| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Departments of 1Medical Biophysics and 2Oncology, University of Western Ontario, and 3London Regional Cancer Centre, London, Ontario, Canada
Submitted 30 October 2003 ; accepted in final form 16 September 2004
 |
ABSTRACT |
|---|
 TOP ABSTRACT MATERIALS AND METHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
microhemodynamics; fluorescence microscopy; video analysis
To quantify changes in microvascular perfusion in animal models of cancer and other diseases, it is necessary to be able to visualize directly the microcirculation in vivo and quantify functional microvascular geometry and blood flow (18). Although noninvasive methods, such as magnetic resonance imaging and ultrasound, have undergone tremendous improvements in temporal and spatial resolution, they remain limited in their ability to clearly resolve vascular structures on the order of diffusion distances (e.g. <200 µm for O2) in vivo and to assess microhemodynamics at the capillary level (8, 15). For these reasons, many groups have chosen in vivo video microscopy (IVVM) to assess microcirculatory function and O2 transport in various disease models (8, 15, 18, 24).
IVVM provides detailed observations of organs and tissues at the microcirculatory level in vivo and in real time. Contrast can be obtained through transillumination of tissues with visible light or through fluorescence excitation of intravenous contrast agents, such as FITC-labeled dextran (15, 18, 24). Because surgical exposure of the tissue is generally required for imaging, this can limit application to end-point studies. However, the development of window chamber models has allowed for serial observation of the same microvascular network over time (4, 15, 17, 21). In all these models, IVVM recordings of microvascular blood flow can be stored for postprocessing to assess microhemodynamic parameters, such as blood flow velocity and RBC supply rate (10, 19).
Although IVVM is capable of providing the desired temporal and spatial resolution, some inherent limitations include 1) restriction of chronic chamber preparations to nonvital organs and tissues in a wound-healing environment (18); 2) motion artifacts, caused by respiratory or cardiac motion, which complicate postprocessing and quantification of microvascular function, particularly with the use of upright microscopes; and 3) limited transillumination through thick tissues, even when an inverted microscope is used. Therefore, a means of directly recording video images of tissue microcirculation in vivo with resolution at the cellular level is needed as well as video analysis tools to permit quantitative assessment of structural and functional microhemodynamic parameters. In this study, we addressed these needs by developing techniques for producing computer-generated images to reveal plasma-perfused vessels, RBC-perfused vessels, and graded distribution of RBC supply.
FITC-labeled dextran was injected intravenously to reveal the plasma space and contrasting nonfluorescent RBCs within organs such as lung and liver. Digitized video sequences were recorded for selected microvascular networks. Physiological background movement was removed using computerized motion correction. Spatial and temporal variations in light intensities due to the passage of RBCs were analyzed to produce a number of distinct functional maps of the microcirculation. These images were used to assess microvascular morphology and function and were further analyzed using quantitative stereology. Differences in normal vasculature were apparent in different organs, and distinct differences in morphology and function were seen between the vessels of normal liver tissue and those of liver metastases.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTMATERIALS AND METHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
To obtain high-contrast views of organ microcirculation, high-molecular-weight (HMW, 250-kDa) fluorescently labeled FITC-dextran (Sigma Chemical, St. Louis, MO) was administered by tail vein injection (0.05 ml, 25 mg/ml). The HMW-FITC-dextran was rapidly distributed throughout the blood plasma volume and remained intravascular for up to 2–3 h (as observed experimentally in the mouse liver and lung) within intact vessels, allowing clear differentiation of microvessels in tissue. The organ of interest was then surgically exposed for viewing by IVVM. As an example, for imaging the liver, the organ was exposed by abdominal incision, and the animal was placed with the liver surface down on a viewing platform of an epifluorescence inverted microscope (Diaphot TMD, Nikon; Fig. 1). Tissue was episcopically illuminated (FITC-dextran; excitation = 490 nm, emission = 520 nm) and viewed using a Panasonic WV1550 Newvicon tube camera. Real-time video was monitored and recorded on SVHS videotape. Selected video sequences were digitized as AVI files at 30 frames/s with 640 x 480 frame resolution and no compression. Postprocessing of video sequences was performed using algorithms written in MATLAB (Mathworks, Natick, MA).
|
|
For Ii(x,y) equal to the pixel intensity at any location (x,y) of 640 x 480 pixel locations in video frame i of N frames
 |
 |
0
 |
Calculation of contrast-to-noise ratio.
To quantify the improvement in identification of functional vessels, changes in vascular contrast resulting from the production of variance images were characterized by the contrast-to-noise ratio: CNR = (
i – e)/
e, where is mean pixel intensity, is standard deviation of the pixel intensities, and subscripts i and e represent intra- and extracellular, respectively. Briefly, using a MATLAB script, equal-sized ROIs from a single frame of a stabilized video sequence were selected from within a vessel and from extravascular tissue adjacent to that vessel. Corresponding ROIs from the variance image based on the sequence were automatically selected by the program. The difference between the mean intravascular pixel intensity and the mean extravascular pixel intensity was used to characterize the vascular contrast in the image, and the standard deviation of the extravascular pixel intensities was used to characterize the level of background noise. The ratio of these values provides an objective measure of perceived vascular contrast in the images.
Nonuniform background correction. Nonuniformity in fluorescence illumination resulted in a brighter image in the center of the field of view. Because variance images reflect the range of light intensities rather than the mean level, they effectively flatten the background brightness. This change was quantified using a MATLAB script performing the following steps. Images were divided into 32 x 32-pixel subsectors. The minimum pixel intensity within each subsector was recorded and mapped to a new image matrix. For images in which bright features of interest are displayed on a dark background, the new image represents an approximation of the background intensity values across the image. This "background map" was then scaled to the original image size with use of standard interpolation methods. The standard deviation of pixel intensity values within the background map was then calculated.
Stereological analysis. Quantitative stereology was used to determine the volume fraction of vessels in the tissue (VV), functional microvessel density (fMVD, JV), branch-point density (BPV), and mean vessel segment length (SL) for tumor tissue and adjacent normal liver tissue.
To measure VV, a 7 x 12-point grid was superimposed on the variance images. The fraction of the grid points falling on vascular structures (PP) was used to estimate area fraction and, thus, as described by the Delesse principle, vascular volume fraction (26). Because of the boundary conditions at the surface of the tissues (normal liver and tumors), vessels seen by IVVM were treated as cylinders lying parallel to the surface. Thus, to correct for overestimation of vascular volume due to the Holmes effect, VV was calculated as PP/[1 + (4t/
D)], where t is the section thickness (20 µm, as empirically determined by optically "slicing" through tissue) and D is average vessel diameter (10 µm, as sampled from the image). The volume overestimate (4t/D) for cylinders was based on the same principle applied to spheres (3t/2D), as described by Weibel (26).
To quantify fMVD, the length of functional vessels (defined as RBC-perfused vessels visible in the variance image) per unit volume of tissue was calculated. A skeleton map of the midlines and branch points of the vessels was created, and vessel length density was calculated as JV = 4IL/t, where IL is the number of intersections of vessels per unit length of randomly oriented test lines superimposed on the image. This measurement was based on the standard relation for randomly oriented vessels [JV = 2QA, where QA is the number of transections per unit area (26)] but was adjusted for measurement of vessels lying parallel to the surface and transected by a plane normal to the surface with an area equal to L x t. The density of branch points was calculated simply as BPV = BPA/t, with BPA measured as the number of vessel bifurcations visible within the sample area, and inasmuch as there are three vessel segments for every two branch points, mean SL was calculated as follows: SL = 2JV/3BPV.
Spatiotemporal analysis.
RBC velocities in individual RBC-perfused vessels were assessed by analysis of space-time images (STIs) (10). For each video sequence, sample lines were manually superimposed on the centerlines of selected vessels located within the field of view. Pixel intensities (If,p) at incremental positions (p = 1 to length) along each line were extracted for each frame of the video sequence (f = 1 to number of frames) and used to create an STI that was fmax columns wide by pmax rows high. The pattern of dark pixels in a column of the STI corresponded to the positions of RBCs within the vessel at that time. Movement of the RBCs resulted in a vertical shift in the pattern for successive columns, producing diagonal bands in the STI. The slope of the bands (
position/frame) was used to calculate the RBC velocity. For this study, RBC velocity in selected vessels was based on an average of five sampled velocity values over the length of the video sequence.
|
RESULTS |
|---|
|
TOPABSTRACTMATERIALS AND METHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
10 µm deeper into the tissue showing RBC-perfused alveolar capillaries.
|
|
|
Vascular contrast enhancement and background correction using variance mapping.
Figure 5A illustrates the low contrast between the vessels and the surrounding tissues that could occur in the single-frame fluorescence images, especially where there was a high density of RBCs in the vessels or where nonuniform illumination varied the brightness of the background. For this image, the CNR was 2.0 ± 1.0, and after blurring to remove local variations in light intensity (Fig. 6A), the standard deviation of gray-scale readings throughout the image was 40.0. Although the maximum image (Fig. 5B) improved the brightness of the vessels, nonuniformity of the background was still evident. Because temporal variation in background intensity was low throughout the sequence, it was uniformly dark in the variance image (Fig. 5C), with an 10-fold reduction in standard deviation to 4.2 (Fig. 6B). The increased vessel brightness against the uniformly dark background of the variance image resulted in a 12-fold increase in CNR to 24 ± 6.
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTMATERIALS AND METHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
Although IVVM provided high-contrast video sequences that could be viewed directly to form qualitative impressions of microvascular morphology and function in normal and diseased tissues, the real power of the technique lay in the postprocessing of those sequences to provide quantitative measurements. To limit the effects of image noise due to physiological motion on postprocessing techniques, motion correction was applied before any other pixel-based analyses. The registration algorithm used for this study proved effective for eliminating residual background motion in the tissues we examined, resulting in stable video sequences suitable for subsequent postprocessing routines.
Functional maps of the microcirculation from video sequences. By generating maps of the microvasculature based on quantitative changes in light intensity at each image location by means of transillumination of thin tissues (11, 14, 16), subjective impressions of RBC motion within microvessels may be quantified and presented as single images (maximum-intensity image, variance image, perfusion image, and difference image). The use of FITC-labeled dextran and epifluorescence illumination to provide plasma-tissue and plasma-RBC contrast extends the utility of the above-described video-processing techniques to thick tissues also. Although contrast marker-based plasma perfusion has been used previously to characterize microvessel function (15, 21, 24), measurements based solely on plasma perfusion overlook the heterogeneities in local RBC distributions that may be present and, therefore, cannot reflect microvascular transport of O2 to tissues.
The maximum-intensity image revealed the lumens of all plasma-perfused vessels, thus mapping the geometry of the patent microvasculature. Most vessels in Fig. 5B were lighter than those in Fig. 5A because of the motion of plasma gaps past each location and in wider vessels where the lumen was clearly defined. Only where RBCs were stationary (arrow) or where the plasma gaps were insignificant did the intensity remain persistently low. Vessels in normal liver tissue exhibiting RBC motion (Fig. 5C) are clearly identified by their high variance, which incidentally includes most of the vessels in the field of view, as represented by the high degree of overlap in the perfusion image (Fig. 5D). Although variance imaging is ideally suited to capillary blood flow, where the RBCs are tightly confined within the vessel walls with plasma gaps only between RBCs, liver sinusoids were often of greater diameter and, in many cases, exhibited flow with distinct axial accumulation. In this case, although the variance image would not reveal the full vessel width, the maximum image would. For the tumors, despite the high vessel density (Fig. 7A), only a few vessels showed RBC motion (Fig. 7B). This was consistent with the data of Dewhirst et al. (7), who used a dorsal skin flap chamber model to study tumor blood flow and estimated that up to 9% of tumor vessels showed plasma perfusion but little or no RBC perfusion.
Although variance images revealed vessels with RBC motion and, thus, the potential for O2 delivery, they were not sensitive to the number or velocity of RBCs that pass a particular location (16). Vessels containing a few slowly moving RBCs, resulting in pixels that were continuously dark for one part of the video sequence and light for the other part, could have as high a variance as a vessel with similarly spaced, more rapidly moving RBCs. For this reason, RBC perfusion within individual capillaries is better represented by the difference image (Fig. 5E). Japee et al. (16) recently described the relation between pixel intensity in the difference image and RBC supply rate (RBCs/s) in transilluminated muscle. Because supply rate is the product of RBC lineal density (RBCs/length) and RBC flow velocity (distance/s) and both of these would increase frame-to-frame differences, difference image intensity would be expected to correlate with RBC supply rate. Care should be taken in interpreting difference images, however, inasmuch as intensity in the difference image would not be an absolute measure of RBC supply. For example, difference image intensity could underestimate RBC supply because of factors such as very high hematocrits (leaving no plasma gaps between RBCs) or regions with poorly focused RBCs. As shown in the difference image for the selected mouse liver sequence (Fig. 5E), a high RBC supply was indicated for only some of perfused capillaries shown in the variance image (Fig. 5C). Although the majority of vessels had RBC flow during the sequence, the magnitude of flows in individual capillaries was quite variable. From direct observation of the video sequence, it was evident that areas with greater RBC flow corresponded to bright areas in the difference image.
The functional maps provide an accurate representation of microvascular morphology but do not give absolute values of blood flow in individual capillaries. RBC velocity is a fundamental measure of blood flow and microvascular function (10, 11, 14, 19), thus providing a quantitative method for characterizing differences in blood flow between diseased and normal tissues (22, 23). To assess absolute RBC velocities and correlate with functional features in the variance and difference images, STI analysis was applied to the video sequence. Although standard video-frame rates limit velocity measurements by this method to 1 mm/s, this limitation is not different for epifluorescence IVVM or transillumination imaging. For higher velocities or measurements in larger vessels, higher frame rates or photometric techniques could be applied to fluorescence contrast imaging. In a comparison of the variance (Fig. 5C) and difference (Fig. 5E) images, some vessels were dark in both, and others were bright in the variance image but dark in the difference image, indicating a low RBC supply, including one vessel with a very low flow velocity (13 µm/s). Vessels with high RBC velocities (>50 µm/s) tended to be bright in variance and difference images. Thus these simultaneous images could be used to provide at a glance 1) all vessels capable of carrying blood plasma, 2) any of those vessels carrying RBCs, and 3) vessels providing elevated RBC supply rates to the tissues. Although one vessel in Fig. 5E with a velocity of 106 µm/s was only moderately bright, Fig. 5A shows that it had a very high hematocrit, which could have affected the variance and difference images. A greatly improved CNR and background uniformity in the variance image (and the difference image) compared with the raw image (Fig. 5A) made these images much easier to interpret.
Quantitative assessment of the functional microvasculature in liver metastases. The analysis tools presented were applied to a video sequence of blood flow in the mouse liver containing experimental liver metastases to generate functional maps highlighting different aspects of RBC-perfused microvasculature in a disease model. The maximum-intensity image (Fig. 7A) revealed extensive intratumoral vascular filling. Tumor vessel morphology was visibly distinct, with tumor vessels (left) having an irregular tortuous structure, clearly different from the normal liver tissue (right), which had a highly branched architecture with smooth vessel walls/lumens. In addition, the metastasis tissue showed increased deposition of FITC-dextran, resulting in a brighter overall appearance, likely a result of increased permeability "leakiness," characteristic of tumor vessels (15), allowing FITC-dextran to diffuse more freely into the tumor tissue.
Quantitative stereology was applied to the maximum and variance images to quantify the observed differences in functional vasculature between the metastasis and normal liver tissues (Table 1). VV and fMVD can be used to characterize the extent of vascular development, whereas BPV and SL can be used to assess the complexity of the microvascular geometry. Despite the large anatomic vascular space indicated by Fig. 7A, Vv was reduced fourfold and fMVD was reduced threefold in the metastasis tissue, suggesting a much lower capacity for RBC supply and O2 delivery.
Vessel network morphology also showed dramatic differences between the metastasis and surrounding normal liver tissues. BPV was fivefold lower in the metastasis tissue, and SL was two times greater. Taken in combination, these values suggest that the geometry of functional tumor vasculature is much simpler than that of the surrounding liver sinusoid. This finding highlights the point that even though there may be angiogenic sprouts and extensive vascular development in rapidly growing tumors or metastases, the morphology of the vasculature actually carrying RBCs may be much simpler.
The difference image provided an objective representation of variability of RBC supply in and around the metastasis tissue. Figure 7C shows that RBC supply was markedly lower within the metastasis (left) than in the surrounding liver vasculature (right). In the region immediately bordering the tumor tissue, microvascular geometry appeared relatively normal (Fig. 7, A and B) but RBC supply was reduced. This suggests that although the invading tumor tissue had not overtaken the adjacent normal vasculature, it imposed boundary conditions that restricted RBC flow in that region and perhaps limited O2 availability for the growing metastatic tumor. This may account for the presence of capillaries in this very small tumor (250 µm diameter).
In summary, animal studies of cancer and other diseases often require methods to assess microvascular perfusion to address O2 delivery to tissues or to study abnormal growth and maintenance of new vessels. The system described here provides a simple and inexpensive means of directly visualizing the functional microcirculation in vital organs of small animals in vivo. Furthermore, detailed quantitative information can be obtained regarding the extent of microvascular perfusion in a variety of tissues and in tumors grown in clinically relevant sites. This novel system provides a means of quantitatively assessing microvascular development and function in normal and diseased tissues and is ideally suited for studies involving vascular remodeling and/or blood flow regulation at the microcirculatory level.
|
GRANTS |
|---|
|
TOPABSTRACTMATERIALS AND METHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
|
REFERENCES |
|---|
|
TOPABSTRACTMATERIALS AND METHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  H. A.J. Struijker-Boudier, A. E. Rosei, P. Bruneval, P. G. Camici, F. Christ, D. Henrion, B. I. Levy, A. Pries, and J.-L. Vanoverschelde Evaluation of the microcirculation in hypertension and cardiovascular disease Eur. Heart J., December 1, 2007; 28(23): 2834 - 2840. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. M. Iga, S. Sarkar, K. M. Sales, M. C. Winslet, and A. M. Seifalian Quantitating Therapeutic Disruption of Tumor Blood Flow with Intravital Video Microscopy Cancer Res., December 15, 2006; 66(24): 11517 - 11519. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 W. Kim, S.-O. Moon, S. Y. Lee, K. Y. Jang, C.-H. Cho, G. Y. Koh, K.-S. Choi, K.-H. Yoon, M. J. Sung, D. H. Kim, et al. COMP-Angiopoietin-1 Ameliorates Renal Fibrosis in a Unilateral Ureteral Obstruction Model J. Am. Soc. Nephrol., September 1, 2006; 17(9): 2474 - 2483. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Visit Other APS Journals Online |
) and y (
) image translation values showing pixel shift for best template match, which is then applied to the entire image to effect stabilization.
