brain research 1583 (2014) 169–178
Available online at www.sciencedirect.com
www.elsevier.com/locate/brainres
Research Report
Infarct volume prediction using apparent diffusion coefficient maps during middle cerebral artery occlusion and soon after reperfusion in the rat Rau´l Tudelaa,n, Guadalupe Soriab,a, Isabel Pe´rez-De-Puigc,d, Dome`nec Rosa,e, Javier Pavı´aa,f, Anna M. Planasc,d a
CIBER de Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN), Barcelona, Spain Experimental MRI 7T Unit, IDIBAPS, Barcelona, Spain c Department of Brain Ischemia and Neurodegeneration, Institut d’Investigacions Biomèdiques de Barcelona (IIBB)-Consejo Superior de Investigaciones Científicas (CSIC), Barcelona, Spain d IDIBAPS, Barcelona, Spain e Biophysics and Bioengineering Laboratory, University of Barcelona, Barcelona, Spain f Nuclear Medicine Department, Hospital Clínic, Barcelona, Spain b
art i cle i nfo
ab st rac t
Article history:
Middle cerebral artery occlusion (MCAO) in rodents causes brain infarctions of variable sizes
Accepted 6 August 2014
that depend on multiple factors, particularly in models of ischemia/reperfusion. This is a major
Available online 14 August 2014
problem for infarct volume comparisons between different experimental groups since unavoid-
Keywords:
able variability can induce biases in the results and imposes the use of large number of subjects.
Ischemic stroke
MRI can help to minimize these difficulties by ensuring that the severity of ischemia is
Magnetic resonance imaging (MRI)
comparable between groups. Furthermore, several studies showed that infarct volumes can be
Stroke animal models
predicted with MRI data obtained soon after ischemia onset. However, such predictive studies
Apparent diffusion coefficient (ADC)
require multiparametric MRI acquisitions that cannot be routinely performed, and data processing using complex algorithms that are often not available. The aim here was to provide a simplified method for infarct volume prediction using apparent diffusion coefficient (ADC) data in a model of transient MCAO in rats. ADC images were obtained before, during MCAO and after 60 min of reperfusion. Probability histograms were generated using ADC data obtained either during MCAO, after reperfusion, or both combined. The results were compared to real infarct volumes, i.e.T2 maps obtained at day 7. Assessment of the performance of the estimations showed better results combining ADC data obtained during occlusion and at reperfusion. Therefore, ADC data alone can provide sufficient information for a reasonable prediction of infarct volume if the MRI information is obtained both during the occlusion and soon after reperfusion. This approach can be used to check whether drug administration after MRI acquisition can change infarct volume prediction. & 2014 Elsevier B.V. All rights reserved.
n Correspondence to: CIBER de Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN), Plataforma d’Imatge Mèdica, CEK Building. IDIBAPS. Rosselló 149-153, Barcelona 08036, Spain. E-mail address: [email protected] (R. Tudela).
http://dx.doi.org/10.1016/j.brainres.2014.08.008 0006-8993/& 2014 Elsevier B.V. All rights reserved.
170
1.
brain research 1583 (2014) 169–178
Introduction
The rat intraluminal middle cerebral artery occlusion (MCAO)/ reperfusion model is broadly used in preclinical studies of stroke (Macrae, 1992; McAuley, 1995). However, several factors account for a rather high degree of inter-animal variability in infarct volumes after transient intraluminal MCAO. The strain (Aspey et al., 2000; Oliff et al., 1995; Prieto et al., 2005; Walberer et al., 2006) and age of the animals (Sutherland et al., 1996), the surgical method (Huang et al., 1999; Zarow et al., 1997), surgical interventions causing infarction in territories of the internal (Kanemitsu et al., 2002) or external (Dittmar et al., 2003) carotid arteries, the extent of collateral circulation (Dittmar et al., 2003) and the efficacy of reperfusion contribute to infarct volume variability. This variability between animals can certainly be a source of bias when the effects of treatments are evaluated. Therefore, identification of the acute lesion and prediction of the subsequent fate of the ischemic tissue would help to compare treatment groups more accurately. Multimodal magnetic resonance imaging (MRI) is nowadays widely available in experimental research and it is an excellent tool for imaging brain damage in ischemia. Various imaging techniques are used to obtain predictive information in experimental stroke (Farr and Wegener, 2010), among which perfusion- and diffusion-weighted MRI are the most useful techniques for detecting signs of acute brain ischemia. Prediction using critical thresholds over apparent diffusion coefficient (ADC) or cerebral blood flow (CBF) maps is used in human stroke (Arenillas, 2002; Thomalla et al., 2003). More complex algorithms have been developed to predict tissue fate in experimental animal studies that are based on a generalized linear model (Wu et al., 2007), the probability of infarct (Shen et al., 2005), improved iterative self-organizing data-analysis algorithm (Lu et al., 2005; Shen and Duong, 2008), and artificial neural network (Huang et al., 2010) or
support vector machine (Huang et al., 2011). These methods provide a reasonably accurate prediction of the final infarct volume determined by T2 MRI or histology one or more days after MCAO. However, many of these algorithms are quite complex to implement and they use multimodal MRI information by combining perfusion and diffusion data, among other parameters. Nevertheless, ADC maps seem to provide valuable information for adequate prediction of the final tissue outcome, particularly in models with reperfusion (Bråtane et al., 2009; Huang et al., 2010; Leithner et al., 2011; Rosso et al., 2011), and they can be easily obtained. Previous predictive studies used permanent MCAO or transient MCAO, but for the latter they used information acquired after reperfusion while they did not use MRI during ischemia. In the present study, we aimed to evaluate the value of acute ADC maps to predict infarct volume after MCAO/ reperfusion in the Wistar rat. We measured ADC during MCAO and shortly after reperfusion since recovery of ADC has been reported after transient ischemia (Ringer et al., 2001).
2.
Results
2.1.
ADC and T2 Maps
For each subject, we constructed several maps with the ADC data obtained during MCAO and after reperfusion and with the final infarct volume determined from the T2 map at day 7. These maps are shown for individual rats (numbered 1 to 11) in Fig. 1. Group A (upper row) (rats 1 to 5) is the group used to compute the probability histograms and group B (bottom row) (rats 6 to 11) is the test group (see Section 5). During MCAO (Fig. 1A), all rats showed a decay of the ADCocc values in the right hemisphere and a corresponding increase in the relative ADC (rADC) values. In all cases the affected volume (see Table 1)
Fig. 1 – Coronal brain slices selected for each rat representation. (A) relative ADC maps at 60 min after occlusion, (B) relative ADC maps at 60 min after reperfusion, (C) difference in the T2 maps between the basal values before the infarct and the values 7 days after occlusion and (D) masks delimiting the final infarct from the T2 difference maps for two groups of rats. The first row shows group A with rats 1 to 5 used to compute the probability histograms. The second row shows group B with rats 6 to 11 as the test group. The number below each brain image is that of the rat.
171
brain research 1583 (2014) 169–178
Table 1 – Evaluation of the performance of estimations for the different rat groups and probability histograms. Infarcted volume (mm3)
Relative error (%)
RMSE (mm3)
R2 coef
Sensitivity (%)
Specificity (%)
Threshold
Group A (5 rats) Prob 1d occ Prob 1d rep Prob 2d Mask T2
115736 105743 109747 108741
32732 53723 1075
28.8 48 10.6
0.65 0.24 0.97
71711 50719 6477
9372 9273 9373
0.36 0.19 0.23
Group B (6 rats) Prob 1d occ Prob 1d rep Prob 2d Mask T2
133723 91753 114726 94731
64745 35718 35720
50.8 42.2 30.3
0.34 0.63 0.66
7179 30725 7078
9073 9273 9372
0.36 0.19 0.23
All Rats Prob 1d occ Prob 1d rep Prob 2d Mask T2
125721 98735 112726 100725
47730 44715 24713
42.3 44.9 23.5
0.41 0.43 0.79
7177 39717 6776
9172 9272 9372
0.36 0.19 0.23
Mean and 95% confidence interval for the infarct volume, as calculated with the different probability histograms and with the T2 mask at day 7. In addition, values for the relative error and the root mean square error (RMSE) of the different estimations are, compared with the final infarct volume using the T2 mask at day 7. Correlation coefficient, Sensitivity and Specificity means and 95% CI are shown for groups A and B (see Methods) and for the group containing all the rats, using the various probability histograms. The final column shows the probability threshold used with each estimation method.
corresponded to less than 10% of the whole brain volume and no statistically significant differences were found between the affected volumes of groups A and B. After reperfusion (Fig. 1B) several behaviors were observed in the rADCrep maps compared to the corresponding rADCocc maps. In 5 out of the 11 rats there was a noticeable recovery of rADCrep values (e.g. rats 2 and 10), while in the other cases the rADCrep values either did not change in relation to the values during MCAO (e.g. rats 1 and 6) or they increased (i.e. progressive ADC decrease) in the zone of the infarct core (e.g. rats 4 and 7). Fig. 1C shows the final infarct for each rat as the difference in the T2 maps (ΔT2) between the basal values before MCAO and 7 days after MCAO. The ΔT2 maps were used to draw the masks shown in Fig. 1D. These masks identified the voxels corresponding to final infarcted tissue. For the rats of group A, this information was then used to compute the probability histograms. Finally, the infarct mask of each rat was compared to the result of the different estimations obtained with the probability histograms (see below).
2.2.
Probability histograms
Normalized histograms (Fig. 2) showing the distribution of rADCocc and rADCrep values were obtained with all the rats of group A. These combined histograms are obtained considering all the values of group A rats joined in one data set. The distribution curve for rADCocc (Fig. 2A) reached a maximum at value 0, which indicates no differences between ADC values during MCAO and the corresponding basal ADC. Both sides of the curve correspond to the points deviating from this exact correspondence to the basal prescan values. The left side of the curve illustrates the deviation from the theoretical prescan value due to natural variability in the
process of matching two images taken at different time points (this variation does not exceed 20%). However, the right side of the curve, instead of being symmetrical to the left side, shows a lateral lobe, corresponding to increased rADC values, between 0.2 and 0.6 (i.e. ADC values showing decline during MCAO). The difference between the areas of the left and right side of the distribution is proportional to the number of ischemic voxels. After reperfusion (Fig. 2A), the right lobe was less pronounced than during MCAO and there was a shift of the main peak from 0, resulting in a less asymmetric distribution than during MCAO. The distribution of the infarcted voxels during occlusion was within the rADC range from 0.2 to 0.6 (with a maximum around 0.4), whereas after reperfusion the distribution of the infarcted voxels was more uniform and had a wider range, from values of 0.1 to over 0.6. The probability distributions of infarct were calculated by dividing the number of infarcted voxels by the number of all the voxels for each rADC value of the histograms. Fig. 2B shows the probability distributions with either the rADCocc or rADCrep values. Then, a two-dimensional (2d) rADC distribution was obtained, using group A rats, by combining the information of the rADCocc and the rADCrep for all the brain voxels and for the infarcted voxels. The ratio between these two distributions provided a two-dimensional probability histogram, as shown in Fig. 3A. All the probability histograms that were generated can be used as look-up tables for a subsequent rat to obtain the probability of infarction for each brain voxel. To check the results of the 2d-probability histograms obtained with the information from the rats in group A, we generated an additional probability histogram that included the rats in group B (n¼ 6) (Fig. 3B). The distribution in this new histogram was very similar to the one obtained with the rats in group A (Fig. 3A), i.e. rADC values had an equivalent
172
brain research 1583 (2014) 169–178
Fig. 2 – 1d histograms. (A) Normalized histogram of the relative ADC values during occlusion (black) and at reperfusion (gray), for all the brain volume voxels (dots) and for only the infarcted ones (triangles), taking into account the mask of the T2 map at 7 days. The histogram is obtained from the relative ADC values of the 5 rats of group A. (B) Infarct probability using only relative ADC during occlusion (black) or at reperfusion (gray) separately. The probability histogram is obtained from the 5 rats of group A.
Fig. 3 – 2d probability histograms. (A) 2d probability histogram obtained with group A (rats 1–5), combining the relative ADC values during occlusion and at reperfusion to obtain the probability of infarct. (B) 2d probability histogram obtained with the group B (rats 6–11). The axes represent the relative ADC values during occlusion (x-axis) and after reperfusion (y-axis). The color intensity level of the points gives the probability of infarct.
probability of infarction in both distributions, and the area with the highest probability of infarction was equally located on the top right for equivalent rADC values. The main difference between the two probability maps was the size of the different areas, indicating a slightly different range of rADC values due to the difference among the experimental groups used. Therefore, the probability distribution shown in Fig. 3A seems to be a suitable distribution for all the rats of both groups and can be used as a look-up table to estimate the final infarct volume of a given rat from its ADC values obtained during and after MCAO.
2.3.
Infarct volume estimation
The probability histograms computed with group A were then used as look-up tables to obtain probability maps of the brain
volume from where the estimated infarct volume for all the rats (n¼ 11) was calculated. For this purpose, we used the various histograms generated in the study: (1) the histogram with the information from rADCocc only (prob-1d-occ, Fig. 4A); (2) the histogram with the information of rADCrep only (prob-1d-rep, Fig. 4B); and (3) the 2d probability histogram that uses the combination of both rADCocc and rADCrep values (prob-2d, Fig. 4C). These probability maps illustrate the location of the infarct, and the higher or lower probability values could be interpreted as an indication of the degree of ischemia at each voxel. The main difference between the first two histograms (prob-1d-occ and prob-1d-rep) is that the predicted infarct volume was smaller when only the values after reperfusion were considered. However, in the latter case (prob-1d-rep), the probability of infarct within the core area had higher values than when using the values during occlusion (prob-1d-occ). In the
brain research 1583 (2014) 169–178
173
Fig. 4 – Coronal brain sections showing the probability maps of infarction for the 11 rats. (A) using only the probability obtained from the relative ADC values during occlusion, (B) only the probability obtained from the relative ADC values after reperfusion, and (C) the probability obtained from the combined relative ADC values during occlusion and after reperfusion.
case of prob-2d, the volume with the highest probability of infarct was similar to the one obtained using data during occlusion (prob-1d-occ), but the probability values within the infarct core were closer to the ones obtained with the values after reperfusion (prob-1d-rep). To estimate the infarct volume first a probability threshold for each rat of the A group was obtained to minimize the difference between the estimated and real infarct volumes. Then the mean of these probability thresholds for the group A of rats was used to predict the infarct volume for all the rats, selecting which voxels are infarcted or not. The predicted
infarct volume was calculated by adding all the voxels marked as infarcted using the probability threshold. Fig. 5 shows the infarct volume (in mm3) estimated using the probabilities obtained with the relative ADC values during occlusion, after reperfusion and both combined. The fact that the infarct volume estimation for rats of group A is neither perfect is explained considering that both the probability histogram and the probability threshold used are obtained from an average of several rats and not just from one, therefore the averaged values may differ from the values of each particular case, and the estimation would differ from the ideal case.
174
brain research 1583 (2014) 169–178
Fig. 5 – Infarcted volume for each rat (in mm3). Measured with the T2 map at 7 days mask (T2 infarct mask), or using the probabilities during occlusion (Prob 1d occlusion), at reperfusion (Prob 1d reperfusion) and both combined (Prob 2d).
To assess whether the various probability maps provided correct estimations of the final infarct volume, we compared for each rat the estimated infarct volumes to the actual infarct volume obtained from the masks of the T2 maps at day 7. Overall, the predicted volume was slightly over-estimated when it was obtained using either 1d probability during occlusion (prob-1d-occ) or 2d probability (prob-2d). By contrast, the estimation of the predicted volume was worse when obtained using only the values after reperfusion (prob-1d-rep). This latter effect is due to the partial ADC recovery after reperfusion, which makes distinction between infarcted and non infarcted voxels less accurate, and the deviation from the predicted volumes increases. The averages and 95% confidence interval (95%CI) for infarct volume estimations in the different rat groups and the various estimation methods, and the final infarct volume measured using the T2 mask at day 7, together with the comparisons with the infarct volume predicted using the probability histograms, are shown in Table 1. The differences between the infarct volume estimations from ADC maps and the real T2 measurements are given in Table 1 as the average relative error and as the root mean square error (RMSE) for the different rat groups and probabilities used. The Table also shows the correlation coefficients R2, and the average and 95%CI of the sensitivity and specificity of the volumes estimated for each group. The final column indicates the probability threshold used. The sensitivity using prob-1d-rep with group B is too low, and makes this predicting method not reliable. However, with the 2d probability (prob-2d) smaller errors were obtained and the correlation was better than with either of the 1d probabilities. Also the sensitivity was higher with the prob-2d than the prob-1d-rep in all the cases. To illustrate the errors in volume estimation for the individual rats we measured the difference (computed as a distance) between the mean rADC value in the ischemic zone at occlusion and the corresponding value at reperfusion. For each rat, we plotted this value against the corresponding real
infarct volume (T2 mask) (Fig. 6), and represented the relative error in infarct volume estimation by the size of the circle representing each value (black circles indicate rats of group A, and white circles indicate rats of group B). We identified 4 rats out of 11 with a high difference between the mean rADC value at occlusion and that at reperfusion (difference 40.15), and small infarct sizes (final infarct volumeo100 mm3). Three of the 4 rats with bad prediction belong to the group B, and two of them behave as outliers since they had a very high difference between the mean rADC value at occlusion and that at reperfusion (40.3). Therefore, it seems that the prediction loses accuracy when there is a high change in the mean values of the ADC from occlusion to reperfusion. To check whether this effect was due to specific features of group A, we also reversed the model using probability histogram of group B (Fig. 3B) to perform the prediction for all rats. By doing this, the relative errors of the group A increased, and they decreased in some rats of group B. However, the two rats of group B with the highest mean rADC difference between occlusion and reperfusion (Fig. 6) still had high relative errors. Therefore, regardless of the reference group used for the estimation, a high difference in mean rADC between occlusion and reperfusion seems to impair the predictive value of the model.
3.
Discussion
In this study we provide a method for infarct volume prediction using MRI data obtained within the first hours after cerebral ischemia/reperfusion in rats. Infarct volume predictions were carried out from the ADC maps obtained at two time points, during and after MCAO, by using the ADC data either separately or combined. These maps are transformed into relative values using an ADC map obtained before MCAO. It seems possible to obtain relative values of ADC using other techniques; one possibility might be to use the contra-lateral
brain research 1583 (2014) 169–178
175
Fig. 6 – Infart volume estimated vs. rADC differences. Distribution of the different infarct volumes estimated with the mask in the T2 map at 7 days respect to the distance between the mean of the rADC values in occlusion and reperfusion, considering only the points in the infarcted volume. The circle diameter and the numbers inside the graphic indicate the relative error in the infarct volume estimation using the 2d probability.
hemisphere values. Measuring CBF can increase the quality of infarct volume estimations (Huang et al., 2010), but requires more acquisitions for the evaluation. The techniques for CBF quantification in rodents may not be widely available in different laboratories, or might not be sufficiently accurate due to low spatial resolution. Using the ADC data only, the sensitivity and specificity are slightly lower than when CBF data are included (Huang et al., 2010). However, the performance of the current estimations is comparable to those obtained using other methods with more complex and timeconsuming algorithms (Huang et al., 2011, 2010), which could be a limitation for their practical application in many experimental laboratories. Here, using ADC data only, we found that the sensitivity was 70% and the specificity 93% when information obtained during MCAO and after reperfusion was used for the estimations. The mean relative error of the infarct volume estimation by using the information from rADC data during MCAO and after reperfusion (35%) is also lower than the relative error obtained only considering the mean infarct volume of group A as predicted volume for all group B rats (50%). The prediction method used here is based on probability histograms created in a group of rats that were used as lookup tables for infarct estimation in other rats. Voxel-by-voxel comparisons were carried out after a registration process to ensure the correspondence between acquisitions. Non-rigid registrations were chosen to account for the fact that the lesion could displace the brain midline due to the presence of edema, since this effect modifies the internal geometry of the brain. We showed that the averaged result obtained from a set of rats could be reliably used to estimate the infarct volume of other rats subjected to the same experimental ischemia model. Since the procedure used here avoids the need to perform manual segmentation to evaluate the final infarct volume, the variability introduced by different users is
prevented and infarct volume prediction could be systematized easily. In general, the final infarct volume was better estimated by using the combined ADC data than by using only the information from a single time point. The estimation seems to be reasonably good in most of the rats excepting for those showing a large difference (40.3) between the mean rADC value at occlusion and that at reperfusion. Therefore, it seems that a big variation in the intensity of the ADC from occlusion to reperfusion impairs the estimation of infarct volume with this method. This variation might not be a frequent phenomenon, but taking it into account could contribute to assess the accuracy of the final volume estimation. In general, the prediction based on ‘ADC at reperfusion only’ is the one that differs more from the final infarct volume showing that a partial ADC recovery makes the estimated lesion at reperfusion less accurate. This corroborates earlier findings showing that reversal of ADC at early reperfusion does not invariably indicate final tissue viability (Neumann-Haefelin et al., 2000; Ringer et al., 2001; Rojas et al., 2006). Overall, the simplified prediction method reported here can provide evidence to show that the expected infarct volume is similar in different experimental groups. Then, the infarct volumes estimated could be compared to the final infarct volumes to evaluate whether treatments carried out after reperfusion may change the course of the prediction.
4.
Summary
This study showed the usefulness of ADC data at early stages during and after MCAO to estimate final infarct volume in rats. We compared the use of input ADC data at two time points (during MCAO and after reperfusion), either individually or combined. We computed the probability histograms in a group of rats, which were subsequently used as look-up
176
brain research 1583 (2014) 169–178
tables to carry out infarct estimation in a different group of rats. The accuracy of the infarct estimation was quantitatively evaluated by comparing it with the final infarct volume obtained from the T2 maps at day 7. Performance measurements showed better results by using information of the ADC during occlusion combined with the ADC after reperfusion, rather than by separately using either the occlusion or reperfusion data. In conclusion, we provide a simplified method, based on ADC maps obtained during and after transient ischemia, to predict final infarct volumes in a stroke model of 90 min intraluminal MCAO in rats.
5.
Experimental procedure
5.1.
Animals and surgery
The study was carried out in 11 adult male Wistar rats (Charles River, Lyon, France), weighing 250–300 g. Rats were housed in cages with controlled temperature (2171 1C) and humidity (55710%), under a 12 h light/12 h dark cycle and with food and water available ad libitum. All experiments were performed in accordance with the local legislation (Decret 214/1997 of July 30th by the ‘Departament d’Agricultura, Ramaderia i Pesca de la Generalitat de Catalunya’) under the approval of the Ethical Committee of the University of Barcelona (CEEA), and in compliance with European legislation. The surgery for intraluminal MCAO was carried out under isoflurane anesthesia, as previously described (Pérez-Asensio et al., 2010). The right MCA was occluded for 90 min. Then the thread was removed and reperfusion was allowed. This period of MCAO usually leads to striatal and cortical damage, but in some cases only striatal infarction is observed. Cortical brain perfusion was assessed by laser Doppler flowmetry and in all the cases occlusion induced a perfusion drop. The perfusion values normalized to the basal registry before surgery (100%) were 32711% and 120751% (mean7SD), during ischemia and reperfusion, respectively.
5.2.
Magnetic resonance imaging
MRI experiments were conducted on a 7T BioSpec 70/30 horizontal animal scanner (Bruker BioSpin, Ettlingen, Germany), equipped with a 12 cm inner diameter actively shielded gradient system (400 mT/m). The receiver coil was a phased-array surface coil for rat brain. Animals were placed in a supine position in a Plexiglas holder with a nose cone for administering anesthetic gases (2% isoflurane in a mixture of oxygen and N2O at a 30:70 ratio) and fixed by a tooth bar, ear bars and adhesive tape. Tripilot scans were carried out for accurate positioning of the animal's head in the isocenter of the magnet. MRI acquisitions were taken at 4 different time points in each rat: before MCAO surgery as the basal scan (prescan, ps), during MCAO (i.e. 60 min after starting the occlusion, occ 60 m), 60 min after reperfusion (rep 60 m), and 7 days after reperfusion (rep 7d).
5.2.1.
ADC maps
ADC scans were acquired during MCAO and 60 min after reperfusion using a pulsed gradient spin echo sequence with
the following acquisition parameters: echo time TE ¼55.07 ms, repetition time TR¼6250 ms, with a total acquisition time of 2 min 55 s, 6 b-values from 100 to 1000 s/mm2, a field of view FOV¼40 40 180 mm3, with a matrix size of 128 128 18 voxels and a resolution of 0.312 0.312 1 mm3/voxel. The same parameters were used for the basal acquisition.
5.2.2.
T2 relaxometry maps
T2 relaxometry scans were carried out 7 days after MCAO with a multi-slice multi-echo acquisition sequence with 16 effective echo times ramping from 11 to 176 ms, TR¼4926 ms, FOV ¼40 40 18 mm3, matrix size 256 256 18 voxels and resolution 0.156 0.156 1 mm3/voxel. The same parameters were used for the basal acquisition.
5.3.
Data analysis
ADC and T2 maps were calculated using Paravision 5.0 software (Bruker BioSpin, Ettlingen, Germany). Later these maps were processed with custom-made algorithms programmed in Matlab (The MathWorks, Inc, Natick, MA, USA) and ImageJ (National Institutes of Health, Bethesda, MD, USA). A binary mask was manually drawn over the T2 map in order to segment the brain from the background in the different acquisitions. Also, a threshold of 275 ms was applied to the T2 maps to reduce the noise. Then, all the maps acquired at different time points for each rat were co-registered to perform a voxel-based analysis.
5.3.1.
Registration process
A registration process was carried out for voxel-by-voxel comparisons between volumes, ensuring correspondence between the same points in images originally showing altered volumes due to brain edema. Non-rigid registration was performed by using Elastix 4.3 (Klein et al., 2010), a program based on Insight Segmentation and Registration Toolkit (ITK). The registration process used the basal scans before surgery as the reference images, whereas the acquisitions after MCAO were used as the moving images. A sequential registration, combining rigid and b-spline registrations, was carried out. Bilinear interpolation was performed to allow correspondence between the various resolutions of ADC and T2 maps. This interpolation also smoothed the images, which was convenient for the subsequent processing.
5.3.2.
ADC probability histogram
Following co-registration, ADC and T2 values at the various time points were used to compute the probability histograms. For each rat, the brain was segmented from the background in all the images using the mask obtained from the basal T2 map. Then a uniform smoothing filter of 3 3 pixels was applied to the original ADC and T2 maps. The relative values of ADC obtained 60 min after MCAO (rADCocc) and ADC at 60 min of reperfusion (rADCrep) were calculated by normalizing to the basal ADC values (obtained before MCAO). The final infarct volume was determined from the difference (ΔT2) between the T2 maps obtained 7 days after reperfusion
brain research 1583 (2014) 169–178
(T2rep 7d) and the corresponding basal T2 map obtained before MCAO (T2ps). For the relative ADC maps, Eq. (1) was used: rADCx ¼ 1
ADCx ADCps
ð1Þ
This calculation gives the relative difference of the ADC values (rADC) for a given time point in relation to the basal acquisition. rADC increases above 0 when the ADC value decreases. A segmentation mask was manually applied to the ΔT2 maps to select the voxels corresponding to infarcted tissue taking as reference intensity differences in the image above 20 ms to delimitate the affected area. This method was chosen for the segmentation, instead of a threshold directly applied to the T2rep 7d image, to avoid selecting noise areas as false infarcted voxels. With this data, a table was computed relating the rADCocc and rADCrep values with the infarct at 7 days (ΔT2). Data representation was carried out with two types of histograms pooling data from a group of rats (group A, see below). First showing the distributions of rADCocc and rADCrep values for the whole brain, and secondly calculating such distributions considering only the infarcted points. The histograms bins were selected of a 0.02 span sampling of the rADC values from 1 to 1. Afterwards, we obtained a probability histogram by dividing the histogram corresponding to the infarcted points by that of the whole brain. Thus, the probability histogram illustrates the proportion of infarcted voxels for each relative ADC value. Two-dimensional histograms were obtained by combining the information of the rADC at occlusion and after reperfusion. A histogram was built with rADCocc values in the x axis and rADCrep values in the y axis and taking into account all the voxels of the whole brain volume. A second histogram was computed in a similar way, but only considering the points that were marked as injured by the ΔT2 mask. From the ratio of both histograms, a new two-dimensional probability histogram was generated showing the proportion of infarcted points for each combination of rADC values. This 2d histogram was also processed to optimize the probability map. A median filter was applied on the two-dimensional probability histogram to eliminate the isolated points of the histogram. This procedure smoothed the probability map providing a more continuous distribution and avoiding the points representing only one specific rat and not an average behavior. Considering that the median filter would also smooth the average probability values, only a 3 3 median filter was used in order to make these differences small and no significant.
5.3.3.
Data analysis
An anatomical probability volume was generated from the rADC maps of each rat, using the probability histogram as a look-up table to obtain an estimation of the final infarct volume. The rats used in the study were separated into two groups: group A comprised the rats (rat 1 to rat 5) used to compute the global probability histograms; and group B was the test group containing rats (rat 6 to rat 11) that were used for infarct estimation by the general probability histograms obtained with the rats of group A. To evaluate the performance of the different probability
177
histograms, we compared the estimated infarct volume with the corresponding real infarct volume. We also performed a specificity and sensitivity analysis, which is useful for evaluating different classifiers. The statistical analysis to compare rat groups A and B was carried out with the Mann-Whitney test. The Kruskal–Wallis test compared the results obtained from the different probability histograms. When significant overall differences were found, a post-hoc analysis was performed with Dunn's multiple comparison test. All data were analyzed with GraphPad (GraphPad Software, Inc., CA, USA) and differences were considered significant if po0.05.
Acknowledgments CIBER-BBN is an initiative funded by the VI National R&D&i Plan 2008–2011, Iniciativa Ingenio 2010, Consolider Program, CIBER Actions and financed by the Instituto de Salud Carlos III with assistance from the European Regional Development Fund. Financed in part by the European Community (FP7/ 2007–2013; grant agreement number 201024), and the Spanish Government (SAF2009–08076, SAF2011-30492, ISCIII PS09/ 00527 and CDTI-CENIT/AMIT).
r e f e r e nc e s
Arenillas, J.F., 2002. Prediction of early neurological deterioration using diffusion- and perfusion-weighted imaging in hyperacute middle cerebral artery ischemic stroke. Stroke 33, 2197–2205. Aspey, B.S., Taylor, F.L., Terruli, M., Harrison, M.J., 2000. Temporary middle cerebral artery occlusion in the rat: consistent protocol for a model of stroke and reperfusion. Neuropathol. Appl. Neurobiol. 26, 232–242. Bra˚tane, B.T., Bastan, B., Fisher, M., Bouley, J., Henninger, N., 2009. Ischemic lesion volume determination on diffusion weighted images vs. apparent diffusion coefficient maps. Brain Res. 1279, 182–188. Dittmar, M., Spruss, T., Schuierer, G., Horn, M., 2003. External carotid artery territory ischemia impairs outcome in the endovascular filament model of middle cerebral artery occlusion in rats. Stroke 34, 2252–2257. Farr, T.D., Wegener, S., 2010. Use of magnetic resonance imaging to predict outcome after stroke: a review of experimental and clinical evidence. J. Cereb. Blood Flow Metab. 30, 703–717. Huang, J., Kim, L.J., Poisik, A., Pinsky, D.J., Connolly Jr., E.S., 1999. Titration of postischemic cerebral hypoperfusion by variation of ischemic severity in a murine model of stroke. Neurosurgery 45, 328–333. Huang, S., Shen, Q., Duong, T.Q., 2010. Artificial neural network prediction of ischemic tissue fate in acute stroke imaging. J. Cereb. Blood Flow Metab., 1–10. Huang, S., Shen, Q., Duong, T.Q., 2011. Quantitative prediction of acute ischemic tissue fate using support vector machine. Brain Res. 1405, 77–84. Kanemitsu, H., Nakagomi, T., Tamura, A., Tsuchiya, T., Kono, G., Sano, K., 2002. Differences in the extent of primary ischemic damage between middle cerebral artery coagulation and intraluminal occlusion models. J. Cereb. Blood Flow Metab. 22, 1196–1204. Klein, S., Staring, M., Murphy, K., Viergever, M.A., Pluim, J.P.W., 2010. Elastix: a toolbox for intensity-based medical image registration. IEEE Trans Med. Imag. 29, 196–205.
178
brain research 1583 (2014) 169–178
Leithner C., Fu¨chtemeier M., Mu¨ller S., Foddis M., Jorks D., Dirnagl U., Royl G. 2011. Can early MRI be used to improve the statistical quality of experimental stroke studies? In: Brain 2011 XXVth International Symposium on Cerebral Blood Flow, Metabolism, and Function, and Xth International Conference on Quantification of Brain Function with PET. Barcelona, Spain. Lu, M., Mitsias, P.D., Ewing, J.R., Soltanian-Zadeh, H., BagherEbadian, H., Zhao, Q., Oja-Tebbe, N., Patel, S.C., Chopp, M., 2005. Predicting final infarct size using acute and subacute multiparametric MRI measurements in patients with ischemic stroke. J. Magn. Res. Imag. 21, 495–502. Macrae, M.I., 1992. New models of focal cerebral ischaemia. Br J. Clin Pharmacol. 34, 302–308. McAuley, M.A., 1995. Rodent models of focal ischemia. Cerebrovas. Brain Metab. Rev. 7, 153–180. Neumann-Haefelin, T., Kastrup, A., de Crespigny, A., Yenari, M.A., Ringer, T., Sun, G.H., Moseley, M.E., 2000. Serial MRI after transient focal cerebral ischemia in rats: dynamics of tissue injury, blood-brain barrier damage, and edema formation. Stroke 31, 1965–1972. Oliff, H.S., Weber, E., Eilon, G., Marek, P., 1995. The role of strain/ vendor differences on the outcome of focal ischemia induced by intraluminal middle cerebral artery occlusion in the rat. Brain Res. 675, 20–26. Pe´rez-Asensio, F.J., De la Rosa, X., Jime´nez-Altayo´, F., Gorina, R., Martı´nez, E., Messeguer, A., Vila, E., Chamorro, A., Planas, A. M., 2010. Antioxidant CR-6 protects against reperfusion injury after a transient episode of focal brain ischemia in rats. J. Cereb Blood Flow Metab. 30, 638–652. Prieto, R., Carceller, F., Roda, J.M., Avendan˜o, C., 2005. The intraluminal thread model revisited: rat strain differences in local cerebral blood flow. Neurol Res. 27, 47–52. Ringer, T.M., Neumann-Haefelin, T., Sobel, R.A., Moseley, M.E., Yenari, M.Y., 2001. Reversal of early diffusion-weighted magnetic resonance imaging abnormalities does not necessarily reflect tissue salvage in experimental cerebral ischemia. Stroke 32, 2362–2369.
Rojas, S., Martı´n, A., Justicia, C., Falco´n, C., Bargallo´, N., Chamorro, A., Planas, A.M., 2006. Modest MRI signal intensity changes precede delayed cortical necrosis after transient focal ischemia in the rat. Stroke 37, 1525–1532. Rosso, C., Colliot, O., Pires, C., Delmaire, C., Valabre`gue, R., Crozier, S., Dormont, D., Baillet, S., Samson, Y., Lehe´ricy, S., 2011. Early ADC changes in motor structures predict outcome of acute stroke better than lesion volume. J. Neurorad. 38, 105–112. Shen, Q., Ren, H., Fisher, M., Duong, T.Q., 2005. Statistical prediction of tissue fate in acute ischemic brain injury. J. Cereb. Blood Flow Metab. 25, 1336–1345. Shen, Q., Duong, T.Q., 2008. Quantitative prediction of ischemic stroke tissue fate. NMR Biomed. 21, 839–848. Sutherland, G.R., Dix, G.A., Auer, R.N., 1996. Effect of age in rodent models of focal and forebrain ischemia. Stroke 27, 1663–1667. Thomalla, G.J., Kucinski, T., Schoder, V., Fiehler, J., Knab, R., Zeumer, H., Weiller, C., Ro¨ther, J., 2003. Prediction of malignant middle cerebral artery infarction by early perfusion- and diffusion-weighted magnetic resonance imaging. Stroke 34, 1892–1899. Walberer, M., Stolz, E., Mu¨ller, C., Friedrich, C., Rottger, C., Blaes, F., Kaps, M., Fisher, M., Bachmann, G., Gerriets, T., 2006. Experimental stroke: ischemic lesion volume and edema formation differ among rat strains (a comparison between Wistar and Sprague-Dawley rats using MRI). Lab Anim. 40, 1–8. Wu, O., Sumii, T., Asahi, M., Sasamata, M., Ostergaard, L., Rosen, B.R., Lo, E.H., Dijkhuizen, R.M., 2007. Infarct prediction and treatment assessment with MRI-based algorithms in experimental stroke models. J. Cereb. Blood Flow Metab. 27, 196–204. Zarow, G.J., Karibe, H., States, B.A., Graham, S.H., Weinstein, P.R., 1997. Endovascular suture occlusion of the middle cerebral artery in rats: effect of suture insertion distance on cerebral blood flow, infarct distribution and infarct volume. Neurol. Res. 19, 409–416.
Available online at www.sciencedirect.com
www.elsevier.com/locate/brainres
Research Report
Infarct volume prediction using apparent diffusion coefficient maps during middle cerebral artery occlusion and soon after reperfusion in the rat Rau´l Tudelaa,n, Guadalupe Soriab,a, Isabel Pe´rez-De-Puigc,d, Dome`nec Rosa,e, Javier Pavı´aa,f, Anna M. Planasc,d a
CIBER de Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN), Barcelona, Spain Experimental MRI 7T Unit, IDIBAPS, Barcelona, Spain c Department of Brain Ischemia and Neurodegeneration, Institut d’Investigacions Biomèdiques de Barcelona (IIBB)-Consejo Superior de Investigaciones Científicas (CSIC), Barcelona, Spain d IDIBAPS, Barcelona, Spain e Biophysics and Bioengineering Laboratory, University of Barcelona, Barcelona, Spain f Nuclear Medicine Department, Hospital Clínic, Barcelona, Spain b
art i cle i nfo
ab st rac t
Article history:
Middle cerebral artery occlusion (MCAO) in rodents causes brain infarctions of variable sizes
Accepted 6 August 2014
that depend on multiple factors, particularly in models of ischemia/reperfusion. This is a major
Available online 14 August 2014
problem for infarct volume comparisons between different experimental groups since unavoid-
Keywords:
able variability can induce biases in the results and imposes the use of large number of subjects.
Ischemic stroke
MRI can help to minimize these difficulties by ensuring that the severity of ischemia is
Magnetic resonance imaging (MRI)
comparable between groups. Furthermore, several studies showed that infarct volumes can be
Stroke animal models
predicted with MRI data obtained soon after ischemia onset. However, such predictive studies
Apparent diffusion coefficient (ADC)
require multiparametric MRI acquisitions that cannot be routinely performed, and data processing using complex algorithms that are often not available. The aim here was to provide a simplified method for infarct volume prediction using apparent diffusion coefficient (ADC) data in a model of transient MCAO in rats. ADC images were obtained before, during MCAO and after 60 min of reperfusion. Probability histograms were generated using ADC data obtained either during MCAO, after reperfusion, or both combined. The results were compared to real infarct volumes, i.e.T2 maps obtained at day 7. Assessment of the performance of the estimations showed better results combining ADC data obtained during occlusion and at reperfusion. Therefore, ADC data alone can provide sufficient information for a reasonable prediction of infarct volume if the MRI information is obtained both during the occlusion and soon after reperfusion. This approach can be used to check whether drug administration after MRI acquisition can change infarct volume prediction. & 2014 Elsevier B.V. All rights reserved.
n Correspondence to: CIBER de Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN), Plataforma d’Imatge Mèdica, CEK Building. IDIBAPS. Rosselló 149-153, Barcelona 08036, Spain. E-mail address: [email protected] (R. Tudela).
http://dx.doi.org/10.1016/j.brainres.2014.08.008 0006-8993/& 2014 Elsevier B.V. All rights reserved.
170
1.
brain research 1583 (2014) 169–178
Introduction
The rat intraluminal middle cerebral artery occlusion (MCAO)/ reperfusion model is broadly used in preclinical studies of stroke (Macrae, 1992; McAuley, 1995). However, several factors account for a rather high degree of inter-animal variability in infarct volumes after transient intraluminal MCAO. The strain (Aspey et al., 2000; Oliff et al., 1995; Prieto et al., 2005; Walberer et al., 2006) and age of the animals (Sutherland et al., 1996), the surgical method (Huang et al., 1999; Zarow et al., 1997), surgical interventions causing infarction in territories of the internal (Kanemitsu et al., 2002) or external (Dittmar et al., 2003) carotid arteries, the extent of collateral circulation (Dittmar et al., 2003) and the efficacy of reperfusion contribute to infarct volume variability. This variability between animals can certainly be a source of bias when the effects of treatments are evaluated. Therefore, identification of the acute lesion and prediction of the subsequent fate of the ischemic tissue would help to compare treatment groups more accurately. Multimodal magnetic resonance imaging (MRI) is nowadays widely available in experimental research and it is an excellent tool for imaging brain damage in ischemia. Various imaging techniques are used to obtain predictive information in experimental stroke (Farr and Wegener, 2010), among which perfusion- and diffusion-weighted MRI are the most useful techniques for detecting signs of acute brain ischemia. Prediction using critical thresholds over apparent diffusion coefficient (ADC) or cerebral blood flow (CBF) maps is used in human stroke (Arenillas, 2002; Thomalla et al., 2003). More complex algorithms have been developed to predict tissue fate in experimental animal studies that are based on a generalized linear model (Wu et al., 2007), the probability of infarct (Shen et al., 2005), improved iterative self-organizing data-analysis algorithm (Lu et al., 2005; Shen and Duong, 2008), and artificial neural network (Huang et al., 2010) or
support vector machine (Huang et al., 2011). These methods provide a reasonably accurate prediction of the final infarct volume determined by T2 MRI or histology one or more days after MCAO. However, many of these algorithms are quite complex to implement and they use multimodal MRI information by combining perfusion and diffusion data, among other parameters. Nevertheless, ADC maps seem to provide valuable information for adequate prediction of the final tissue outcome, particularly in models with reperfusion (Bråtane et al., 2009; Huang et al., 2010; Leithner et al., 2011; Rosso et al., 2011), and they can be easily obtained. Previous predictive studies used permanent MCAO or transient MCAO, but for the latter they used information acquired after reperfusion while they did not use MRI during ischemia. In the present study, we aimed to evaluate the value of acute ADC maps to predict infarct volume after MCAO/ reperfusion in the Wistar rat. We measured ADC during MCAO and shortly after reperfusion since recovery of ADC has been reported after transient ischemia (Ringer et al., 2001).
2.
Results
2.1.
ADC and T2 Maps
For each subject, we constructed several maps with the ADC data obtained during MCAO and after reperfusion and with the final infarct volume determined from the T2 map at day 7. These maps are shown for individual rats (numbered 1 to 11) in Fig. 1. Group A (upper row) (rats 1 to 5) is the group used to compute the probability histograms and group B (bottom row) (rats 6 to 11) is the test group (see Section 5). During MCAO (Fig. 1A), all rats showed a decay of the ADCocc values in the right hemisphere and a corresponding increase in the relative ADC (rADC) values. In all cases the affected volume (see Table 1)
Fig. 1 – Coronal brain slices selected for each rat representation. (A) relative ADC maps at 60 min after occlusion, (B) relative ADC maps at 60 min after reperfusion, (C) difference in the T2 maps between the basal values before the infarct and the values 7 days after occlusion and (D) masks delimiting the final infarct from the T2 difference maps for two groups of rats. The first row shows group A with rats 1 to 5 used to compute the probability histograms. The second row shows group B with rats 6 to 11 as the test group. The number below each brain image is that of the rat.
171
brain research 1583 (2014) 169–178
Table 1 – Evaluation of the performance of estimations for the different rat groups and probability histograms. Infarcted volume (mm3)
Relative error (%)
RMSE (mm3)
R2 coef
Sensitivity (%)
Specificity (%)
Threshold
Group A (5 rats) Prob 1d occ Prob 1d rep Prob 2d Mask T2
115736 105743 109747 108741
32732 53723 1075
28.8 48 10.6
0.65 0.24 0.97
71711 50719 6477
9372 9273 9373
0.36 0.19 0.23
Group B (6 rats) Prob 1d occ Prob 1d rep Prob 2d Mask T2
133723 91753 114726 94731
64745 35718 35720
50.8 42.2 30.3
0.34 0.63 0.66
7179 30725 7078
9073 9273 9372
0.36 0.19 0.23
All Rats Prob 1d occ Prob 1d rep Prob 2d Mask T2
125721 98735 112726 100725
47730 44715 24713
42.3 44.9 23.5
0.41 0.43 0.79
7177 39717 6776
9172 9272 9372
0.36 0.19 0.23
Mean and 95% confidence interval for the infarct volume, as calculated with the different probability histograms and with the T2 mask at day 7. In addition, values for the relative error and the root mean square error (RMSE) of the different estimations are, compared with the final infarct volume using the T2 mask at day 7. Correlation coefficient, Sensitivity and Specificity means and 95% CI are shown for groups A and B (see Methods) and for the group containing all the rats, using the various probability histograms. The final column shows the probability threshold used with each estimation method.
corresponded to less than 10% of the whole brain volume and no statistically significant differences were found between the affected volumes of groups A and B. After reperfusion (Fig. 1B) several behaviors were observed in the rADCrep maps compared to the corresponding rADCocc maps. In 5 out of the 11 rats there was a noticeable recovery of rADCrep values (e.g. rats 2 and 10), while in the other cases the rADCrep values either did not change in relation to the values during MCAO (e.g. rats 1 and 6) or they increased (i.e. progressive ADC decrease) in the zone of the infarct core (e.g. rats 4 and 7). Fig. 1C shows the final infarct for each rat as the difference in the T2 maps (ΔT2) between the basal values before MCAO and 7 days after MCAO. The ΔT2 maps were used to draw the masks shown in Fig. 1D. These masks identified the voxels corresponding to final infarcted tissue. For the rats of group A, this information was then used to compute the probability histograms. Finally, the infarct mask of each rat was compared to the result of the different estimations obtained with the probability histograms (see below).
2.2.
Probability histograms
Normalized histograms (Fig. 2) showing the distribution of rADCocc and rADCrep values were obtained with all the rats of group A. These combined histograms are obtained considering all the values of group A rats joined in one data set. The distribution curve for rADCocc (Fig. 2A) reached a maximum at value 0, which indicates no differences between ADC values during MCAO and the corresponding basal ADC. Both sides of the curve correspond to the points deviating from this exact correspondence to the basal prescan values. The left side of the curve illustrates the deviation from the theoretical prescan value due to natural variability in the
process of matching two images taken at different time points (this variation does not exceed 20%). However, the right side of the curve, instead of being symmetrical to the left side, shows a lateral lobe, corresponding to increased rADC values, between 0.2 and 0.6 (i.e. ADC values showing decline during MCAO). The difference between the areas of the left and right side of the distribution is proportional to the number of ischemic voxels. After reperfusion (Fig. 2A), the right lobe was less pronounced than during MCAO and there was a shift of the main peak from 0, resulting in a less asymmetric distribution than during MCAO. The distribution of the infarcted voxels during occlusion was within the rADC range from 0.2 to 0.6 (with a maximum around 0.4), whereas after reperfusion the distribution of the infarcted voxels was more uniform and had a wider range, from values of 0.1 to over 0.6. The probability distributions of infarct were calculated by dividing the number of infarcted voxels by the number of all the voxels for each rADC value of the histograms. Fig. 2B shows the probability distributions with either the rADCocc or rADCrep values. Then, a two-dimensional (2d) rADC distribution was obtained, using group A rats, by combining the information of the rADCocc and the rADCrep for all the brain voxels and for the infarcted voxels. The ratio between these two distributions provided a two-dimensional probability histogram, as shown in Fig. 3A. All the probability histograms that were generated can be used as look-up tables for a subsequent rat to obtain the probability of infarction for each brain voxel. To check the results of the 2d-probability histograms obtained with the information from the rats in group A, we generated an additional probability histogram that included the rats in group B (n¼ 6) (Fig. 3B). The distribution in this new histogram was very similar to the one obtained with the rats in group A (Fig. 3A), i.e. rADC values had an equivalent
172
brain research 1583 (2014) 169–178
Fig. 2 – 1d histograms. (A) Normalized histogram of the relative ADC values during occlusion (black) and at reperfusion (gray), for all the brain volume voxels (dots) and for only the infarcted ones (triangles), taking into account the mask of the T2 map at 7 days. The histogram is obtained from the relative ADC values of the 5 rats of group A. (B) Infarct probability using only relative ADC during occlusion (black) or at reperfusion (gray) separately. The probability histogram is obtained from the 5 rats of group A.
Fig. 3 – 2d probability histograms. (A) 2d probability histogram obtained with group A (rats 1–5), combining the relative ADC values during occlusion and at reperfusion to obtain the probability of infarct. (B) 2d probability histogram obtained with the group B (rats 6–11). The axes represent the relative ADC values during occlusion (x-axis) and after reperfusion (y-axis). The color intensity level of the points gives the probability of infarct.
probability of infarction in both distributions, and the area with the highest probability of infarction was equally located on the top right for equivalent rADC values. The main difference between the two probability maps was the size of the different areas, indicating a slightly different range of rADC values due to the difference among the experimental groups used. Therefore, the probability distribution shown in Fig. 3A seems to be a suitable distribution for all the rats of both groups and can be used as a look-up table to estimate the final infarct volume of a given rat from its ADC values obtained during and after MCAO.
2.3.
Infarct volume estimation
The probability histograms computed with group A were then used as look-up tables to obtain probability maps of the brain
volume from where the estimated infarct volume for all the rats (n¼ 11) was calculated. For this purpose, we used the various histograms generated in the study: (1) the histogram with the information from rADCocc only (prob-1d-occ, Fig. 4A); (2) the histogram with the information of rADCrep only (prob-1d-rep, Fig. 4B); and (3) the 2d probability histogram that uses the combination of both rADCocc and rADCrep values (prob-2d, Fig. 4C). These probability maps illustrate the location of the infarct, and the higher or lower probability values could be interpreted as an indication of the degree of ischemia at each voxel. The main difference between the first two histograms (prob-1d-occ and prob-1d-rep) is that the predicted infarct volume was smaller when only the values after reperfusion were considered. However, in the latter case (prob-1d-rep), the probability of infarct within the core area had higher values than when using the values during occlusion (prob-1d-occ). In the
brain research 1583 (2014) 169–178
173
Fig. 4 – Coronal brain sections showing the probability maps of infarction for the 11 rats. (A) using only the probability obtained from the relative ADC values during occlusion, (B) only the probability obtained from the relative ADC values after reperfusion, and (C) the probability obtained from the combined relative ADC values during occlusion and after reperfusion.
case of prob-2d, the volume with the highest probability of infarct was similar to the one obtained using data during occlusion (prob-1d-occ), but the probability values within the infarct core were closer to the ones obtained with the values after reperfusion (prob-1d-rep). To estimate the infarct volume first a probability threshold for each rat of the A group was obtained to minimize the difference between the estimated and real infarct volumes. Then the mean of these probability thresholds for the group A of rats was used to predict the infarct volume for all the rats, selecting which voxels are infarcted or not. The predicted
infarct volume was calculated by adding all the voxels marked as infarcted using the probability threshold. Fig. 5 shows the infarct volume (in mm3) estimated using the probabilities obtained with the relative ADC values during occlusion, after reperfusion and both combined. The fact that the infarct volume estimation for rats of group A is neither perfect is explained considering that both the probability histogram and the probability threshold used are obtained from an average of several rats and not just from one, therefore the averaged values may differ from the values of each particular case, and the estimation would differ from the ideal case.
174
brain research 1583 (2014) 169–178
Fig. 5 – Infarcted volume for each rat (in mm3). Measured with the T2 map at 7 days mask (T2 infarct mask), or using the probabilities during occlusion (Prob 1d occlusion), at reperfusion (Prob 1d reperfusion) and both combined (Prob 2d).
To assess whether the various probability maps provided correct estimations of the final infarct volume, we compared for each rat the estimated infarct volumes to the actual infarct volume obtained from the masks of the T2 maps at day 7. Overall, the predicted volume was slightly over-estimated when it was obtained using either 1d probability during occlusion (prob-1d-occ) or 2d probability (prob-2d). By contrast, the estimation of the predicted volume was worse when obtained using only the values after reperfusion (prob-1d-rep). This latter effect is due to the partial ADC recovery after reperfusion, which makes distinction between infarcted and non infarcted voxels less accurate, and the deviation from the predicted volumes increases. The averages and 95% confidence interval (95%CI) for infarct volume estimations in the different rat groups and the various estimation methods, and the final infarct volume measured using the T2 mask at day 7, together with the comparisons with the infarct volume predicted using the probability histograms, are shown in Table 1. The differences between the infarct volume estimations from ADC maps and the real T2 measurements are given in Table 1 as the average relative error and as the root mean square error (RMSE) for the different rat groups and probabilities used. The Table also shows the correlation coefficients R2, and the average and 95%CI of the sensitivity and specificity of the volumes estimated for each group. The final column indicates the probability threshold used. The sensitivity using prob-1d-rep with group B is too low, and makes this predicting method not reliable. However, with the 2d probability (prob-2d) smaller errors were obtained and the correlation was better than with either of the 1d probabilities. Also the sensitivity was higher with the prob-2d than the prob-1d-rep in all the cases. To illustrate the errors in volume estimation for the individual rats we measured the difference (computed as a distance) between the mean rADC value in the ischemic zone at occlusion and the corresponding value at reperfusion. For each rat, we plotted this value against the corresponding real
infarct volume (T2 mask) (Fig. 6), and represented the relative error in infarct volume estimation by the size of the circle representing each value (black circles indicate rats of group A, and white circles indicate rats of group B). We identified 4 rats out of 11 with a high difference between the mean rADC value at occlusion and that at reperfusion (difference 40.15), and small infarct sizes (final infarct volumeo100 mm3). Three of the 4 rats with bad prediction belong to the group B, and two of them behave as outliers since they had a very high difference between the mean rADC value at occlusion and that at reperfusion (40.3). Therefore, it seems that the prediction loses accuracy when there is a high change in the mean values of the ADC from occlusion to reperfusion. To check whether this effect was due to specific features of group A, we also reversed the model using probability histogram of group B (Fig. 3B) to perform the prediction for all rats. By doing this, the relative errors of the group A increased, and they decreased in some rats of group B. However, the two rats of group B with the highest mean rADC difference between occlusion and reperfusion (Fig. 6) still had high relative errors. Therefore, regardless of the reference group used for the estimation, a high difference in mean rADC between occlusion and reperfusion seems to impair the predictive value of the model.
3.
Discussion
In this study we provide a method for infarct volume prediction using MRI data obtained within the first hours after cerebral ischemia/reperfusion in rats. Infarct volume predictions were carried out from the ADC maps obtained at two time points, during and after MCAO, by using the ADC data either separately or combined. These maps are transformed into relative values using an ADC map obtained before MCAO. It seems possible to obtain relative values of ADC using other techniques; one possibility might be to use the contra-lateral
brain research 1583 (2014) 169–178
175
Fig. 6 – Infart volume estimated vs. rADC differences. Distribution of the different infarct volumes estimated with the mask in the T2 map at 7 days respect to the distance between the mean of the rADC values in occlusion and reperfusion, considering only the points in the infarcted volume. The circle diameter and the numbers inside the graphic indicate the relative error in the infarct volume estimation using the 2d probability.
hemisphere values. Measuring CBF can increase the quality of infarct volume estimations (Huang et al., 2010), but requires more acquisitions for the evaluation. The techniques for CBF quantification in rodents may not be widely available in different laboratories, or might not be sufficiently accurate due to low spatial resolution. Using the ADC data only, the sensitivity and specificity are slightly lower than when CBF data are included (Huang et al., 2010). However, the performance of the current estimations is comparable to those obtained using other methods with more complex and timeconsuming algorithms (Huang et al., 2011, 2010), which could be a limitation for their practical application in many experimental laboratories. Here, using ADC data only, we found that the sensitivity was 70% and the specificity 93% when information obtained during MCAO and after reperfusion was used for the estimations. The mean relative error of the infarct volume estimation by using the information from rADC data during MCAO and after reperfusion (35%) is also lower than the relative error obtained only considering the mean infarct volume of group A as predicted volume for all group B rats (50%). The prediction method used here is based on probability histograms created in a group of rats that were used as lookup tables for infarct estimation in other rats. Voxel-by-voxel comparisons were carried out after a registration process to ensure the correspondence between acquisitions. Non-rigid registrations were chosen to account for the fact that the lesion could displace the brain midline due to the presence of edema, since this effect modifies the internal geometry of the brain. We showed that the averaged result obtained from a set of rats could be reliably used to estimate the infarct volume of other rats subjected to the same experimental ischemia model. Since the procedure used here avoids the need to perform manual segmentation to evaluate the final infarct volume, the variability introduced by different users is
prevented and infarct volume prediction could be systematized easily. In general, the final infarct volume was better estimated by using the combined ADC data than by using only the information from a single time point. The estimation seems to be reasonably good in most of the rats excepting for those showing a large difference (40.3) between the mean rADC value at occlusion and that at reperfusion. Therefore, it seems that a big variation in the intensity of the ADC from occlusion to reperfusion impairs the estimation of infarct volume with this method. This variation might not be a frequent phenomenon, but taking it into account could contribute to assess the accuracy of the final volume estimation. In general, the prediction based on ‘ADC at reperfusion only’ is the one that differs more from the final infarct volume showing that a partial ADC recovery makes the estimated lesion at reperfusion less accurate. This corroborates earlier findings showing that reversal of ADC at early reperfusion does not invariably indicate final tissue viability (Neumann-Haefelin et al., 2000; Ringer et al., 2001; Rojas et al., 2006). Overall, the simplified prediction method reported here can provide evidence to show that the expected infarct volume is similar in different experimental groups. Then, the infarct volumes estimated could be compared to the final infarct volumes to evaluate whether treatments carried out after reperfusion may change the course of the prediction.
4.
Summary
This study showed the usefulness of ADC data at early stages during and after MCAO to estimate final infarct volume in rats. We compared the use of input ADC data at two time points (during MCAO and after reperfusion), either individually or combined. We computed the probability histograms in a group of rats, which were subsequently used as look-up
176
brain research 1583 (2014) 169–178
tables to carry out infarct estimation in a different group of rats. The accuracy of the infarct estimation was quantitatively evaluated by comparing it with the final infarct volume obtained from the T2 maps at day 7. Performance measurements showed better results by using information of the ADC during occlusion combined with the ADC after reperfusion, rather than by separately using either the occlusion or reperfusion data. In conclusion, we provide a simplified method, based on ADC maps obtained during and after transient ischemia, to predict final infarct volumes in a stroke model of 90 min intraluminal MCAO in rats.
5.
Experimental procedure
5.1.
Animals and surgery
The study was carried out in 11 adult male Wistar rats (Charles River, Lyon, France), weighing 250–300 g. Rats were housed in cages with controlled temperature (2171 1C) and humidity (55710%), under a 12 h light/12 h dark cycle and with food and water available ad libitum. All experiments were performed in accordance with the local legislation (Decret 214/1997 of July 30th by the ‘Departament d’Agricultura, Ramaderia i Pesca de la Generalitat de Catalunya’) under the approval of the Ethical Committee of the University of Barcelona (CEEA), and in compliance with European legislation. The surgery for intraluminal MCAO was carried out under isoflurane anesthesia, as previously described (Pérez-Asensio et al., 2010). The right MCA was occluded for 90 min. Then the thread was removed and reperfusion was allowed. This period of MCAO usually leads to striatal and cortical damage, but in some cases only striatal infarction is observed. Cortical brain perfusion was assessed by laser Doppler flowmetry and in all the cases occlusion induced a perfusion drop. The perfusion values normalized to the basal registry before surgery (100%) were 32711% and 120751% (mean7SD), during ischemia and reperfusion, respectively.
5.2.
Magnetic resonance imaging
MRI experiments were conducted on a 7T BioSpec 70/30 horizontal animal scanner (Bruker BioSpin, Ettlingen, Germany), equipped with a 12 cm inner diameter actively shielded gradient system (400 mT/m). The receiver coil was a phased-array surface coil for rat brain. Animals were placed in a supine position in a Plexiglas holder with a nose cone for administering anesthetic gases (2% isoflurane in a mixture of oxygen and N2O at a 30:70 ratio) and fixed by a tooth bar, ear bars and adhesive tape. Tripilot scans were carried out for accurate positioning of the animal's head in the isocenter of the magnet. MRI acquisitions were taken at 4 different time points in each rat: before MCAO surgery as the basal scan (prescan, ps), during MCAO (i.e. 60 min after starting the occlusion, occ 60 m), 60 min after reperfusion (rep 60 m), and 7 days after reperfusion (rep 7d).
5.2.1.
ADC maps
ADC scans were acquired during MCAO and 60 min after reperfusion using a pulsed gradient spin echo sequence with
the following acquisition parameters: echo time TE ¼55.07 ms, repetition time TR¼6250 ms, with a total acquisition time of 2 min 55 s, 6 b-values from 100 to 1000 s/mm2, a field of view FOV¼40 40 180 mm3, with a matrix size of 128 128 18 voxels and a resolution of 0.312 0.312 1 mm3/voxel. The same parameters were used for the basal acquisition.
5.2.2.
T2 relaxometry maps
T2 relaxometry scans were carried out 7 days after MCAO with a multi-slice multi-echo acquisition sequence with 16 effective echo times ramping from 11 to 176 ms, TR¼4926 ms, FOV ¼40 40 18 mm3, matrix size 256 256 18 voxels and resolution 0.156 0.156 1 mm3/voxel. The same parameters were used for the basal acquisition.
5.3.
Data analysis
ADC and T2 maps were calculated using Paravision 5.0 software (Bruker BioSpin, Ettlingen, Germany). Later these maps were processed with custom-made algorithms programmed in Matlab (The MathWorks, Inc, Natick, MA, USA) and ImageJ (National Institutes of Health, Bethesda, MD, USA). A binary mask was manually drawn over the T2 map in order to segment the brain from the background in the different acquisitions. Also, a threshold of 275 ms was applied to the T2 maps to reduce the noise. Then, all the maps acquired at different time points for each rat were co-registered to perform a voxel-based analysis.
5.3.1.
Registration process
A registration process was carried out for voxel-by-voxel comparisons between volumes, ensuring correspondence between the same points in images originally showing altered volumes due to brain edema. Non-rigid registration was performed by using Elastix 4.3 (Klein et al., 2010), a program based on Insight Segmentation and Registration Toolkit (ITK). The registration process used the basal scans before surgery as the reference images, whereas the acquisitions after MCAO were used as the moving images. A sequential registration, combining rigid and b-spline registrations, was carried out. Bilinear interpolation was performed to allow correspondence between the various resolutions of ADC and T2 maps. This interpolation also smoothed the images, which was convenient for the subsequent processing.
5.3.2.
ADC probability histogram
Following co-registration, ADC and T2 values at the various time points were used to compute the probability histograms. For each rat, the brain was segmented from the background in all the images using the mask obtained from the basal T2 map. Then a uniform smoothing filter of 3 3 pixels was applied to the original ADC and T2 maps. The relative values of ADC obtained 60 min after MCAO (rADCocc) and ADC at 60 min of reperfusion (rADCrep) were calculated by normalizing to the basal ADC values (obtained before MCAO). The final infarct volume was determined from the difference (ΔT2) between the T2 maps obtained 7 days after reperfusion
brain research 1583 (2014) 169–178
(T2rep 7d) and the corresponding basal T2 map obtained before MCAO (T2ps). For the relative ADC maps, Eq. (1) was used: rADCx ¼ 1
ADCx ADCps
ð1Þ
This calculation gives the relative difference of the ADC values (rADC) for a given time point in relation to the basal acquisition. rADC increases above 0 when the ADC value decreases. A segmentation mask was manually applied to the ΔT2 maps to select the voxels corresponding to infarcted tissue taking as reference intensity differences in the image above 20 ms to delimitate the affected area. This method was chosen for the segmentation, instead of a threshold directly applied to the T2rep 7d image, to avoid selecting noise areas as false infarcted voxels. With this data, a table was computed relating the rADCocc and rADCrep values with the infarct at 7 days (ΔT2). Data representation was carried out with two types of histograms pooling data from a group of rats (group A, see below). First showing the distributions of rADCocc and rADCrep values for the whole brain, and secondly calculating such distributions considering only the infarcted points. The histograms bins were selected of a 0.02 span sampling of the rADC values from 1 to 1. Afterwards, we obtained a probability histogram by dividing the histogram corresponding to the infarcted points by that of the whole brain. Thus, the probability histogram illustrates the proportion of infarcted voxels for each relative ADC value. Two-dimensional histograms were obtained by combining the information of the rADC at occlusion and after reperfusion. A histogram was built with rADCocc values in the x axis and rADCrep values in the y axis and taking into account all the voxels of the whole brain volume. A second histogram was computed in a similar way, but only considering the points that were marked as injured by the ΔT2 mask. From the ratio of both histograms, a new two-dimensional probability histogram was generated showing the proportion of infarcted points for each combination of rADC values. This 2d histogram was also processed to optimize the probability map. A median filter was applied on the two-dimensional probability histogram to eliminate the isolated points of the histogram. This procedure smoothed the probability map providing a more continuous distribution and avoiding the points representing only one specific rat and not an average behavior. Considering that the median filter would also smooth the average probability values, only a 3 3 median filter was used in order to make these differences small and no significant.
5.3.3.
Data analysis
An anatomical probability volume was generated from the rADC maps of each rat, using the probability histogram as a look-up table to obtain an estimation of the final infarct volume. The rats used in the study were separated into two groups: group A comprised the rats (rat 1 to rat 5) used to compute the global probability histograms; and group B was the test group containing rats (rat 6 to rat 11) that were used for infarct estimation by the general probability histograms obtained with the rats of group A. To evaluate the performance of the different probability
177
histograms, we compared the estimated infarct volume with the corresponding real infarct volume. We also performed a specificity and sensitivity analysis, which is useful for evaluating different classifiers. The statistical analysis to compare rat groups A and B was carried out with the Mann-Whitney test. The Kruskal–Wallis test compared the results obtained from the different probability histograms. When significant overall differences were found, a post-hoc analysis was performed with Dunn's multiple comparison test. All data were analyzed with GraphPad (GraphPad Software, Inc., CA, USA) and differences were considered significant if po0.05.
Acknowledgments CIBER-BBN is an initiative funded by the VI National R&D&i Plan 2008–2011, Iniciativa Ingenio 2010, Consolider Program, CIBER Actions and financed by the Instituto de Salud Carlos III with assistance from the European Regional Development Fund. Financed in part by the European Community (FP7/ 2007–2013; grant agreement number 201024), and the Spanish Government (SAF2009–08076, SAF2011-30492, ISCIII PS09/ 00527 and CDTI-CENIT/AMIT).
r e f e r e nc e s
Arenillas, J.F., 2002. Prediction of early neurological deterioration using diffusion- and perfusion-weighted imaging in hyperacute middle cerebral artery ischemic stroke. Stroke 33, 2197–2205. Aspey, B.S., Taylor, F.L., Terruli, M., Harrison, M.J., 2000. Temporary middle cerebral artery occlusion in the rat: consistent protocol for a model of stroke and reperfusion. Neuropathol. Appl. Neurobiol. 26, 232–242. Bra˚tane, B.T., Bastan, B., Fisher, M., Bouley, J., Henninger, N., 2009. Ischemic lesion volume determination on diffusion weighted images vs. apparent diffusion coefficient maps. Brain Res. 1279, 182–188. Dittmar, M., Spruss, T., Schuierer, G., Horn, M., 2003. External carotid artery territory ischemia impairs outcome in the endovascular filament model of middle cerebral artery occlusion in rats. Stroke 34, 2252–2257. Farr, T.D., Wegener, S., 2010. Use of magnetic resonance imaging to predict outcome after stroke: a review of experimental and clinical evidence. J. Cereb. Blood Flow Metab. 30, 703–717. Huang, J., Kim, L.J., Poisik, A., Pinsky, D.J., Connolly Jr., E.S., 1999. Titration of postischemic cerebral hypoperfusion by variation of ischemic severity in a murine model of stroke. Neurosurgery 45, 328–333. Huang, S., Shen, Q., Duong, T.Q., 2010. Artificial neural network prediction of ischemic tissue fate in acute stroke imaging. J. Cereb. Blood Flow Metab., 1–10. Huang, S., Shen, Q., Duong, T.Q., 2011. Quantitative prediction of acute ischemic tissue fate using support vector machine. Brain Res. 1405, 77–84. Kanemitsu, H., Nakagomi, T., Tamura, A., Tsuchiya, T., Kono, G., Sano, K., 2002. Differences in the extent of primary ischemic damage between middle cerebral artery coagulation and intraluminal occlusion models. J. Cereb. Blood Flow Metab. 22, 1196–1204. Klein, S., Staring, M., Murphy, K., Viergever, M.A., Pluim, J.P.W., 2010. Elastix: a toolbox for intensity-based medical image registration. IEEE Trans Med. Imag. 29, 196–205.
178
brain research 1583 (2014) 169–178
Leithner C., Fu¨chtemeier M., Mu¨ller S., Foddis M., Jorks D., Dirnagl U., Royl G. 2011. Can early MRI be used to improve the statistical quality of experimental stroke studies? In: Brain 2011 XXVth International Symposium on Cerebral Blood Flow, Metabolism, and Function, and Xth International Conference on Quantification of Brain Function with PET. Barcelona, Spain. Lu, M., Mitsias, P.D., Ewing, J.R., Soltanian-Zadeh, H., BagherEbadian, H., Zhao, Q., Oja-Tebbe, N., Patel, S.C., Chopp, M., 2005. Predicting final infarct size using acute and subacute multiparametric MRI measurements in patients with ischemic stroke. J. Magn. Res. Imag. 21, 495–502. Macrae, M.I., 1992. New models of focal cerebral ischaemia. Br J. Clin Pharmacol. 34, 302–308. McAuley, M.A., 1995. Rodent models of focal ischemia. Cerebrovas. Brain Metab. Rev. 7, 153–180. Neumann-Haefelin, T., Kastrup, A., de Crespigny, A., Yenari, M.A., Ringer, T., Sun, G.H., Moseley, M.E., 2000. Serial MRI after transient focal cerebral ischemia in rats: dynamics of tissue injury, blood-brain barrier damage, and edema formation. Stroke 31, 1965–1972. Oliff, H.S., Weber, E., Eilon, G., Marek, P., 1995. The role of strain/ vendor differences on the outcome of focal ischemia induced by intraluminal middle cerebral artery occlusion in the rat. Brain Res. 675, 20–26. Pe´rez-Asensio, F.J., De la Rosa, X., Jime´nez-Altayo´, F., Gorina, R., Martı´nez, E., Messeguer, A., Vila, E., Chamorro, A., Planas, A. M., 2010. Antioxidant CR-6 protects against reperfusion injury after a transient episode of focal brain ischemia in rats. J. Cereb Blood Flow Metab. 30, 638–652. Prieto, R., Carceller, F., Roda, J.M., Avendan˜o, C., 2005. The intraluminal thread model revisited: rat strain differences in local cerebral blood flow. Neurol Res. 27, 47–52. Ringer, T.M., Neumann-Haefelin, T., Sobel, R.A., Moseley, M.E., Yenari, M.Y., 2001. Reversal of early diffusion-weighted magnetic resonance imaging abnormalities does not necessarily reflect tissue salvage in experimental cerebral ischemia. Stroke 32, 2362–2369.
Rojas, S., Martı´n, A., Justicia, C., Falco´n, C., Bargallo´, N., Chamorro, A., Planas, A.M., 2006. Modest MRI signal intensity changes precede delayed cortical necrosis after transient focal ischemia in the rat. Stroke 37, 1525–1532. Rosso, C., Colliot, O., Pires, C., Delmaire, C., Valabre`gue, R., Crozier, S., Dormont, D., Baillet, S., Samson, Y., Lehe´ricy, S., 2011. Early ADC changes in motor structures predict outcome of acute stroke better than lesion volume. J. Neurorad. 38, 105–112. Shen, Q., Ren, H., Fisher, M., Duong, T.Q., 2005. Statistical prediction of tissue fate in acute ischemic brain injury. J. Cereb. Blood Flow Metab. 25, 1336–1345. Shen, Q., Duong, T.Q., 2008. Quantitative prediction of ischemic stroke tissue fate. NMR Biomed. 21, 839–848. Sutherland, G.R., Dix, G.A., Auer, R.N., 1996. Effect of age in rodent models of focal and forebrain ischemia. Stroke 27, 1663–1667. Thomalla, G.J., Kucinski, T., Schoder, V., Fiehler, J., Knab, R., Zeumer, H., Weiller, C., Ro¨ther, J., 2003. Prediction of malignant middle cerebral artery infarction by early perfusion- and diffusion-weighted magnetic resonance imaging. Stroke 34, 1892–1899. Walberer, M., Stolz, E., Mu¨ller, C., Friedrich, C., Rottger, C., Blaes, F., Kaps, M., Fisher, M., Bachmann, G., Gerriets, T., 2006. Experimental stroke: ischemic lesion volume and edema formation differ among rat strains (a comparison between Wistar and Sprague-Dawley rats using MRI). Lab Anim. 40, 1–8. Wu, O., Sumii, T., Asahi, M., Sasamata, M., Ostergaard, L., Rosen, B.R., Lo, E.H., Dijkhuizen, R.M., 2007. Infarct prediction and treatment assessment with MRI-based algorithms in experimental stroke models. J. Cereb. Blood Flow Metab. 27, 196–204. Zarow, G.J., Karibe, H., States, B.A., Graham, S.H., Weinstein, P.R., 1997. Endovascular suture occlusion of the middle cerebral artery in rats: effect of suture insertion distance on cerebral blood flow, infarct distribution and infarct volume. Neurol. Res. 19, 409–416.
