Rca

FIGURE 4 (a) Source biplane angiograms: the upper pair depicts the left coronary artery system (LCA) in time-equivalent projections from right and left anterior oblique (RAO/LAO) views; the lower pair shows the right coronary artery system of the same subject in a second acquisition. (b) Visualization of the 3D model reconstructed from the angiograms of Fig. 4a. The right coronary artery (RCA) is in the front, at the same level behind it to the left circumflex artery (LCX), and to the right the left anterior descending artery (LAD). Images courtesy of Andreas Wahle, University of Iowa.

image intensifier camera

X-rav tube

FIGURE 5 Schematic of the micro-CT scanner.

image intensifier camera

X-rav tube

FIGURE 5 Schematic of the micro-CT scanner.

graphy (micro-CT), required to quantify the geometric and biomechanical properties of the pulmonary arterial tree, and to apply these tools to determine the relative contributions of vessel remodeling (e.g., narrowing of the lumen of the arterial tree due to decreased unstressed vessel diameters and/or decreased vessel distensibility) and changes in the vascular tree structure (e.g., decrease in number of parallel vessels) in causing the elevated pulmonary vascular resistance in pulmonary hypertension of different etiologies. To date, we have built the micro-CT scanner, developed image acquisition and reconstruction protocols and algorithms to produce highresolution 3D image volumes of excised, contrast-enhanced rat lungs, experimented with several methods for image processing and data analysis, and applied the methods to a first model of pulmonary hypertension: chronic hypoxia.

The micro-CT scanner is shown schematically in Fig. 5. The main system components are the demountable microfocal X-ray tube to the right, which is capable of producing focal spots as small as three microns in diameter, the precision specimen stage in the center, and the detector to the left, which consists of a high-resolution 9-7-5" image intensifier coupled to a 10242, 12-bit CCD camera. Figure 6 shows a photo of the scanner with a rat lung in place for imaging. Control and experimental rats in which pulmonary hypertension has been induced by exposure to chronic hypoxia (11% 02 for 3 weeks) are studied. The rats are anesthetized, the trachea and pulmonary artery cannulated, and the lungs excised. The vessels are flushed with a physiological saline solution to remove the blood, then the lungs are suspended by the cannulas from the top of a thin-walled polyacetate cylinder, as can be seen in Fig. 6 where the cannulas are just visible at the top. A brominated perfluorocarbon (Perflubron; perfluor-ooctyl bromide) contrast agent is then introduced through the arterial cannula. Because of the surface tension at the perfluorocarbon-aqueous interface, when introduced into the pulmonary artery, the contrast agent fills only the arterial tree and can be held at a variety of intra-arterial pressures spanning the physiological range. We image either whole lungs or left lung lobes at magnifications ranging from 3.5 x to 9 x .

FIGURE 6 Photo of micro-CT scanner with rat lung in place for imaging. X-ray source is to the right, specimen in the middle with arterial and tracheal cannulas visible, and image intensifier to the left.

Typical technique factors for image acquisition are 85 kVp, 30 microamps, and 3 minutes total scanning time for 360 views.

Projection preprocessing includes correcting the pincushion distortion of the image intensifier using a polynomial unwarping technique in which the coefficients are determined from the image of a precision grid of ball bearings (BB phantom) [105-107]. The rotation axis is then centered with the aid of a sinogram and the illumination nonuniformity corrected by floodfield division. Images are reconstructed, typically onto a 5123 grid, using the Feldkamp conebeam filtered backprojection algorithm [108] and a Shepp-Logan filter [109]. Figure 7 shows as-acquired views of a mouse lung from two representative angles on the left. The corresponding, fully preprocessed projections are shown in the center, and surface-shaded renderings, again from the same view angle, are shown on the right. The mouse lung arterial tree structure could be contained in an 8-mm sphere, and these projections were acquired at a magnification of about 12 x using the 5" image intensifier input diameter. The top panel of Fig. 8 shows a projection and surface shaded rendering of a rat lung viewed from the same angle. The rat lung structure would fit within a 2.5-cm sphere and was imaged at a magnification of about 4 x using the 7" image intensifier input. The bottom panel of Fig. 8 shows a surface shaded rendering of another rat lung on the left and of a dog lung on the right. Inspection of Figs. 7 and 8 reveals distinct interspecies differences in the pulmonary arterial tree morphometry, at least on a gross scale (the distribution vessels).

The segmentation algorithm used to produce the renderings of Figs. 7 and 8 is based on region growing and connectivity list routines, similar to the seeded region growing with gray-level hysteresis described earlier. The image volume is traversed until a voxel is found within a specified gray-scale window. This voxel is the first seed point for the region growing and is added to a list of connected voxels, and the corresponding bit in a previously initialized binary volume is toggled. If a seed point is found, the routine repeatedly checks whether the 26 neighbors fall within the specified gray-scale range and whether they were previously detected as vessel voxels. If they do and were not, they are added to the connectivity list and the corresponding bit in the binary volume is toggled. Processing continues until no more vessel voxels are detected in the first region, in which case scanning of the volume continues until the next seed point is found. The routine builds a new connectivity list for each new seed point it finds and will only accept voxels as seed points if they are not already included on a connectivity list. When volume scanning is complete, all vessels within the gray-scale window will be members of one and only one connectivity list. The user then specifies the minimum size of connectivity list to be accepted in the segmentation (the smallest connected region falling within the gray-scale window that is to be accepted as vessel), and the connectivity lists of all smaller regions are eliminated and the corresponding bits in the binary volume toggled off. Thus, the binary volume becomes a segmented volume of the vascular tree. The renderings themselves are

FIGURE 7 As-acquired (left) and fully preprocessed (center) projections of mouse lung. Right: Surface shaded renderings of mouse lung viewed from same angles. See also Plate 27.
FIGURE 8 Top: Rat lung projection (left) and surface shaded rendering (right) viewed from same angle. Bottom: Surface shaded rendering of rat (left) and dog (right) pulmonary arterial trees. See also Plate 28.

produced using IDL (Interactive Data Language) software [110].

One of the problems with quantitative analysis of a structure as complex as the pulmonary arterial tree is the sheer amount of data available in the 3D images. Pending complete automation of the segmentation and measurement processes, we are exploring rational procedures for data reduction. In the past several methods have been used to summarize the data on segment lengths, diameters and numbers derived from arterial casts. Discussions as to the advantages of the different methods center around the kinds of information retained in the morphometric summary [9,111]. These vessel classification schemes can be categorized as centrifugal or centripetal ordering. In centrifugal ordering, vessels are classified according to generation proceeding from the pulmonary artery to the peripheral, precapillary vessels [73,112]. This has a simple logic when applied to a symmetrical homogeneous tree, but as larger variations in the daughter diameter ratios are encountered as one moves along the actual asymmetrical heterogeneous tree, it results in grouping together vessels of very different diameters in the same generation. In addition, centrifugal ordering has difficulty coping with the wide range in the number of generations between the inlet and periphery between short and long pathways. As an alternative, centripetal ordering has been advocated. For example, in Strahler ordering, the classification begins with the precapillary arter-ioles and proceeds up the tree with the order increasing toward the pulmonary artery [7,8,28,111]. Centripetal ordering has some advantages for summarizing the morphometry of a heterogeneous asymmetrical tree, but among the drawbacks for imaging data is that imaging systems typically have limited resolution, which makes starting with the smallest vessels problematic. It is interesting that, despite the various ordering systems used, the published corrosion cast data [6,9,31-33] from several species appear to be quite consistent with respect to linearity and slope if one plots, for example, the log of the number of segments having a certain diameter vs log segment diameter or the log of segment length vs log segment diameter. This property of pulmonary arterial tree geometry suggests alternatives to the problem of morphometric summary. For example, viewed as a fractal structure, the three vessel segments forming a bifurcation can be considered the repeating, scale-independent element comprising the vascular tree [113,114]. Then, the diameters of the parent (D0) and daughter vessels (D1 and D2) from a number of bifurcations can be used to estimate a value of z defined by D0z = D1 z + D2z. The distribution of z is then the morphometric characterization of the tree [115,116]. This approach is practical for imaging data, but it has also required large numbers of measurements and much of the information about the connectivity of the tree is lost.

Ritman et al. used a method based on analysis of main trunks of the pulmonary arterial tree, which is particularly useful in the context of volumetric CT data [117]. Their results suggest that a reasonable number of measurements of segment lengths and diameters can be exploited to investigate hypotheses about tree structure. One possible approach is to analyze the longest pathway, or main trunk of the tree from the inlet to the smallest resolvable precapillary arterioles. Since this principal pathway contains vessel segments spanning the entire range of diameters present in the structure, its taper might be representative of the tree as a whole. Figure 9 shows a rendering of the rat pulmonary arterial tree in the top left panel. The four longest pathways through the tree are indicated by the colored lines. The bottom left panel shows plots for the four pathways of vessel segment diameter vs distance from the branchpoint off the principal pathway. The top right panel shows a rendering of the pruned principal pathway with the subsidiary pathways rearranged to align the precapillary arterioles at the distal tip. In the bottom right panel, vessel segment diameter is plotted vs distance from the arterial inlet for all four pathways. Curves fitted to the four data sets are statistically indistinguishable, illustrating the correspondence of the tapers of these four pathways. Our results indicate that the various pathways through the tree are self-similar in the sense that all subtrees distal to vessel segments of a given diameter are statistically equivalent [118], and we will exploit the self-similarity of the tree to extrapolate morphometric and mechanical parameters derived from measurements on a small number of pathways to the structure as a whole.

Accurate measurement of vessel segment diameters is a simple concept, but is hampered in practice by a number of complicating factors related to the physics of the imaging system and nuances of the image analysis and data extraction procedures. Volume averaging causes the densities of vessels less than about twice the width of the imaging system point spread function (PSF), which is degraded by the finite focal spot size, specimen stage imperfections, detector pixel size, and the reconstruction algorithm, to be smeared out over a considerably larger volume than they in fact occupy [119]. When subresolution tungsten wires are reconstructed to characterize the system PSF, the cross-sectional image of the wire affects the densities of several pixels (the number of pixels depends upon magnification, which in turn is dictated by imaging geometry), even though each pixel represents more than 10 microns, even at the highest magnifications we have employed (35 x ). Therefore, we have developed a model of the ideal vessel cross-section convolved with the imaging system PSF to calibrate the imaging system for vessel diameter measurements, assuming that artery cross-sections are circular. Figure 10 illustrates how an ideal vessel lumen cross-section larger than twice the full width at half maximum (FWHM) of the system PSF, when convolved with the PSF, yields a density profile whose width at FWHM is a good approximation to the lumen diameter and whose height represents the attenuation characteristics of the contrast agent. A small ideal vessel with width about equal to the PSF, on the other hand, when convolved with the PSF, yields a density trace whose width at FWHM significantly overestimates the vessel diameter and whose height underestimates the contrast agent density. But the areas under the density profiles are proportional to the vessel diameters since, in spite of volume averaging, contrast agent density is conserved in the reconstruction. We have imaged wires and contrast agent-filled tubes of precisely known diameters and found an exactly linear relationship with zero intercept between the area under density profiles taken orthogonal to the wire or tube axis and the true diameter. Therefore, as long as at least one cylindrical object of known diameter is contained within the reconstruction, we can accurately measure vessels down to subpixel diameters using a density profile area vs diameter calibration curve.

Before developing a reliable calibration method for small vessel diameter measurements, we measured vessels larger than the system PSF using the FWHM criterion. In order to measure a statistically significant number of smaller vessels, high magnification images of distal portions of the lung were obtained as shown in Fig. 11. Density traces across vessels like the one indicated by arrows on the left panel were analyzed, as shown in the right panel, by fitting the line-scan data to a model projection of an ideal circular vessel riding on a nonzero and nonuniform baseline. The width of the model vessel was then taken as the vessel diameter.

Since the arterial trees are reconstructed on cubic-voxel grids of isotropic resolution, it is most straight forward to access orthogonal (axial, sagittal, or coronal) slices from which to extract density profiles and diameter measurements, as shown

FIGURE 9 Top left panel illustrates the four longest pathways through a rendering of the rat pulmonary arterial tree. Bottom left panel shows plots of vessel segment diameter vs. distance from branchpoint off the principal pathway. Top right panel shows pathways rearranged to align the precapillary arterioles. Bottom right panel shows plots of vessel segment diameter vs. distance from pulmonary arterial inlet. The fitted curves are statistically indistinguishable. See also Plate 29.

FIGURE 9 Top left panel illustrates the four longest pathways through a rendering of the rat pulmonary arterial tree. Bottom left panel shows plots of vessel segment diameter vs. distance from branchpoint off the principal pathway. Top right panel shows pathways rearranged to align the precapillary arterioles. Bottom right panel shows plots of vessel segment diameter vs. distance from pulmonary arterial inlet. The fitted curves are statistically indistinguishable. See also Plate 29.

for axial slices in the left panel of Fig. 12. By interactively "flying through" the stack of transaxial slices, bifurcation locations can be reliably and reproducibly identified as the xyz coordinates where two daughters have just separated, as shown on the right of Fig. 12 in the third panel from top. After recording the coordinates of all branch points for daughters leaving a pathway, the diameters of the vessel segments comprising the pathway can be measured by accessing the slice equidistant between the contiguous branch points defining the segment. However, as is evident in Fig. 12 and shown in Fig. 13, many vessel segment orientations deviate significantly from the vertical. In Fig. 13 transaxial (yellow) and orthogonal (lavender) slices through a nearly horizontal vessel are indicated on the left panel. The vessel cross-section in the transaxial slice is a high-aspect-ratio ellipse, as shown in the top right panel. It is nearly impossible to make reliable and reproducible diameter measurements from this image data. We have therefore developed software to access oblique slices orthogonal to the vessel segment at the midpoint between consecutive bifurcations off the pathway, as illustrated by the

FIGURE 10 Left. Ideal vessel lumen cross section larger than twice the FWHM of the system PSF when convolved with the PSF yields a density profile whose width at FWHM is a good approximation to the lumen diameter and whose height represents the attenuation characteristics of the contrast agent. Right. Small ideal vessel with width about equal to the PSF when convolved with the PSF yields a density trace whose width at FWHM significantly overestimates the vessel diameter and whose height underestimates the contrast agent density. But the areas under the density profiles are proportional to the vessel diameters.

FIGURE 10 Left. Ideal vessel lumen cross section larger than twice the FWHM of the system PSF when convolved with the PSF yields a density profile whose width at FWHM is a good approximation to the lumen diameter and whose height represents the attenuation characteristics of the contrast agent. Right. Small ideal vessel with width about equal to the PSF when convolved with the PSF yields a density trace whose width at FWHM significantly overestimates the vessel diameter and whose height underestimates the contrast agent density. But the areas under the density profiles are proportional to the vessel diameters.

lavender plane of Fig. 13. The vessel cross-section viewed in the orthogonal slice is nearly circular, facilitating accurate diameter measurement.

A major goal of our current research is to extract functionally relevant morphometric parameters from 3D images of arterial tree structures. Our current approach is to analyze the principal pathway from the inlet to the periphery, measuring the diameters of all segments of the main trunk and of all the daughter branches immediately off the main trunk.

The upper curve in Fig. 14 shows a plot of vessel segment diameter vs distance from the pulmonary arterial inlet for the principal pathway of a hypertensive rat lung imaged at high pressure. The lower plot shows the same relationship for the smaller daughter vessel segments branching immediately off the main trunk. Morphometric parameters available from such plots include the parameters of the fitted curves (indicators of slope and curvature) and the dispersion of the data about the fits. The latter is greater for the daughter segments, since the

FIGURE 11 Shows method for measuring small vessel diameters from high-magnification projection images. The model vessel diameter is indicated as 50 microns in the right panel.

FIGURE 12 Left. Transaxial slices to which many vessels are nearly orthogonal can readily be accessed from the reconstruction. However, many vessel segments can be seen to deviate significantly from the vertical. Right. Bifurcation locations can be identified as the xyz coordinates where two daughters have just separated as shown in the third panel from top. See also Plate 30.

FIGURE 12 Left. Transaxial slices to which many vessels are nearly orthogonal can readily be accessed from the reconstruction. However, many vessel segments can be seen to deviate significantly from the vertical. Right. Bifurcation locations can be identified as the xyz coordinates where two daughters have just separated as shown in the third panel from top. See also Plate 30.

Tr$nM*n5l Slice

Tr$nM*n5l Slice

FIGURE 13 Transaxial (yellow) and orthogonal (lavender) slices through a nearly-horizontal vessel are indicated on the left panel. The vessel cross-section in the transaxial slice is a high-aspect-ratio-ellipse, as shown in the top right panel, whereas the cross section in the orthogonal slice is nearly circular. See also Plate 31.

FIGURE 14 Plot of vessel segment diameter vs. distance from the pulmonary arterial inlet for the principal pathway of a hypertensive rat lung imaged at high pressure. The lower plot shows the same relationship for the smaller daughter vessel segments branching immediately off the main trunk.

D(x, P) = a[l + aP(x)](1 - bx)c where a is the global distensibility and a, b, and c are parameters of the fit. In the case of one normal and one hypertensive rat the results were as below:

Blood Pressure Health

Blood Pressure Health

Your heart pumps blood throughout your body using a network of tubing called arteries and capillaries which return the blood back to your heart via your veins. Blood pressure is the force of the blood pushing against the walls of your arteries as your heart beats.Learn more...

Get My Free Ebook


Post a comment