstrated that beading as a consequence of acute ischemia is caused by the combination of oxygen–glucose deprivation and depolarization.6 Interestingly, on reperfusion, the beading in dendrites was remediated (axons were not studied), which coincides with observations of MD lesion reversal in stroke after reperfusion.32 Although beading can, but does not necessarily, lead to cell death,30–33 understanding how beading relates to the diffusion MRI changes may have implications to improve detection of injury in stroke and potentially other disorders because it could guide the development of methods that are specifically tailored to detect beading or membrane potential changes. Also, because transient MD reductions of similar magnitude to ischemia have been observed with spreading depression in healthy brain and peri-ischemic regions,34 these results implicate that transient reversible beading may underpin this key phenomenon or other acute neurological injuries with reductions of MD.
In summary, a new diffusion MRI acquisition strategy applied to acute stroke patients for the first time has shown attenuated reduction of water diffusion in lesions at short diffusion times and diffusion time dependencies supporting the hypothesis that MD reductions of water after ischemia, at least in WM, are because of both neurite (axon) beading and swelling.
Sources of Funding
We thank the Heart and Stroke Foundation of Alberta, North West Territories, and Nunavut for operating funds, Alberta Innovates– Health Solutions and the Natural Sciences and Engineering Research Council (NSERC) for salary support, and the Research and Education Initiative Fund, a component of the Advancing a Healthier Wisconsin endowment at the Medical College of Wisconsin (Dr Budde).
- Received January 16, 2015.
- Revision received May 5, 2015.
- Accepted May 29, 2015.
- © 2015 American Heart Association, Inc.
- Budde MD,
- Frank JA.
- Murphy TH,
- Li P,
- Betts K,
- Liu R.
- Novikov DS,
- Jensen JH,
- Helpern JA,
- Fieremans E.
- Syková E,
- Nicholson C.
- Andrew RD,
- Labron MW,
- Boehnke SE,
- Carnduff L,
- Kirov SA.
- Risher WC,
- Ard D,
- Yuan J,
- Kirov SA.
- Zeng LH,
- Xu L,
- Rensing NR,
- Sinatra PM,
- Rothman SM,
- Wong M.
- Labeyrie MA,
- Turc G,
- Hess A,
- Hervo P,
- Mas JL,
- Meder JF,
- et al
- Takeuchi H,
- Mizuno T,
- Zhang G,
- Wang J,
- Kawanokuchi J,
- Kuno R,
- et al