Magnetic Resonance Imaging Protocol for Stroke Onset Time Estimation in Permanent Cerebral Ischemia

Long Abstract

MRI provides a sensitive and specific imaging tool to detect acute ischemic stroke by means of a reduced diffusion coefficient of brain water. In a rat model of ischemic stroke, differences in quantitative T₁ and T₂ MRI relaxation times (qT₁ and qT₂) between the ischemic lesion (delineated by low diffusion) and the contralateral non-ischemic hemisphere increase with time from stroke onset. The time dependency of MRI relaxation time differences is heuristically described by a linear function and thus provides a simple estimate of stroke onset time. Additionally, the volumes of abnormal qT₁ and qT₂ within the ischemic lesion increase linearly with time providing a complementary method for stroke timing. A (semi)automated computer routine based on the quantified diffusion coefficient is presented to delineate acute ischemic stroke tissue in rat ischemia. This routine also determines hemispheric differences in qT₁ and qT₂ relaxation times and the location and volume of abnormal qT₁ and qT₂ voxels within the lesion. Uncertainties associated with onset time estimates of qT₁ and qT₂ MRI data vary from ± 25 min to ± 47 min for the first 5 hours of stroke. The most accurate onset time estimates can be obtained by quantifying the volume of overlapping abnormal qT₁ and qT₂ lesion volumes, termed ‘V^overlap’ (± 25 min) or by quantifying hemispheric differences in qT₂ relaxation times only (± 28 min). Overall, qT₂ derived parameters outperform those from qT₁. The current MRI protocol is tested in the hyperacute phase of a permanent focal ischemia model, which may not be applicable to transient focal brain ischemia.

Introduction

Brain tissue is particularly vulnerable to ischemia due to the high dependence of the oxidative phosphorylation for ATP synthesis and limited energy reserves. Ischemia results in subtle time-dependent ionic changes in intracellular and extracellular spaces that lead to redistribution of brain water pools, release of excitotoxic neurotransmitters and ultimately initiation of destructive processes ¹. In focal ischemia, tissue damage spreads beyond the initial core if blood flow is not restored within a certain time frame ². The time of stroke onset is currently one of the key criteria in clinical decisions for the pharmacotherapy of ischemic stroke, including recanalization by thrombolytic agents ³. Consequently, many patients are automatically ineligible for thrombolytic therapy due to unknown symptom onset time, owing to the stroke occurring during sleep (‘wake-up stroke’), lack of witness, or being unaware of symptoms ^4,5. A procedure that determines the time of stroke onset is therefore required so that such patients may be considered for thrombolysis.

MRI probes water in vivo. Dynamics of which are severely perturbed by acute ischemic energy failure ⁶. Most notably, the diffusion of water governed by translational (thermal) motion of water molecules is reduced in the early moments of ischemia due to the energy failure ⁷. This in turn results in anoxic depolarization of neural cells ⁸. Diffusion MRI (DWI) has become a gold standard diagnostic imaging modality for acute stroke ⁹. The DWI signal increases rapidly in response to ischemia allowing ischemic tissue to be identified but does not show any time-dependency during the first few hours of ischemic stroke ¹⁰. Likewise, quantitative measures of water diffusion such as the apparent diffusion coefficient (ADC) or the trace of the diffusion tensor (D_av) decrease rapidly in ischemic tissue, but show no relationship with time from stroke onset in animal stroke models ¹⁰ and patients ¹¹.

Quantitative MRI (qMRI) relaxation parameters, qT₁, qT₂ and qT_1ρ, are governed by rotational motion and the exchange of water hydrogen atoms and show complex time-dependent changes in brain parenchyma following ischemic energy failure ⁶. Such time-dependent changes enabled stroke onset time to be estimated in in patients ¹² and animal models of ischemia ^13–15. In rat focal stroke, qT_1ρ increases almost instantaneously after ischemia onset and continues linearly for at least 6 hours ^13,14. qT₁ relaxation times also increase in a time-dependent fashion in ischemic brain tissue which can be described by two time constants: an initial fast phase followed by a slow phase lasting for hours ^8,16. Because of this biphasic increase, the use of qT₁ in stroke timing may be more complicated than that of qT_1ρ MRI ¹⁵. qT₂ relaxation times also show a bi-phasic change in rat focal stroke, whereby there is an initial shortening within the first hour, followed by a linear increase with time ¹³. The initial shortening can be explained by two parallel running factors including: (i) the buildup of deoxyhemoglobin resulting in the so-called ‘negative blood oxygenation level dependent effect’ and (ii), the shift of extracellular water into the intracellular space ^17,18. The time-dependent increase in qT₂ is likely due to cytotoxic and/or vasogenic edema with subsequent breakdown of intracellular macromolecular structures ¹⁸. Both qT_1ρ and qT₂ data provide accurate estimates of stroke onset time in preclinical models ¹⁴. qT₂ ¹² and T₂-weighted signal intensities ^19,20 have also been exploited for stroke onset time estimation in clinical settings.

In addition to hemispheric differences in quantitative relaxation times, the spatial distribution of elevated relaxation times within the ischemic region may also serve as surrogates for stroke onset time ¹⁴. In rat models of stroke, regions with elevated qT_1ρ, qT₂ and qT₁ relaxation times are initially smaller than the diffusion defined ischemic lesion but increase with time ^14,15,21. Hence quantification of the spatial distribution of elevated relaxation times as a percentage of the size of the ischemic lesion also enables stroke onset time to be estimated ^14,15. Here we describe the protocol to determine stroke onset time in a rat model of stroke using qMRI parameters.

Protocol

Animal procedures were conducted according to European Community Council Directives 86/609/EEC guidelines and approved by the Animal Care and Use Committee of the University of Eastern Finland, Kuopio, Finland.

1. Animal Model

• 1.1.Anesthetize male Wistar rats weighing 300 – 400 g with isoflurane in N₂/O₂ flow (70%/30%) through a facemask for the duration of the operation and MRI experiments. Induce anesthesia in ventilated hood. Maintain isoflurane levels between 1.5 and 2.4%. Lack of response to pinch reflex is taken as a sign of sufficient depth for surgical anesthesia. Monitor depth of anesthesia during MRI from breathing frequency via a pneumatic pillow under the torso (SA Instruments Inc, Stony Brook, NY). Use isoflurane scavenger attached to magnet bore.
• 1.2.Perform permanent middle cerebral artery occlusion (MCAO) to induce focal ischemic stroke. Use the intraluminal thread model for MCAO and carry out the operation according to methods given by Longa et al. ²².
  • 1.2.1.Leave the occluding thread (silicon-PTFE tempered monofil filament, diameter 0.22 mm) in place for the duration of the MRI experiment.
• 1.3.Analyze arterial blood gases and pH (e.g. blood analyser i-Stat Co, East Windsor, NJ, USA)
  • 1.3.1.During MRI monitor: breathing rate with the pneumatic pillow placed under the torso and rectal temperature using a rectal temperature monitoring system (SA Instruments Inc, NY). Maintain core temperature close to 37 °C using a water heating pad under the torso.
  • 1.3.2.Immediately after MCAO, secure rat in a cradle at the center of the magnet bore using a rat head holder (RAPID Biomedical GmbH, Germany). Inject 2 mL of saline intraperitoneally before transferring the rats into the magnet bore.

2. MRI

• 2.1.Acquire MRI data using a 9.4T/31 cm (with the 12 cm gradient insert) horizontal magnet interfaced to a Direct Drive console (Agilent, CA) equipped with an actively decoupled linear volume transmitter and quadrature receiver coil pair (RAPID Biomedical GmbH, Germany).
• 2.2.Scan each rat for up to 5 hours post MCAO. At hourly intervals (60, 120, 180, 240 min post MCAO), acquire 12 congruently sampled (0.5 mm slice-gap, slice thickness = 1 mm, field of view = 2.56 cm x 2.56 cm) coronal slices of the trace of diffusion tensor (2.2.1.), Carr-Purcell-Meiboom-Gill T₂ (2.2.2.) and Fast Low Angle Shot T₁ (2.2.3.).
  • 2.2.1.Obtain the trace of the diffusion tensor images (D_av= 1/3 trace [D]) with three bipolar gradients along each axis (the duration of the diffusion gradient = 5 ms, the diffusion time = 15 ms) and three b-values (0, 400 and 1400 s/mm-²s), where, Δ = 15 ms, ∂ = 5 ms, echo time (TE) = 36 ms, repetition time (TR) = 4000 ms and acquisition time = 7.36 minutes.
  • 2.2.2.Obtain the Carr-Purcell-Meiboom-Gill T₂ sequence with 12 echoes for T₂ quantification, where echo-spacing = 10 ms, TR = 2000 ms and acquisition time = 4.20 minutes.
  • 2.2.3.Obtain the Fast Low Angle Shot (FLASH) for T₁, where the time from inversion to the first FLASH sequence (T₁₀) is 7.58 ms, with 600 ms increments of 10 inversions up to 5407.58 ms, TR = 5.5 ms, time between inversion pulses (T_relax) = 10 s and acquisition time = 8.20 minutes.

3. Image Processing

• 3.1.Computation of relaxometry and ADC maps: Compute qT₂, qT₁ and ADC maps using Matlab functions provided on University of Bristol website [DOI: pending], for which the input is a file path to the location of the MRI data.
  • 3.1.1.For T₂ data, apply Hamming filtering in k-space prior to reconstruction of images (or in image domain by convolution, with equivalent results but computationally less efficient). Compute qT₂ maps by taking the logarithm of each time series and solving on a voxel-wise basis by linear least-squares. (A bi-exponential fit can also be performed to the T₂ decays, but the voxel-wise F-tests tests revealed that voxels within the image for which additional parameters could not be justified).
  • 3.1.2.For T₁ data, apply Hamming filtering in k-space prior to reconstruction of images. Perform T₁ fitting according to methods given in ²³. To deal with the problem of unknown sign (due to the use of magnitude images), the lowest-intensity point may be either excluded, or estimated in the course of fitting with similar results.
  • 3.1.3.For Diffusion-weighted data, apply Hamming filtering by convolution in image domain (this is more straightforward due to the segmented k-space trajectory). Fit ADC maps by the method in¹³.

• 3.2.Identification of ischemic tissue
  • 3.2.1.Identify ischemic tissue on reciprocal D_av images (1/D_av) as this provides clear contrast for lesion identification. To generate ischemic volumes of interest (VOI), define ischemic tissue as voxels with values one median absolute deviation above the median value of the whole-brain 1/D_av distribution. To identify homologous regions in the non-ischemic hemisphere, reflect the ischemic VOI about the vertical axis. Manually adjust non-ischemic VOIs to avoid including voxels containing cerebrospinal fluid.
  • 3.2.2.
    In order to determine the relationship of qT₁ and qT₂ with time post MCAO, for each rat and time-point load the ischemic and non-ischemic VOIs onto qT₁ and qT₂ maps. Extract mean relaxation times and calculate the percentage difference in qT₁ and qT₂ between hemispheres (ΔT₁ and ΔT₂) using the following equation:

    $$\Delta T_x=100*\left(\frac{\left(T_x^{ischaemic}-T_x^{contralateral}\right)}{\left(T_x^{ischaemic}+T_x^{contralateral}\right)}\right)$$

    Where T_x is the chosen parameter, qT₁ or qT₂. $T_x^{ischaemic}$ refers to the mean relaxation time of the ischemic VOI and $T_x^{contralateral}$ the mean relaxation time in the non-ischemic VOI. The non-ischemic VOI should be used so that each rat serves as its own control.
  • 3.2.3.Use the following criteria to identify voxels with elevated qT₁ and qT₂: any voxels within the ischemic VOI with relaxation times exceeding the median relaxation time of the qT₁ or qT₂ distribution in the non-ischemic VOI by more than one half-width at half-maximum (HWMH). These criteria mean relaxation times must have been in the 95^th percentile or higher to be classified as ‘high’. Use of the median relaxation time of the non-ischemic VOI allows each rat to serve as its own control.
  • 3.2.4.To picture the spatial distribution of relaxation time changes within regions of decreased diffusion, identify and color code voxels with elevated qT₁ or qT₂ as well as voxels with both elevated qT₁ and qT₂ termed ‘qT₁ and qT₂ overlap’.
  • 3.2.5.To determine the size of the lesion according to qT₁ and qT₂, compute the parameter f ( as introduced by Knight et al. ¹⁸) from MRI data acquired for each rat and time-point. f₁ and f₂ represent the number of voxels with high qT₁ or qT₂ (respectively) as a percentage of the size of the ischemic VOI.
    Use the following equation to calculate f₁ and f₂:

    $$f_x=100*\left(\frac{N^{High}-N^{Low}}{N^{Lesion}}\right)$$

    Where x refers to the relaxation time (qT₁ or qT₂), N^High refers to the number of ‘high’ relaxation time voxels in the ischemic VOI, N^Low is the number of ‘low’ relaxation time voxels in the ischemic VOI and N^Lesion, the total number of voxels within the ischemic VOI. Criteria for identifying voxels with elevated qT₁ and qT₂ are outlined in section 3.2.3. ‘Low’ voxels are voxels with relaxation times less than the median qT₁ or qT₂ of the non-ischemic VOI by one HWHM. The subtraction of N^Low allows for decreases in relaxation times due to ischemia or other pathologies ¹⁷. Determine the extent of ‘qT₁ and qT₂ overlap’ by calculating the volume of overlapping elevated qT₁ and qT₂ as a percentage of whole brain volumes, hereby referred to as ‘V^Overlap’. Use the following equation:

    $$V^{Overlap}=100*\left(\frac{N^{Overlap}}{N^{Wholebrain}}\right)$$

    Where, N^Overlap refers to the number of voxels within the ischemic VOI with both ‘high’ qT₁ and ‘high’ qT₂ N^Wholebrain and represents the total number of voxels in the whole rat brain. Determine the number of voxels in the rat brain by manually creating a VOI around the whole brain on qT₂ relaxometry maps.

4. Verification of ischemic lesion with triphenyletrazolium chloride (TTC)

• 4.1.Immediately after decapitation, carefully extract the rat brain from the skull. Complete this procedure within 10 minutes from the time the rat was decapitated.
• 4.2.Store brains in refrigerated 0.01 M phosphate buffered saline (PBS) before using a rat brain slicer matrix to section the brain into serial 1 mm-thick coronal slices.
• 4.3.After sectioning, incubate each brain slice in 20 mL PBS containing TTC at 37 °C for 30 min in the dark, as recommended in ²⁴. Although 1% TTC concentration is acceptable, use 0.5% for improved contrast.
  • 4.3.1.Cover the containers of the sections in foil to keep it dark.
• 4.4.After incubation, remove the TTC solution using a pipette and wash slices in three changes of PBS.
• 4.5.Immediately photograph slices using a standard light microscope and a digital camera.

5. Statistical Analysis

• 5.1.Carry out statistical analysis using Matlab and statistical software such as Statistical Package for the Social Sciences (SPSS) or Microsoft Excel.
• 5.2.Determine relationship of MR parameters with time
  • 5.2.1.Perform Pearson’s correlations on pooled rat data to determine the relationship of ΔT₁, ΔT_2, f₁ and f₂ and V^Overlap with time post MCAO.
  • 5.2.2.For parameters that show a significant linear relationship (p < .05), perform linear least square regression to determine whether stroke onset time can be predicted by quantifying the parameter of interest. Use the root mean square error (RMSE) to assess the accuracy of onset time estimates.
• 5.3.Quantification of lesion size
  • 5.3.1.To compare lesion sizes according to different qMRI parameters, conduct one-way related ANOVAs and Fisher’s least significant difference post-hoc on the average number of voxels in the ischemic VOI and the average number of voxels with high qT₁ and qT₂. Differences are considered significant at p < .05. If sphericity assumptions are not met according to Mauchly’s sphericity test, degrees of freedom and significance values should be corrected according to Greenhouse Geisser estimates.

Representative Results

Across rats the blood gas profiles were as follows: SO₂ 95.8 ±3.2%, P_aCO₂ 51.6 ± 2.9 mmHg, and pH 7.30± 0.04.

Typical D_av, qT₂ and qT₁ images from a central slice of a representative rat at 4 time points post MCAO are shown in the first 3 panels of Figure 1a. Images in the other panels of Figure 1 show the automatically detected ischemic lesion in red and regions within the ischemic lesion with elevated qT₁, qT₂ and regions with V^overlap are shown in green. Up to 2 hours post-MCAO, regions with high qT₁ within the ischemic lesion were significantly larger than high qT₂ (p < .01), but converged with time (Figure 1). The ischemic lesion by D_av was also larger than regions of high qT₁ (p < .05) and qT₂ (p < 05) in the first two hours.

Figure 1. Changes in qMRI parameters due to ischemic stroke in an example rat.

Panel (a) shows example qMRI images over a 4-hour period of ischemia. The first four columns show D_av maps, qT₂ maps, qT₁ maps and T₂ weighted images respectively. Remaining columns show the D_av maps with various representations of the automatically segmented lesions. The automatically detected D_av lesion is shown in column 5 in red. In column 6 the D_av lesion is shown in red with voxels with high qT₂ shown in green. In column 7 the D_av lesion is shown in red with high qT₁ voxels shown in green. In column 8 the D_av lesion is shown in red with voxels with both elevated qT₁ and qT₂ (V^overlap) shown in green. Panel (b) shows the distribution of qT₂ in the D_av lesion as a function of time post MCAO, as well as the non-lesion qT₂ distribution at time zero. Panel (c) shows the corresponding qT₁ distributions, the adjacent legend pertaining to both panels (b) and (c). Panel (d) shows a TTC-stained brain slice after the animal was sacrificed at 6 hours post-MCAO.

The time dependencies of qMRI parameters are shown in Figure 2. All qMRI parameters were significant predictors of time post MCAO (ΔT_1: R² = 0.71, ΔT₂: R² = 0.75, f₁: R² = 0.53, f₂: R² = 0.82, V^overlap: R² = 0.87). Based on the RMSE for each parameter, uncertainties associated with estimates of time since stroke onset were ± 37 min for ΔT₁, ± 28 min for ΔT₂, ± 47 min for f₁, ±34 min for f₂, and ±25 min for V^overlap. Thus V^overlap gave the most accurate estimate of time since stroke onset.

Figure 2. The relationships between time post-MCAO and qMRI parameters relevant to the timing of ischemia.

Panel (a) shows f₁,(b) the f₂ parameter, (c), V^overlap, (d),ΔT₁ and (e), ΔT₂. Best fit for each parameter (solid red line) and RMSE error bars (solid black lines) are shown. Dotted lines represent each of the induvial 5 rats subjected to MCAO.

TTC staining of brains samples around 6 hours after MCAO verified irreversible ischemic damage predominantly in grey matter (Figure 1d).

Discussion

The current protocol for estimation of stroke onset time in rats uses quantitative diffusion and relaxation time MRI data rather than signal intensities of respective weighted MR contrast images¹⁹. Recent evidence points to inferior performance of image intensities in estimating stroke onset time ^{14, 25}. In the ‘diffusion-positive’ stroke lesion our MRI protocol provides stroke onset times from qT₁ and qT₂ MRI data with an accuracy of half an hour or so. It is a general trend that qT₂ data outperforms that of qT₁. The best accuracy for onset time determination is obtained from the volume of overlapping elevated qT₁ and qT₂ (V^overlap).

The images in Figure 1 demonstrate that while the reduced diffusion coefficient appears rather uniform, regions with abnormal qT₁ and qT₂ are heterogeneously scattered within the ischemic lesion. This finding is in accordance with previous observations and is likely due to different sensitivities of these qMRI parameters to pathophysiological changes caused by ischemia ⁶. This suggests qMRI parameters may be informative of tissue status and supports notions that DWI over-estimates ischemic damage ²⁶. Indeed, recent preclinical evidence points toward the heterogeneity of ischemic damage within diffusion defined lesions²⁷. Thus the combination of diffusion, qT₁ and qT₂ potentially provides information on stroke onset time and tissue status, both of which are clinically useful for treatment decisions regarding patients with unknown onset.

V^overlap and f₂ gave the most accurate estimates of stroke onset time. The benefit of quantifying relaxation times is that unlike signal intensities they are insensitive to inherent variations caused by technical factors such as magnetic field inhomogeneities and proton density ⁶, including the expected magnetic field variation within the ischemic lesion ¹⁸. Reduced uncertainty associated with onset time estimates of qT₁ and f₁ are likely due to the aforementioned bi-phasic response of qT₁ to ischemia, which contributes to the shallow slope of the time-dependent qT₁ change ^8,15,16. The MRI data shown (Figure 2) are in accordance with previous works ^13,14, in that the time courses of relaxation time differences between the ischemic and contralateral non-ischemic brain are adequately described by linear functions. It is, however, important to note that the underpinning hydrodynamic changes due to ischemia are not linear^1,18.

The current MRI protocol for stroke timing is demonstrated in rat subjected to permanent ischemia using the Longa et al. procedure²². In our experience the Longa et al. procedure fails to induce MCAO in 10-20% of rats, however, as ADC is used to verify presence of ischemia, the experiments can be terminated prematurely. Failure to induce MCAO is often due to imperfect occluder thread. A further factor resulting in experimental failures is that MCAO is a severe procedure causing death of up to 20% of rats during a prolonged MRI session.

The stroke onset timing protocol applies only to permanent ischemia. In rat focal ischemia with reperfusion, the relationship between D_av and qT₁ or qT₂ will dissociate as D_av recovers, but for qT₁ and qT₂ may or may not, depending on duration of ischemia prior to reperfusion^{8, 28}. Additionally, the evolution of ischemic damage is likely to be more variable in stroke patients due to individual differences in factors affecting microcirculation such as age and co-morbidities (e.g., diabetes, hypertension, heart disease). These factors will inevitably influence the time dependence of f₁, f₂ and V^overlap in human strokes and thus requires investigation in clinical settings.

To conclude, qMRI parameters provide estimates of stroke onset time. V^overlap and f₂ provide the most accurate estimates and may also be informative of tissue status. qMRI could therefore be clinically beneficial in terms of aiding treatment decisions for patients with unknown onset time. An issue to be considered here is that the gray-to-white-matter ratio in rat brain is much higher than in humans, and hydrodynamics in these brain tissue types may vary¹⁸. Nevertheless, further investigation into the time dependence of f₂, V^overlap and qT₂ in hyper acute stroke patients is therefore warranted.

Short Abstract.

A protocol for stroke onset time estimation in a rat model of stroke exploiting quantitative magnetic resonance imaging (qMRI) parameters is described. The procedure exploits diffusion MRI for delineation of the acute stroke lesion and quantitative T₁ and T₂ (qT₁ and qT₂) relaxation times for timing of stroke.