Validation of In Vivo Magnetic Resonance Imaging Blood–Brain Barrier Permeability Measurements by Comparison With Gold Standard Histology
Background and Purpose—We sought to validate the blood–brain barrier permeability measurements extracted from perfusion-weighted MRI through a relatively simple and frequently applied model, the Patlak model, by comparison with gold standard histology in a rat model of ischemic stroke.
Methods—Eleven spontaneously hypertensive rats and 11 Wistar rats with unilateral 2-hour filament occlusion of the right middle cerebral artery underwent imaging during occlusion at 4 hours and 24 hours after reperfusion. Blood–brain barrier permeability was imaged by gradient echo imaging after the first pass of the contrast agent bolus and quantified by a Patlak analysis. Blood–brain barrier permeability was shown on histology by the extravasation of Evans blue on fluorescence microscopy sections matching location and orientation of MR images. Cresyl-violet staining was used to detect and characterize hemorrhage. Landmark-based elastic image registration allowed a region-by-region comparison of permeability imaging at 24 hours with Evans blue extravasation and hemorrhage as detected on histological slides obtained immediately after the 24-hour image set.
Results—Permeability values in the nonischemic tissue (marginal mean±SE: 0.15±0.019 mL/miṅ100 g) were significantly lower compared to all permeability values in regions of Evans blue extravasation or hemorrhage. Permeability values in regions of weak Evans blue extravasation (0.23±0.016 mL/miṅ100 g) were significantly lower compared to permeability values of in regions of strong Evans blue extravasation (0.29±0.020 mL/miṅ100 g) and macroscopic hemorrhage (0.35±0.049 mL/miṅ100 g). Permeability values in regions of microscopic hemorrhage (0.26±0.024 mL/miṅ100 g) only differed significantly from values in regions of nonischemic tissue (0.15±0.019 mL/miṅ100 g).
Conclusions—Areas of increased permeability measured in vivo by imaging coincide with blood–brain barrier disruption and hemorrhage observed on gold standard histology.
Hemorrhagic transformation is a serious complication of ischemic stroke that can increase the risk of mortality up to 11 times.1 Combined data from 6 major stroke trials showed that severe hemorrhage with significant mass effect occurs in 5% of patients with stroke treated with tissue plasminogen activator within 3 hours after symptom onset and in up to 6% in patients treated between 3 and 6 hours.2 Hemorrhagic transformation is a multifactorial phenomenon,3 with damage to the blood–brain barrier (BBB) and subsequent vascular leakage being considered 1 of the contributing mechanisms.4,5 Hemorrhagic transformation can occur spontaneously but is more often triggered by reperfusion.3 Oxidative stress occurs early after ischemia and is accrued by reperfusion. It causes damage to lipid-rich membranes in the BBB and consequently leads to vascular leakage and possibly vascular disruption in ischemic brain tissue.6,7 Additionally, oxidative stress stimulates inflammatory cytokine production and protease secretion by microglia, infiltrating leukocytes and resident cells of the neurovascular unit.8,9 As these neuroinflammatory mechanisms become activated, alterations in cytokine profiles, adhesion molecule expression, and tight junction components mediate further vascular leakage.3 Although many proteases are expressed in the brain under normal and ischemic conditions, both animal and human studies suggest that the matrix metalloproteinase family and the tissue plasminogen activator system play a central role in activating the proteolysis cascade.10–13
Early detection of a damaged BBB by imaging could potentially be used to identify patients who are more likely to have hemorrhagic transformation develop and would impact the selection of patients with ischemic stroke for acute reperfusion therapies.14–18 Measurement of BBB permeability using perfusion imaging relies on applying a mathematical model to the time–signal curves of an intravascular, partly permeating tracer recorded on the images. A mathematical model is always a simplified representation of the reality summarized by an equation. As long as assumptions underlying the model are respected, results are supposed to be correct but cannot necessarily be considered as definite, and validation against a gold standard is required.
The goal of this study is to validate BBB permeability measurements extracted from perfusion-weighted MRI through a relatively simple and frequently applied model, the Patlak model,19,20 by comparison with gold standard histology in a rat model of ischemic stroke.
Materials and Methods
The experimental animals were cared for in accordance with the Animal Welfare Act, and the experiments were conducted in compliance with our institutional guidelines for animal research and with the approval of the University of California San Francisco Committee on Animal Research.
Spontaneously hypertensive (SHR) male rats and Wistar male rats weighing 220 to 280 g (Charles River Laboratory, Hollister, CA for Wistar rats and Portage, MI for SHR rats) were subjected to a 2-hour filament occlusion of the right middle cerebral artery. MRI was obtained during occlusion and 4 hours and 24 hours after reperfusion.
All experiments—surgery and MRI—were conducted while the animals were anesthetized. Anesthesia was induced by isoflurane 3.5% and then maintained at 2%.
Immediately after the 24-hour MRI scan, the animal was injected with Evans blue (Sigma Chemical; 0.6 mL of a 2% solution in saline), which circulated for 30 minutes. After the circulation period, the animals were euthanized by decapitation and the brain was quickly removed and then prepared for histological processing.
Reversible middle cerebral artery occlusion was performed according to a previously described technique.21 In brief, we inserted a precoated suture (7.0; Ethicon) through the external carotid artery into the internal carotid artery. The suture was advanced 18 to 23 mm past the external carotid artery–internal carotid artery bifurcation to occlude the middle cerebral artery. Middle cerebral artery occlusion and associated ischemic injury were confirmed during occlusion by perfusion-weighted and diffusion-weighted MRI. Reperfusion of the middle cerebral artery territory was achieved by withdrawal of the suture 2 hours after occlusion.
The MRI was performed with a 2-T Bruker Omega CSI system (Bruker Instruments) equipped with Acustar S-150 self-shielded gradients (±20 g/cm, 15-cm inner diameter). The animals were placed supine on a plastic support with ear and bite bars to minimize head motion during breathing. The head was inserted into a custom 5.5-cm diameter birdcage transmit–receive imaging coil. A water-recirculating warming pad was wrapped around the animal below the neck to maintain body temperature at 37°C, as monitored by an intrarectal thermocouple.
The MRI protocol was the same at each time point (during occlusion and 4 hours and 24 hours after reperfusion) and included the sequences listed in Supplemental Table 1 (http://stroke.ahajournals.org). All images were obtained in the coronal plane, were obtained as 8 consecutive 2-mm slices (if not otherwise specified), and had an image matrix size of 128×128 pixels.
To achieve quantitative hemodynamic measurements of cerebral perfusion and permeability, 2 boluses were injected. The first bolus of contrast was administered to measure permeability and served as a preload bolus for the perfusion scans performed with a second bolus of contrast. As shown previously,22 this preloading minimizes the effect of contrast leakage for the perfusion imaging.
The brain parenchymal tissue was segmented by intensity thresholding on the T2-weighted images. All subsequent processing was limited to the voxels belonging to the brain parenchymal tissue mask determined by segmentation.
Perfusion maps were computed using commercially available software (Brain Perfusion, Extended Brilliance Workspace; Philips Healthcare) that was modified in a tailored fashion for the purpose of this research project. This software relies on the central volume principle, which is the most accurate for low injection rates of iodinated contrast material.23 After motion correction and noise reduction by an anisotropic edge-preserving spatial filter, the software applies curve-fitting by least mean squares to obtain mathematical descriptions of the time–density curves for each pixel. The cerebral blood volume (CBV) map is calculated from the area under the time–density curves compared to a similarly obtained venous reference curve.24 A closed-form deconvolution is then applied to calculate the mean transit time map.25 The deconvolution operation requires a reference arterial input function that is manually selected. This closed-form deconvolution provides mean transit time maps even for animal data with a mean transit time in brain tissue of 1.5 seconds, close to the sampling rate of 1 second, which is a known challenge for all perfusion algorithms. The cerebral blood flow was computed as: cerebral blood flow=CBV/mean transit time.
For the computation of permeability maps, the superior sagittal sinus was manually selected to provide the reference time–concentration curve for the intravascular contrast agent concentration. Permeability maps were computed by Patlak analysis,19,20 which provides the estimation of a local CBV together with the blood-to-brain transfer constant k1. The analysis included an additional stabilizing step in which the area under the reference curve was scaled such that a target median CBV of 3 mL/100 g was obtained in the contralateral hemisphere for all animals at all time points.
Frozen tissue sections of 50 μm were cut on a cryostat (Leica Microsystems) from bregma −5 mm to +2 mm,26 covering all regions possibly affected by infarction. Every 500 μm, 2 sections were collected and used for Cresyl-violet staining and fluorescent microscopy analyses.
Image Analysis: Infarction and Hemorrhage
Sections stained with Cresyl-violet were examined microscopically with varying magnifications to delineate infarction and to identify different types of hemorrhagic transformation. The infarction was identified on the Cresyl-violet section on light microscopy as a region of pallor that contained shrunken cell bodies characteristic of neuronal cell death.
Macroscopic hemorrhage was defined as blood visible without magnification and confirmed by higher magnification (20×). Microscopic hemorrhage was defined as blood only visible by microscopy (20×)27 showing tissue characteristics of a petechial bleed (Supplemental Figure I, http://stroke.ahajournals.org). Such regions usually showed colocation of extravascular red blood cells in the structurally preserved brain tissue with or without plugging of cerebral microvessels with red blood cells, as well as formation of small blood clots by the extravasated blood.
Image Analysis: Extravasation of Evans Blue
Fluoroscopy images were used to identify the Evans blue marker on all sections. Images were examined on a standard fluorescent microscope using a Nikon CY3 filter set. Images were digitally captured with a digital microscope camera AxioCam IC (Carl Zeiss AG). Strong and weak appearances of Evans blue were distinguished by identifying regions where >50% or <50% of visible cells had taken up the marker.
Alignment of In Vivo Imaging and Histology
For all histology sections, overview images showing the whole brain were created using an in-house Matlab application (The MathWorks) that allows stitching of multiple low-power images. All regions of interest (ROI) identified directly at the microscope (infarction, macroscopic and microscopic hemorrhage, strong and weak extravasations of Evans blue) were transferred onto the overview images using Matlab.
The in vivo and histology images were obtained almost simultaneously, ≈30 minutes apart. However, to address a possible evolution in the degree of hemispheric brain swelling and deformation of the specimen occurring during the histology processing, in vivo and histology image data were coregistered based on an anatomic atlas of the rat brain.26 Eighteen distinct slices from this atlas were selected that axially covered all imaged brain regions. A set of 20 landmarks was selected that was visible on histology sections and the T2-weighted MRI images. A multistep elastic registration process was designed and implemented so that individual brain hemispheres and manually defined ROI on all slices were aligned to their corresponding atlas slices. ROI defined on ground truth microscopy slides were made available on the in vivo permeability and perfusion maps (Figure 1).
Once registration was completed, permeability values in the ROI delineating the infarction and hemorrhages (macroscopic and microscopic) were recorded. Cerebral blood flow, CBV, and mean transit time values within these ROI were also recorded. Cerebral blood flow, CBV, mean transit time, and permeability values were also recorded in the contralateral nonischemic hemisphere.
To assess the accuracy of the registration process, 1 experienced observer (A.H.) marked the position of visible anatomic structures separately and in a blinded fashion on selected Cresyl-violet sections and on the atlas. The average distance between the location of these structures on the Cresyl-violet sections and their location on the atlas, considered as the reference, was calculated.
Permeability values in the macroscopic and microscopic hemorrhagic ROI and in the strong and weak Evans blue ROI were compared using a linear mixed regression model to values in the contralateral nonischemic hemisphere or values in the infarcted hemisphere, but outside the drawn ROI. The linear mixed regression model approach reflects the structure of repeated data and takes correlations of multiple measurements per individual rat into account. Marginal means for the different ROI were reported with standard errors (±SE).
Nonparametric Mann-Whitney U tests were used to compare quantitative data between SHR and Wistar rat groups. Quantitative data were described using average, SD, and range. All statistical tests were performed as 2-sided and P<0.05 was considered statistically significant.
Eleven male SHR rats and 11 male Wistar rats weighing 220 to 280 g were subjected to a 2-hour filament occlusion of the right middle cerebral artery. MRI was obtained during occlusion and 4 hours and 24 hours after reperfusion. The present study focuses on the MRI obtained 24 hours after reperfusion, just before injection of Evans blue and euthanization. The MRI during occlusion was used to confirm the inclusion of the animals in the study. The MRI obtained at 4 hours after reperfusion was used to assess the degree of reperfusion.
Animals were excluded from the study if the perfusion imaging showed no reperfusion or if the diffusion-weighted imaging lesion during occlusion involved only subcortical regions, but not the cortex. Seven normal Wistar rats and 1 SHR rat were excluded because of these criteria. One additional SHR rat was excluded because it died soon after reperfusion because of vessel perforation. The characteristics of the remaining 9 male SHR rats and 4 male Wistar rats are described in Table 1.
The same final infarction pattern was obtained in the Wistar and SHR rats, both on diffusion-weighted imaging images at 24 hours and on the Cresyl-violet sections. The infarction patterns for Wistar and SHR rats are presented in Supplemental Figure II (http://stroke.ahajournals.org) and show similar distribution of the infarcted area in both groups of rats.
Four hours after removal of the suture, reperfusion was examined on perfusion-weighted imaging at 4 hours after removal of the suture (Supplemental Figure III, http://stroke.ahajournals.org). All animals but 2 showed reperfusion from 72% up to 99% (Table 1). One SHR rat did not have reperfusion at all because the coating of the occluding filament remained at the origin of the middle cerebral artery.
Macroscopic and Microscopic Hemorrhage on Histology
Macroscopic hemorrhage occurred more frequently in SHR rats compared to Wistar rats. Macroscopic hemorrhage covered 1% of the hemisphere in SHR rats and 0.1% of the hemisphere in Wistar rats (statistically significant difference; P=0.02; Table 1). Microscopic hemorrhage tended to be present at 24 hours after reperfusion with a higher frequency in the SHR rat group, but the difference was not statistically significant (5% in SHR rats versus 2% in Wistar rats; P=0.22; Table 1).
Evans Blue Extravasation on Histology
Strong rates of Evans blue extravasation covered 5% of the hemisphere in Wistar rats and 2% in SHR rats (P=0.41). Weak Evans blue extravasation covered 7% of the hemisphere in Wistar rats and 3% in SHR rats (P=0.23; Table 1).
Alignment of In Vivo Imaging and Histology
From a total of 297 observations on 33 systematically selected Cresyl-violet slices, we determined the overall accuracy of registration as 0.34±0.28 mm. This corresponds to ≈1.1 voxels on the permeability maps. Accuracy of registration was 0.13±0.14 mm (≈0.4 voxels), 0.38±0.22 mm (≈1.3 voxels), and 0.30±0.27 mm (≈1 voxel) in the cortex, preoptic area, and striatum, respectively.
Permeability Values in Different Types of ROI
Both Wistar and SHR rats showed an average of 5% of increased permeability values at the scan 24 hours after reperfusion. There was no statistical difference of the 2 groups (P=0.70; Table 1). Permeability values were measured in the 6 different types of ROI: nonischemic, infarct side but outside the drawn ROI, strong extravasation of Evans blue, weak extravasation of Evans blue, microscopic hemorrhage, and macroscopic hemorrhage.
Permeability values in the nonischemic tissue (0.15±0.019 mL/miṅ100 g) and permeability values on the infarct side outside the ROI (0.18±0.018 mL/miṅ100 g) were significantly lower compared to permeability values in regions of Evans blue extravasation or hemorrhage. Permeability values in regions of weak Evans blue extravasation (0.23±0.016 mL/miṅ100 g) were significantly lower compared to permeability values of in regions of strong Evans blue extravasation (0.29±0.020 mL/miṅ100 g) and macroscopic hemorrhage (0.35±0.049 mL/miṅ100 g). Permeability values in regions of microscopic hemorrhage (0.26±0.024 mL/miṅ100 g) only differed significantly from values in regions of nonischemic tissue Permeability values in the nonischemic tissue (marginal mean±SE: 0.15±0.019 mL/miṅ100 g) and permeability values on the infarct side outside the ROI (0.18±0.18 mL/miṅ100 g) were significantly lower compared to all permeability values in regions of Evans blue extravasation or hemorrhage. Permeability values in regions of weak Evans blue extravasation (0.23±0.016 mL/miṅ100 g) were significantly lower compared to permeability values of in regions of strong Evans blue extravasation (0.29±0.020 mL/miṅ100 g) and macroscopic hemorrhage (0.35±0.049 mL/miṅ100 g). Permeability values in regions of microscopic hemorrhage (0.26±0.024 mL/miṅ100 g) only differed significantly from values on the infarct side outside the ROI (0.18±0.18 mL/miṅ100 g) or values in regions of nonischemic tissue (0.15±0.019 mL/miṅ100 g; Figure 2, Table 2). The permeability values in the different ROI were not significantly different in SHR and Wistar rats.
The purpose of this study was to validate in vivo BBB permeability measurements extracted from perfusion-weighted MRI by the same software that has been used in acute stroke patients28–31 in comparison with gold standard histology. This software relies on the Patlak model to calculate BBB permeability values and has been used to predict clinically significant hemorrhagic transformation in acute stroke patients receiving tissue plasminogen activator.32
Our results indicate that all areas of BBB disruption (Evans blue extravasation and hemorrhages) identified on histology have significantly increased permeability on imaging compared to nonaffected brain. Permeability imaging is also able to grade the degree of BBB disruption, because regions with strong Evans blue extravasation show higher permeability on imaging compared to regions with weak Evans blue extravasation. Permeability values on imaging were similar in regions with strong Evans blue extravasation and in regions with macroscopic hemorrhage.
For the present study, we focused on the 24-hour time point to validate the BBB permeability measurements. At 24 hours after reperfusion, histological signs of an ischemic injury have developed, the endothelium is severely damaged, and BBB disruption has occurred. The ruptured BBB was detected by injecting Evans blue immediately after the last imaging scan and circulated for 30 minutes. This study is a cross-sectional study comparing imaging and histology obtained almost simultaneously at 24 hours does not allow inferences to be made about the causal relationship between increased BBB permeability and hemorrhagic transformation in the setting of acute ischemic stroke.
To achieve our validation goal, we thought it was important to work on a dataset that included intact BBB values together with a wide range of permeable and disrupted BBB values. The hemispheric stroke model selected for this study reliably induced ischemic stroke with a diffusion lesion that affected 79% of the selected hemisphere in SHR rats and 80% in Wistar rats. The 2 different breeds, Wistar and SHR rats, were selected so that the same 2-hour middle cerebral artery occlusion could result in different degrees of BBB disruption and of hemorrhagic transformation, despite similar infarct size at 24 hours after reperfusion to brain tissue. SHR rats were more likely to have macroscopic hemorrhage develop compared to Wistar rats, which is in agreement with previously performed studies.27,33,34 However, the permeability values were not significantly different in SHR and Wistar rats, which might be because of the small sample size. Further investigation will be necessary to address this issue.
Whereas this study shows the sensitivity of in vivo BBB measurements by imaging, it also identifies limitations. Imaging assessment of the BBB permeability cannot distinguish between different reasons for contrast agent extravasation, namely permeation through an open BBB in otherwise intact vasculature and leakage of contrast agent through damaged vascular walls.
An additional limitation of this study is that Gd-DTPA (molecular weight, 552 Da) and Evans blue tagged albumin (≈68 kDa) differ in size and distribution volumes. Different extravasation mechanisms for both agents are also described in in vitro and in vivo studies under the condition of hypoxia.35,36 However, a reperfusion injury after 2-hour middle cerebral artery occlusion produces extensive damage to the whole neurovascular unit, including tight junctions and endothelial cells. Therefore, even if the amount of extravasated Gd-DTPA and Evans blue-tagged albumin may be different because of their respective sizes, we observed that the locations of Evans blue leakage on histology and areas with increased permeability values on Gd-DTPA imaging were the same, and as such our comparison of the 2 is justified.
Gd-DTPA can be linked to bovine serum albumin and Evans blue, and has similar properties and a similar extravasation mechanism as Evans blue. Gd-bovine serum albumin-Evans blue has been studied in the setting of BBB permeability in ischemic stroke37 but was not used in this study to adhere to the same imaging method already used in patients.
In conclusion, this study validates that regions of increased permeability measured in vivo by imaging coincide with BBB disruption and hemorrhage observed in gold standard histology. The imaging method that was used to assess BBB permeability disruption was exactly the same as the one used previously in patients with acute ischemic stroke to predict hemorrhagic transformation after administration of tissue plasminogen activator.
Sources of Funding
This study was supported by a seed grant from the University of California San Francisco Department of Radiology and Biomedical Imaging, NIH R01 NS27713 (W.L.Y), P01 NS44155 (W.L.Y., H.S.), NIH R21 NS070153 (H.S.), and AHA 10GRNT3130004 (H.S.).
J.B. is an employee of Philips Healthcare.
The online-only Data Supplement is available at http://stroke.ahajournals.org/cgi/content/full/STROKEAHA.110.597997/DC1.
- Received July 28, 2010.
- Accepted February 1, 2011.
- © 2011 American Heart Association, Inc.
- Berger C,
- Fiorelli M,
- Steiner T,
- Schabitz WR,
- Bozzao L,
- Bluhmki E,
- et al
- Gürsoy-Ozdemir Y,
- Can A,
- Dalkara T
- Rosell A,
- Ortega-Aznar A,
- Alvarez- Sabín J,
- Fernández-Cadenas I,
- Ribó M,
- Molina CA,
- et al.,
- Montaner J
- Lin K,
- Kazmi KS,
- Law M,
- Babb J,
- Peccerelli N,
- Pramanik BK
- Hjort N,
- Wu O,
- Ashkanian M,
- Sølling C,
- Mouridsen K,
- Christensen S,
- et al
- Longa EZ,
- Weinstein PR,
- Carlson S,
- Cummins R
- Wintermark M,
- Sesay M,
- Barbier E,
- Borbély K,
- Dillon WP,
- Eastwood JD,
- et al
- Ladurner G,
- Zilkha E,
- Iliff D,
- du Boulay GH,
- Marshall J
- George Paxinos CW
- Ding G,
- Nagesh V,
- Jiang Q,
- Zhang L,
- Zhang ZG,
- Li L,
- et al
- Hom J,
- Dankbaar JW,
- Schneider T,
- Cheng SC,
- Bredno J,
- Wintermark M
- Dankbaar JW,
- Hom J,
- Schneider T,
- Cheng SC,
- Lau BC,
- van der Schaaf I,
- et al
- Hom J,
- Dankbaar JW,
- Soares BP,
- Schneider T,
- Cheng SC,
- Bredno J,
- et al
- Neumann-Haefelin C,
- Brinker G,
- Uhlenküken U,
- Pillekamp F,
- Hossmann KA,
- Hoehn M
- Sood R,
- Yang Y,
- Taheri S,
- Candelario-Jalil E,
- Estrada EY,
- Walker EJ,
- et al
- Nagaraja TN,
- Karki K,
- Ewing JR,
- Croxen RL,
- Knight RA