Depiction of Infarct Frequency Distribution by Computer-Assisted Image Mapping in Rat Brains With Middle Cerebral Artery Occlusion
Comparison of Photothrombotic and Intraluminal Suture Models
Background and Purpose Histopathologic analysis of experimental brain damage has traditionally been performed by measuring areas of infarction and/or selective neuronal alterations on a section-by-section basis in individual animals. For series containing multiple replicate animals, quantitation of tissue injury is typically performed at similar coronal levels throughout an experimental group. A means of facilitating pictorial group comparisons of these histopathologic alterations between different series of replicate studies is highly desirable.
Methods We introduce a newly designed approach to achieve this goal, based on a linear affine transformation that is used to map corresponding sections at the same anatomic level into a common template to yield a frequency distribution map depicting the aggregate data set. We have applied this approach to compare the histopathologic features of two models of middle cerebral artery (MCA) occlusion in rats: (1) photothrombotically induced permanent distal MCA occlusion in spontaneously hypertensive rats (SHR) and (2) temporary MCA occlusion by intraluminal suture in Wistar rats.
Results The brains of SHR rats with permanent distal MCA occlusion showed a high frequency of infarction involving the dorsolateral and lateral portions of the ipsilateral neocortex, whereas Wistar rats with 90-minute MCA suture occlusion showed a zone of infarction largely concentrated in the dorsolateral portion of the ipsilateral caudoputamen. Infarct frequency distributions for the two animal groups were compared statistically at three corresponding anatomic levels by Fisher’s exact test; the resulting statistical parametric maps are shown.
Conclusions With the use of frequency distribution maps, the pattern of trends within a group can be observed coronally or three-dimensionally. One can directly access data as to numbers of rats with infarction for any point on the map. Studies performed under different experimental conditions can also be compared with one another by means of the generated data sets.
Histopathologic analysis of perfusion-fixed, paraffin-embedded brain material has long been regarded as the “gold standard” for the quantitative assessment of tissue injury in experimental models of focal cerebral ischemia.1 Traditionally, this has been accomplished by measuring areas of pannecrosis and/or selective neuronal alterations in stained coronal sections of individual animals; numeric integration of sequential areas then yields the total affected volumes.1 In many experimental settings, one wishes to compare lesion areas or volumes across multiple replicate animals, eg, control versus ischemic groups or untreated animals versus those treated with putative neuroprotective agents.2 3 As part of that analysis, it is highly desirable to have a pictorial means of identifying which tissue regions differ from one experimental group to another. Finally, the ability to display mean trends in the data sets and to obtain 3-D representations of his topathologic injury would further enhance the power of data analysis because such images could be analyzed in conjunction with other pictorial data sets obtained, for example, from brain atlases or from autoradiographic studies of local cerebral blood flow or glucose utilization in comparable animals.4
In this report we introduce a novel method for the histopathologic analysis of experimental cerebral infarction in the rat, and we demonstrate its application to the pathophysiological study of two different models of focal cerebral ischemia.
A common approach for analyzing patterns of focal cerebral infarction in paraffin-embedded histological material is to use a camera lucida attached to a light microscope to trace section contours and to outline tissue regions affected by infarction and/or selective ischemic neuronal changes. These drawings can then be computer digitized. In an experimental series consisting of N replicate brains studied at a particular anatomic level, there are thus N corresponding sections for which the extracted morphological information can be rendered in a binary digital image format by assigning a value of 1 to the image contour and to pathologically affected regions and a value of 0 to unaffected areas and background. We wish to deform these individual sections so that they can each be mapped into a standard predefined anatomic template to study the aggregate distribution of infarction or ischemic cell change at specific anatomic locations.
Several existing image-mapping and image-registration methods have been proposed to handle different circumstances.5 Among these schemes are feature-matching–based methods, as well as correlation-based and moment-based methods. The matching of feature points appears to be an accurate way of mapping or registering image sections. Feature-matching methods (also termed control-point methods) search for special edge or pixel landmarks representing a particular feature in two image sections. This method works well if control points can be found. However, it is a difficult task to select the same feature points from two image sections. Usually, an operator must choose a set of feature points from one section and compute the local cross-correlation to determine the corresponding feature points on the other image section. If fewer feature points can be extracted from both image sections, this approach becomes less robust. Correlation methods have long been used for image mapping and image registration.6 7 Cross-correlation is a measure of the similarity of two images in terms of their spatial distributions of gray-level values (image intensities); hence, pixel-by-pixel calculation is required on two mapped image planes. A moment-based method handles image registration only.
We have developed and validated a novel method for the mapping of image sections that overcomes most of the difficulties experienced by existing mapping/registration methods.8 9 This new method, termed disparity analysis, is based on a linear affine model to analyze point-to-point disparities in two images. It is a direct method that estimates scaling, translation, and rotation parameters simultaneously without transformation. The disparity approach is computationally fast, and it takes into consideration shape differences of coronal sections. It is a general and flexible method that uses the same basic principle to deal with different situations, such as damaged or asymmetrical sections. The method can be applied to boundaries and/or image intensities (optical densities). In the case of mapping histological sections, we use section-boundary information. Below, we briefly describe the method and its application to the comparative analysis of histopathology in two models of focal cerebral ischemia in the rat. The details of the mathematical derivation can be found in a previous publication.8 Considering that the morphological information has been extracted and converted to a digital image format, we shall focus the following discussion on image planes.
We assume a linear relationship between corresponding points on two mapped section contours, as follows: where A is a 2×2 affine matrix; x, a point on the boundary of a coronal histological section, is defined as xT=(x, y); and vectors x′ and c are defined in a similar manner.
The physical meaning of this linear affine model is as follows: 2-D translation is represented by the vector c, and the affine matrix A is responsible for 2-D rotation and 2-D linear shape changes. In fact, the affine matrix A can be uniquely decomposed into 2-D rotation, 2-D angular deformation (shearing), and height and width changes10 : In other words, for a given A, the 2-D rotation angle θ, the deformation angle α, and the height and width changes (scale factors) lx and ly can be computed uniquely from the matrix elements. More terms could be added to Equation 1, eg, higher-order terms, to form a nonlinear relationship7 between matched sections. We note that shape differences between corresponding histological sections from different animals are limited; thus, a linear model is appropriate, and adding higher-order terms would only complicate the computation. This model can be applied not only to section matching but also to section registration for 3-D reconstruction as well. In the case of section registration (alignment), we take A=Ar and c computed from the two sections. If section mapping (matching) is required, we take the general form of A and vector c. As long as the matrix A and vector c are known, registration and mapping can then be easily accomplished by this linear transformation. Both registration and mapping calculations use the same computational process.
Coronal sections from ischemic brains may have size differences between hemispheres caused by brain edema, atrophy, or infarction. One can calculate transformation parameters, A and vector c, for each hemisphere and map the two hemispheres separately. To validate the applicability of the affine transformation model under those conditions, we performed the following simulation: We first digitized a brain atlas11 as our reference template (Fig 1A⇓). We then defined two anatomic areas (Par1 and CPu) by filling their boundaries according to the atlas (Fig 1B⇓). To simulate brain edema, deformation (simulated swelling) was artificially applied eccentrically to the right hemisphere. This size and shape change is obvious when the atlas diagram is superimposed (Fig 1C⇓). Finally, we applied affine transformation to map the deformed section back into the reference template using section contour information only. (This is not a simple reverse transformation because the simulated swelling was induced eccentrically, whereas the transformation coordinates are centrally located in the cerebral hemisphere.8 9 ) As shown in Fig 1D⇓, the designated anatomic regions of the previously deformed hemisphere fall within the expected boundaries of the atlas diagram. Quantitatively, the initial sizes of the two designated anatomic areas before deformation were 23.2 (Par1) and 18.8 (CPu) mm2. After artificial deformation, these values became 25.3 and 22.9 mm2, respectively. After affine transformation was applied, the two areas were restored to 22.4 and 19.9 mm2, respectively, ie, to within 4% to 6% difference of their initial values. This example illustrates that the affine transformation model yields a tolerable error even in the presence of hemispheric size alterations of 9% to 22% produced by disease. If the mapped sections differed from the template by more than this amount, a linear affine transformation model might not be suitable. We are currently working on nonlinear transformation matching strategies to overcome larger regional distortions caused by edema or atrophy.
Materials and Methods
We have applied this approach to analyze histological data derived from two studies that used different models of MCA occlusion in anesthetized rats.
Series 1: Photothrombotic Distal MCA Occlusion
This procedure was performed in 8 male SHR (weight, 270 to 345 g). Anesthesia was induced with 4% halothane and maintained with 0.5% halothane in a 70%/30% mixture of nitrous oxide and oxygen delivered through an endotracheal tube. Vascular catheters were inserted into the femoral arteries and veins. Arterial blood pressure was monitored continually and recorded on a polygraph, and arterial blood gases were periodically measured. Rats were ventilated mechanically on a small-animal respirator. Pancuronium bromide (initial dose, 0.35 mg/kg IV, followed by 0.1 mg/kg every half hour) was administered for muscle relaxation. Rectal temperature was maintained at 37°C by a thermostatically regulated heating pad. The temperature of the right temporalis muscle was monitored with a thermistor and was maintained at 36°C by a servo-controlled heating lamp placed near the head.
Rats were then mounted in a stereotaxic head-holder, and a 1.5-cm vertical incision was made between the right eye and ear. With the aid of a Zeiss operating microscope, a burr hole approximately 3 mm in diameter was made with a high-speed drill 1 mm rostral to the anterior junction of the zygomatic and squamosal bones. Care was taken not to injure the dura mater. The distal segment of the right MCA was thus exposed above the rhinal fissure.1 12 The photosensitizing dye rose bengal (15 mg/mL) was first administered intravenously in a dose of 20 mg/kg over a 90-second interval. Immediately thereafter, the distal MCA was occluded photochemically at three different points by simultaneous irradiation with a 562-nm beam from an argon laser–activated dye laser (Coherent, Inc) at a power of 20 mW, focused on the MCA. Details of this method have been recently summarized.13
Series 2: Reversible MCA Occlusion by Intraluminal Suture
For these studies, fasted male Wistar rats (n=17) were anesthetized with 3.5% halothane in a 70%/30% mixture of nitrous oxide and oxygen. Animals were orotracheally intubated and ventilated mechanically. Heating lamps were used to maintain rectal and temporalis muscle temperatures at 37°C to 38°C. Femoral vessels were cannulated, and mean arterial blood pressure, arterial blood gases, and plasma glucose were measured. A ventral midline incision was made in the neck, and the right carotid arterial system was identified. A 3-0 nylon suture, blunted at the tip by heating near a flame and coated with a solution of 0.1% poly-l-lysine, was introduced retrogradely through the external carotid artery into the internal carotid artery and MCA according to the procedure of Zea Longa et al.14 The suture was inserted 18 to 20 mm from the bifurcation of the common carotid artery according to the animal’s weight. Animals were allowed to awaken after MCA occlusion and were evaluated for neurological status by a standard behavioral battery at 60 minutes. These tests assessed forelimb placement to visual, tactile, and proprioceptive stimuli and assessed postural reflexes on tail suspension. Rats that failed to exhibit neurological abnormalities were excluded from further study. At 90 minutes after MCA occlusion, rats were reanesthetized with halothane and the suture was removed.
Three days after MCA occlusion, rats of each series were reanesthetized with halothane and perfused transcardially at a pressure of 120 mm Hg with FAM (40% formaldehyde/glacial acetic acid/methanol, 1:1:8 [vol]) for 20 minutes after a brief saline wash. Brains were left in situ overnight at 4°C before removal. Coronal brain blocks were then processed for paraffin embedding. Coronal sections 10 μm in thickness were cut and stained with hematoxylin and eosin for histopathological analysis. To quantify infarct volume, coronal sections were viewed microscopically at low power (×1), and the section contours as well as the outline of the zone of cerebral infarction were traced onto paper with the aid of a camera lucida microscope attachment. These drawings were then video digitized and saved as digital images. A value of 1 was assigned to each pixel inside the traced infarct region and on the section contour, and the remainder of the pixels were given a value of 0. All computational procedures were performed on a MicroVAX 3600 (32 MB RAM); image display and analysis were performed on a VAX Station 3200 (8-bit color plane, 16 MB RAM; Digital Equipment Corp).
Histopathologic Findings in Series 1
The brains of SHR with permanent distal photothrombotic MCA occlusion (for 3 days) were remarkable for a confluent zone of pannecrosis involving the dorsolateral and lateral portions of the ipsilateral frontoparietal neocortex but sparing subcortical structures. The infarct volume (stereotaxically corrected for shrinkage during tissue processing) averaged 37.0±9.7 mm3 (SD). Fig 2A⇓ shows digitized drawings of the infarct area at the coronal level bregma −1.3 mm (level of fimbria15 ) in the 8 rats of this series. The mean infarct area at this level was 8.84±3.6 mm2 (SD).
Histopathologic Findings in Series 2
Wistar rats with 90-minute MCA occlusion followed by recirculation for 3 days showed a zone of infarction intermixed with selective ischemic neuronal alterations largely involving the dorsolateral portion of the ipsilateral caudoputamen, with inconstant involvement of the overlying lateral and dorsolateral frontoparietal neocortex. Total infarct volume averaged 149.8±49.7 mm3 (SD). Fig 2B⇑ depicts the distribution of pathological changes in digitized drawings at coronal level bregma −1.3 mm (level of fimbria15 ) in 9 of the 17 rats of this series. The mean lesion area at this level in the 17 rats was 30.4±22.7 mm2 (SD). The volume of infarction was calculated by numerical integration by an investigator who was blinded to the experimental groups.
3-D Maps of Lesion Distribution
Frequency Maps of Lesion Distribution
Fig 4⇓ illustrates the ability of the computer-assisted image-mapping method to yield maps revealing the frequency distribution of infarction. To generate these maps, the corresponding digitized coronal drawings obtained in the individual rats of each series (see Fig 2⇑) were mapped into a common anatomic template (one preselected section from the experimental group) by linear affine transformation (see “Theory”), and pixel-by-pixel based summation was performed. These images, shown for series 1 and 2 in Fig 4⇓, clearly reveal the frequency distributions of pathological changes and thus provide immediate anatomic information as to which loci are consistently versus less commonly affected by the respective focal ischemic insults. Fig 4⇓ also juxtaposes the corresponding atlas image at each coronal level, derived by digitizing the functional-anatomic brain atlas of Zilles.11 Superimposition of this digitized atlas image (by the same mapping method) would thus reveal the reference anatomic loci affected by the ischemic process.
Statistical Comparison of Series 1 and 2
To illustrate the capability of this method to yield statistical insights on comparative lesion distribution between the two series, the following procedures were performed: First, the digitized images of series 1 and 2 were mapped, at each coronal level of interest, into a common anatomic template. Next, Fisher’s exact test16 was used to compare the two series at each image-pixel location. Fig 5⇓ shows the resulting statistical parametric map of 1−P, where P represents the statistical level of interseries difference computed from Fisher’s exact test. These images immediately reveal that the major differences in lesion distribution between the series are attributable to the lack of subcortical involvement in series 1 and the relatively less extensive involvement of neocortex in series 2, in which MCA occlusion was temporary rather than permanent.
In this study we have applied a novel method of affine transformation to map regional histopathologic data into a common anatomic template. With the use of image-mapping strategies, anatomic maps of infarct frequency distribution could be produced. The stacking of these closely spaced maps in turn provides a means of visualizing patterns of ischemic injury in 3-D space. From these computer maps, one can directly access data on numbers of rats showing damage for any image pixel. These maps also permit areas and volumes of injury to be directly computed. In addition, these data sets provide a means of performing statistical comparisons of lesion distribution between two series, by means of standard statistical procedures applied on a pixel-by-pixel basis.
This method, in our view, has great generalized utility. Potential applications would include the following: (1) the ability to discern those anatomic regions significantly protected by anti-ischemic neuroprotective agents17 (comparison of treated versus untreated rats); (2) pixel-by-pixel comparison of histopathologic images with autoradiographic data sets of local cerebral blood flow or glucose utilization derived from studies in replicate animal groups4 ; and (3) the means to analyze functional-anatomic zones affected by local ischemic lesions by using the atlas-overlay capability of this method (Fig 4⇑). Each of these applications is currently being implemented in our laboratory.
Selected Abbreviations and Acronyms
|MCA||=||middle cerebral artery|
|SHR||=||spontaneously hypertensive rats|
This study was supported by National Institutes of Health grants NS-05820 and NS-23244. We wish to thank Drs Brant D. Watson and Hiroshi Yao for their expertise in applying the photothrombotic model of MCA occlusion and Susan Kraydieh for providing superb technical assistance.
Reprint requests to Myron D. Ginsberg, MD, Department of Neurology (D4-5), University of Miami School of Medicine, PO Box 016960, Miami, FL 33101. E-mail email@example.com.
- Received August 9, 1995.
- Revision received January 19, 1996.
- Accepted January 22, 1996.
- Copyright © 1996 by American Heart Association
Markgraf CG, Kraydieh S, Prado R, Watson BD, Dietrich WD, Ginsberg MD. Comparative histopathologic consequences of photothrombotic occlusion of the distal middle cerebral artery in Sprague-Dawley and Wistar rats. Stroke.. 1993;24:286-293.
Yao H, Ginsberg MD, Watson BD, Prado R, Dietrich WD, Kraydieh S, Busto R. Failure of MK-801 to reduce infarct volume in thrombotic middle cerebral artery occlusion in rats. Stroke.. 1993;24:864-871.
Altmann J, Reitbock HJP. A fast correlation method for scale- and translation-invariant pattern recognition. IEEE Trans Pattern Anal Machine Intelligence.. 1984;6:46-57.
Maguire GQ Jr, Noz ME, Lee EM, Schimpf JH. Correlation methods for tomographic images using two or three dimensional techniques. In: Program and Abstracts of the Ninth International Conference on Information Processing in Medical Imaging; October 14-16, 1985; Washington, DC; 266-279.
Young TY, Wang YL. Analysis of three dimensional rotation and linear shape changes. Pattern Recognition Lett.. 1984;2:239-242.
Zilles K. The Cortex of the Rat. Heidelberg, Germany: Springer-Verlag; 1985.
Watson BD, Dietrich WD, Prado R, Nakayama H, Kanemitsu H, Futrell NN, Yao H, Markgraf CG, Wester P. Concepts and techniques of experimental stroke induced by cerebrovascular photothrombosis. In: Ohnishi ST, ed. Membrane-Linked Diseases: Volume 4: CNS Trauma: Laboratory Techniques and Recent Advancement. CRC Press. In press.
Zea Longa EL, Weinstein PR, Carlson S, Cummins R. Reversible middle cerebral artery occlusion without craniectomy in rats. Stroke.. 1989;20:84-91.
Paxinos G, Watson C. The Rat Brain in Stereotaxic Coordinates. Sydney, Australia: Academic Press; 1982.
Siegel S. Nonparametric Statistics for the Behavioral Sciences. New York, NY: McGraw-Hill Publishing Co; 1956.
Ginsberg MD. Neuroprotection in brain ischemia: an update: parts I and II. Neuroscientist.. 1995;1:95-103, 164-175.