Composition of Human Thrombus Assessed by Quantitative Colorimetric Angioscopic Analysis
Background Angioscopy surpasses other diagnostic tools, such as angiography and intravascular ultrasound, in detecting arterial thrombus. This capability arises in part from the unique ability of angioscopy to assess true color during imaging. In practice, hardware-induced chromatic distortions and the subjectivity of human color perception substantially limit the theoretic potential of angioscopic color. We used a novel application of tristimulus colorimetry to quantify thrombus color to both aid in its detection and assess its composition.
Methods and Results A series of human thrombus models were constructed in vitro. Spatial homogeneity was ensured by light and electron microscopy. Quantitative colorimetric angioscopic analysis demonstrated excellent measurement reproducibility (mean difference, 0.07% to 0.17%), unaffected by illuminating light intensity (coefficient of variation, 0.21% to 3.67%). Colorimetric parameters C1 and C2 were strongly correlated (r=.99, P<.0001) with thrombus erythrocyte concentration. Principal components analysis transformed these parameters into a single value, the thrombus erythrocyte index, with little (0.06%) loss of content. Measured and predicted concentrations were similar (mean difference, 0.16 erythrocytes per 1 ng). Randomly ordered images were also subjected to visual analysis by three experienced angioscopists, with suboptimal levels of both intraobserver (mean κ=0.63) and interobserver (mean κ=0.48) agreement. In addition, visual ranking resulted in a Kendall rank coefficient of 0.72 to 0.76 versus a perfect 1.00 from quantitative measurement.
Conclusions Quantitative colorimetric angioscopic analysis provides a new, objective, and reproducible analytic tool for assessing angioscopic images of human thrombus. Even under ideal circumstances, experienced angioscopists do a poor job of assessing color (and therefore composition) of human thrombi. This technique can, for the first time, provide quantitative information of thrombus composition during routine diagnostic imaging.
The presence of thrombus in the coronary arterial tree has long been recognized by pathologists as a common finding in postmortem hearts.1 2 However, it was the classic work of DeWood and colleagues3 in 1980 that firmly established the etiological role of thrombus in the genesis of acute myocardial infarction. More recent investigations have also identified thrombus as a prominent feature of unstable angina.4 5 6 7 8 This recognition has led to the successful treatment of this common disorder with agents that interfere with the clotting process.9 10 11 12 13 14 15 Current thinking now places thrombus at a pivotal position in nearly all acute coronary syndromes.16
Part of the past controversy regarding intracoronary thrombosis arises from the difficulty of its detection in the living heart. Contrast angiography remains the most widely used technique for characterizing coronary pathology. However, thrombus that projects into the vascular lumen is generally surrounded by contrast medium, resulting in a roentgenographic silhouette that frequently masks its presence. Mural thrombus avoids this problem through its adherence to the vessel wall. Its detection is far from assured, however, because angiography can characterize only the arterial lumen and not the adjacent structures. Finally, even if a filling defect is visualized, angiography cannot reliably distinguish thrombus from other common intracoronary structures such as plaques, dissections, or intimal flaps.
Newer imaging techniques such as intravascular ultrasound have the advantage of producing topographic rather that silhouette representations and can interrogate tissues beneath the luminal border.17 Thrombus, though, possesses an acoustic density that differs only slightly from nearby normal and abnormal tissues, making a positive identification particularly difficult.18 It is also possible to use approaches that specifically search for the presence or absence of thrombus. Indium-labeled platelet imaging represents a good example of a procedure that has been quite successful in identifying thrombus localized to the chambers of the heart.19 20 Attempts to extrapolate these impressive results to the detection of thrombus confined to the coronary vasculature have been disappointing, however, likely because of the small mass of thrombus involved.21
Angioscopy has been shown to represent a useful technique in the recognition of intracoronary thrombus.7 22 23 24 25 26 27 28 29 30 31 Indeed, thrombus detection may prove to be the most important clinical capability of this emerging diagnostic procedure. Angioscopy presents several features that make it uniquely suited to this role. First, its interrogative capabilities are confined to the intraluminal space and surrounding endoluminal surface. This is precisely the area where occlusive coronary thrombi are generated and located. Second, unlike other techniques, angioscopy can, in a single image, visualize a major portion of the surface of a thrombus. This enhances one’s ability to distinguish thrombus from other protruding structures such as atheroma. Third, angioscopy provides a spatial resolution, measured in our laboratory, of ≈50 μm. This compares favorably to resolutions of ≈200 μm for angiography32 and ≈150 μm for intravascular ultrasound.33 Finally, angioscopy represents the only diagnostic technique in cardiology capable of providing information on true color. Because the color of thrombus frequently differs from that of surrounding structures, better diagnostic specificity is realized.
In the current study, we investigated whether the composition of a human thrombus could be ascertained through characterization of the light reflected from its surface. This was accomplished using a new tool developed in our laboratory specifically for the quantification of images obtained during angioscopy.
Creation of Thrombus Model
A series of static human thrombus models were constructed in vitro. Blood was collected at room temperature from a single healthy donor using a large-bore intravenous needle. The donor was not using chronic medications or birth control pills and had not taken aspirin for at least 30 days before donation. All studies were performed in accordance with the guidelines of the Institutional Review Board. An anticoagulant, consisting of 0.8% citric acid, 2.2% sodium citrate, and 2.25% glucose, was continually mixed with the blood during collection achieving a final concentration of 6%. Immediately after collection, platelet-rich plasma was created by centrifugation of anticoagulated blood at 450g for 20 minutes at 23°C. After removal of the supernatant, the remaining mixture was centrifuged at 20 000g for 10 minutes at 4°C to produce platelet-free plasma. Appropriate volumes of platelet-rich and platelet-free plasma were combined to create plasma with a plasma platelet concentration identical to that of the original anticoagulated blood sample. The integrity of the erythrocytes subjected to the separation process was verified by the confirmation of a plasma-free hemoglobin concentration of <10 μg/L obtained from all samples of blood and plasma.
The thrombi were formed in a series of 12-mm-diameter cylindrical polystyrene containers held in a 37°C water bath. By use of a precision micropipette (rated accuracy ±1.0%), appropriate volumes of the plasma mixture and blood were combined to produce 18 different erythrocyte concentrations. The final target concentrations of erythrocytes in the cylinders, relative to the original anticoagulated blood, were 100%, 80%, 60%, 40%, 30%, 25%, 20%, 15%, 12%, 8%, 6%, 4%, 2%, 1%, 0.5%, 0.2%, 0.1%, and 0%. After mixing, each mixture was maintained at 37°C for 45 minutes before clotting to enhance platelet activation. Before clotting, a 700-μL aliquot of each mixture was sent for assessment of the concentration of erythrocytes, leukocytes, and fibrinogen. Clotting was initiated with the addition of 230 μL of 0.2 mol/L calcium chloride, resulting in a final volume of blood mixture of 4530 μL in each cylinder. During clotting, each cylinder was gently agitated at 3-minute intervals to prevent cell settling, and each was capped to prevent the loss of CO2 with a resultant change in pH. All clots were maintained at 37°C for a minimum of 18 hours to allow sufficient time for full clot retraction.
The formed thrombus models were removed, accurately weighed, and submerged in Michaelis buffer (sodium acetate and sodium 5,5-diethylbarbiturate titrated to a pH of 7.42)34 contained in a Petri dish. The serum was used for determining the concentration of remaining formed elements and fibrinogen. Each thrombus model was then photographed using constant lighting conditions, exposure settings, and object distance.
Composition of Thrombus Models
The concentration of erythrocytes, leukocytes, platelets, and fibrin contained in each thrombus model was computed by use of the differences in volume and concentration of the blood/plasma mixture before clotting and the serum after clotting. These values were computed in absolute terms as a quantity per thrombus and in relative terms per gram of thrombus material. Spatial homogeneity was assessed by light microscopy of a 5-μm-thick vertical section cut along a diameter of each cylindrical thrombus model. Examination of these sections prepared with Goldner stain permitted assessment of the relative content of each formed element at the top surface (remote from container wall), at the bottom and side surfaces (adjacent to container wall), and at various depths within the thrombus model. The ultrastructure was assessed with electron microscopy of glutaraldehyde-fixed sections of each thrombus model.
A custom test setup was designed and constructed for in vitro imaging of the thrombus models (Fig 1⇓). Angioscopic imaging was accomplished with a new 4.5F Baxter ImageCath coronary angioscope. Before testing, a camera white balance was performed during imaging of an equal-energy white color standard (Spectralon, Labsphere); the resulting white image was also recorded for the derivation of correction factors detailed below. Two 0.3-mm-diameter steel pins were mounted at the tip of the setup device just outside the image field. These pins were brought into contact with the surface of the thrombus model undergoing angioscopic imaging, thereby maintaining a fixed and reproducible 1.7-mm distance from angioscope tip to object surface. All imaging was accomplished with the angioscope tip and thrombus model submerged in a buffer solution to reduce surface reflection and more closely approximate in vivo imaging conditions. A flat black background was used to minimize reflection of any traversing light energy. The thrombi were illuminated through the angioscope with a 300-W xenon light source (Baxter model OPTX 300), and the image was captured on an Envision model 2705 single-chip, 1/3–in (8.38-mm) CCD video camera with the automatic shutter control disabled to prevent autoregulation of light level. To permit subsequent off-line analysis, all images were recorded on 0.5-in super VHS videotape using the “Y/C” video signal. Imaging was performed in a fully darkened room with the angioscope shielded from stray light emanating from the light source ventilation ports or video monitor screens.
All thrombi models were oriented so that imaging was performed on the top surface of the clot, an area not in contact with the container wall during the final stages of clotting. Each thrombus model was imaged four times, once at each of four light intensity settings. The settings used were held constant for all testing and were selected to provide a wide range of light intensities while maintaining the mean luminosity values within the valid working range as previously established in our laboratory.35 To assess the homogeneity of the thrombus model, each thrombus was bisected after completion of surface imaging and a new image obtained from its center.
Image Processing and Analysis System
Tristimulus colorimetric measurements were made from the images of the thrombus models using the quantitative colorimetric angioscopic analysis system developed in our laboratory. Detailed descriptions of this system have been published previously.35 36 In brief, video images obtained either directly from the camera or from videotape were digitally sampled by use of a fast 24-bit (8 bit per channel) analog-to-digital conversion system. This system permits real-time (1/25 second) digitization in 16.7 million colors and provides a spatial resolution of 720×576 for PAL (720×480 for NTSC). Single-frame images were acquired and stored on a microcomputer in a 640×480×24-bit-deep bit map format. A custom image analysis program permitted the user to select a rectangular region of interest intended for quantitative colorimetric analysis. Red, green, and blue (RGB) color values for each picture element (pixel) contained within the selected region were computed. These values were corrected for errors inherent to the hardware imaging chain by use of three sequentially applied algorithms (color-dependent offset value, software-derived “white balance,” and exponential gamma correction). The corrected RGB values were transformed into two other chromaticity coordinate systems. The first, known as the HSI color space, allows a color to be completely described by hue, saturation, and intensity. The second, the C diagram, was developed in our laboratory specifically for angioscopic imaging.35 In this system, color that is independent of light intensity is completely defined by two parameters, C1 (an intensity-independent red index) and C2 (an intensity-independent green index). These colorimetric parameters were normalized to provide a full range of 0 to 1.
All image analyses were performed without knowledge of the composition of the thrombus models. To maximize objectivity, a predefined and reproducible region of interest for quantitative measurement was used for all images.
Visual Interpretation of Thrombus Models
To assess the accuracy of visual discrimination of color between the thrombus models, three experienced angioscopists were asked to blindly grade and order the color of all images of the erythrocyte thrombus models containing a thrombus erythrocyte concentration of 8 cells per 1 ng or less. None of these individuals were involved with either the creation or imaging of the thrombus models. To accomplish this task, each expert was presented with 44 images of the 11 thrombus models, each recorded at four different light intensities. The order of presentation was set randomly by a computer-generated random number table, and the images were shown sequentially for 5 seconds with a 5-second pause between images. The experts were asked to order the images from the greatest to least concentration of erythrocytes, as well as assign each thrombus model to one of five ordered color categories: red, whitish-red, red-white, reddish-white, and white.
Continuous variables were analyzed by use of the two-tailed t test. Linear and nonlinear regression techniques were used to explore the relations between different colorimetric indexes, as well as in the construction of a model predictive of thrombus erythrocyte concentration. Multidimensional data were reduced to a single dimension by principal components analysis. To assess observer agreement, the κ statistic was used to assess the magnitude of differences found both between and within observers. The accuracy of ranking of angioscopic images was analyzed using Kendall’s rank correlation. Two-way ANOVA was used to investigate the relation between illuminating light intensity and visual ranking of thrombus model color. Plus-minus values represent mean±SD. Mean differences associated with a value of P<.05 were deemed statistically significant.
Thrombus Model Creation
A total of 18 human thrombus models were created in vitro under static flow conditions. The composition of each was designed to reflect decreasing concentrations of erythrocytes relative to the number of platelets. This was accomplished by sequentially decreasing the number of erythrocytes in the preclot mixture while maintaining a nearly constant plasma platelet concentration. Table 1⇓ provides the target proportion of whole anticoagulated blood planned in the final mixture, as well as the actual measured concentration of elements and fibrinogen determined before clotting. The values reflect a serially diminishing hematocrit and concentrations of erythrocytes and leukocytes while maintaining a steady plasma fibrinogen and platelet concentration.
A comparison between planned and observed erythrocyte parameters is provided in Table 2⇓. Note that the differences between measured and predicted values were small, averaging −0.07±0.34% for hematocrit and −0.42±0.59% for hemoglobin. This confirms the accuracy of the procedure in achieving the desired erythrocyte concentrations.
The final composition of each thrombus model was computed by use of the difference in absolute amounts of its component constituents between the preclot mixture and the postclot serum (Table 3⇓). In the first 12 thrombus models, a measurable number of erythrocytes remained in the serum after clotting. Nearly all the platelets present in the preclotting mixture were incorporated into the thrombus models, with measurable serum quantities found in only 3 samples. All measurable quantities of fibrinogen were consumed in all samples during the clotting process.
Although the volume of preclot mixture was held constant for each thrombus model, the final weight of the thrombi after clotting differed significantly (Table 4⇓), ranging from 0.35 to 1.72 g. To correct for these size differences, the absolute number of formed elements contained in each thrombus model was normalized by weight as shown in Table 4⇓. As expected, the resulting concentration of erythrocytes differed more than a 100-fold from thrombus 1 to 18 (9.59 versus 0.08 per 1 ng), with more modest differences observed in platelet concentrations (0.47 versus 1.74 per 1 ng).
Assessment of Measurement Variability
Each thrombus model was imaged during submersion by use of an angioscope model in common clinical use. The circular image obtained at the object-to-catheter distance used in this study was ≈3.4 mm in diameter. Colorimetric measurement reproducibility was assessed by analysis of three sequentially acquired images obtained from each of 11 randomly selected thrombus models. The values obtained were found to be reproducible during serial imaging, with an SD of 0.002 for C1 values and 0.001 for C2 values. This corresponded to a mean absolute difference of 0.17±0.13% and 0.07±0.05% for C1 for C2, respectively.
It is also important to ensure that the colorimetric values obtained are relatively independent of color intensity, because this latter parameter is highly dependent on both the brightness of the illumination source and the object-to-catheter distance. For this evaluation, serial images were obtained while illumination intensity was varied at the light source over the entire range of clinically useful levels of brightness. Colorimetric analysis of the 72 images obtained (4 per thrombus model) revealed a mean SD below 0.007 for all four colorimetric parameters. Corresponding coefficients of variation (mean±SD) were 1.00±0.79% and 3.67±1.53% for C1 and C2 and were even smaller for hue and saturation (0.21±0.32% and 2.67±3.15%, respectively). Hence, illumination intensity had little impact on the colorimetric values obtained in this study.
Assessment of Color Heterogeneity
Color heterogeneity of the thrombi was assessed in two ways. First, local heterogeneity was examined with 25 contiguous regions of interest positioned in a 5×5 square matrix and applied to images obtained from each thrombus model, with the size of each region measuring ≈0.32 mm (true scale) in length and width. Colorimetric analysis of all these 450 regions produced a mean coefficient of variation per thrombus model of 2.71% for C1 and 4.07% for C2. Second, to examine possible differences between the surface and interior of the thrombus, each thrombus model was imaged at its center after bisection as well as at its surface. The colorimetric values obtained from these two different areas revealed a mean difference of −0.04±0.06 for C1 and +0.02±0.03 for C2. These data suggest that the color assessed by quantitative colorimetric angioscopic analysis is relatively homogeneous throughout each thrombus model and that measurement at one site is representative of the thrombus overall. Histological homogeneity was verified by microscopy (Figs 2⇓ and 3⇓).
Relation of Colorimetric Measurements to Thrombus Models
The C1 and C2 colorimetric values obtained from the thrombus models are graphically depicted in Fig 4⇓. Starting from thrombus 9, C1 (the red index) demonstrates a progressive decrease, and C2 (the green index) demonstrates a progressive increase in value with decreasing amounts of erythrocytes in the preclot mixture. These values converge near but not at the white point (C1=C2=0.333). In contrast, the thrombus models generated from mixtures containing 100% to 15% blood (thrombus 1 through 8) reveal no apparent change over this range.
Fig 5⇓ displays the relation between the color intensity value and thrombus number. A minor linear increase in intensity is evident (r=.74) as the number of erythrocytes in the preclot mixture decreases. Although this trend is statistically significant (P<.001), the extent of scatter of the individual values is large enough to preclude useful predictive power (standard error of the estimate, 13.8). Thus, colorimetric differentiation of the thrombus models was well characterized by C1 and C2 values, with little useful discriminating information provided by intensity.
Optimal Characterization of Thrombus Color
The results of the quantitative colorimetric measurements obtained from the thrombus models can be described using several different color coordinate systems. Fig 6⇓ shows a polar plot of the colorimetric values of the thrombus model, with hue presented in angular values and saturation in radial values. Saturation varied over most (75.0%) of its possible range and provided most of the variability that distinguished one thrombus model from another.
The C1/C2 chromaticity diagram provides an alternative descriptive format (Fig 7⇓). With this triangular coordinate system, red colors are located in the lower right corner, green in the upper left corner, and blue in the lower left corner. The thrombus models in the current study range in color from a strong red (thrombus 1) to a white with a slight cyan-blue tint (thrombus 18).
Creation of a Predictive Model
It is important to note that in Fig 7⇑ the observed values occupy a nearly linear (r=.998) distribution. This pattern permitted the reduction of the two-dimensional data shown in this figure into a single dimension, thereby permitting the differentiation of the thrombi using a more convenient single parameter. Principal components analysis was used to achieve this transformation. This process was accomplished with little loss of predictive accuracy, because the first principal component in this model by itself accounted for 99.94% of the total observed variance. The resulting model was weighed more heavily toward C1 than C2, with linear coefficients of .907 and −.422, respectively.
A final goal was the transformation of this colorimetry-derived parameter into a more direct and clinically relevant descriptor of thrombus composition such as the number of erythrocytes contained within a nanogram of thrombus material. This allowed the creation of a final model that predicted the concentration of erythrocytes in a given thrombus based solely on measured C1 and C2 values. The final predictive equation is provided below: where TEI represents the colorimetry-derived thrombus erythrocyte index. Fig 8⇓ graphically compares the colorimetric estimates of erythrocyte concentrations of the thrombus models with actual measured values. The mean difference between estimated and measured values was small (0.16 erythrocytes per 1 ng).
Comparison of Quantitative and Visual Categorization and Ranking
κ statistics were used to assess the magnitude of observer agreement in the 44 randomly ordered images. Using a five-point ordinal color scale, the assessment of intraobserver variability revealed κ values ranging from 0.57 to 0.73, with a mean of 0.63. Analogous κ values for interobserver variability were somewhat lower, with a mean of 0.48 (range, 0.34 to 0.70). None of the individual κ values obtained fell into a range generally considered to indicate good observer agreement (ie, κ>0.75).
In addition to comparing color categories, the sequentially displayed images were ranked, allowing comparisons not only between different observers but also between observers and the quantitative analysis system. Each angioscopist was asked to rank the 44 randomly presented images from the highest to lowest concentration of erythrocytes on the basis of color. A second reading of the entire set of images immediately followed the first. These same images were also subjected to quantitative colorimetric analysis, during which they were ranked on the basis of computed thrombus erythrocyte index values. Calculation of the Kendall rank correlation coefficient permitted assessment of the accuracy of the ranking using the measured thrombus erythrocyte concentration as the standard for comparison. A Kendall coefficient of 1.00 indicates a completely accurate ranking, whereas a value of 0.00 is anticipated for a ranking done solely by chance. In this study, the coefficients obtained from the first rankings of the three trained observers were 0.68, 0.74, and 0.76 (mean, 0.72). The values for the second reading were each slightly better (0.72, 0.75, and 0.81 with a mean of 0.76), suggesting the possibility of a training effect. Compared with these scores, the quantitative colorimetric angioscopic analysis system produced a Kendall rank correlation coefficient of 1.00, indicating a completely accurate ranking of all 44 angioscopic images.
Because each thrombus model was imaged with four different light intensities, it was possible to ascertain whether light intensity itself systematically influences the visual interpretation or quantitative measurement of thrombus color and composition. Similar mean thrombus rankings at each intensity would be expected if visual interpretation were unaffected. However, the mean visual rankings decreased progressively (24.6 to 23.6 to 22.7 to 21.6) as the light intensity decreased from its highest to its lowest value (P=.007). This trend was seen individually for all three observers as well. This finding suggests that trained observers tend to rate the erythrocyte concentration as erroneously high in images that are dimly lit and erroneously low in images that are brightly lit. This pattern of error was not observed with the quantitative colorimetric analysis system, which reported values uninfluenced by light intensity.
Importance of Thrombus Composition
Intracoronary thrombus should not be viewed as a single entity with uniform characteristics and behavior. Rather, its composition may vary considerably, depending on the clinical milieu present at the time of its creation. Pathologists have for decades dichotomized thrombi into the categories of “red” and “white” on the basis of their gross appearance.37 The color difference appears to arise from the relative number of erythrocytes and platelets present. This distinction represents more than a pathological convenience. In 1992, Mizuno and coworkers23 reported that intracoronary thrombi were angioscopically visualized in 94% of all patients who presented with acute coronary syndromes. Importantly, in 10 of 14 patients with unstable angina, the observed thrombi were grayish-white, whereas in all 15 patients with acute myocardial infarction, the thrombi were reddish. Hence, thrombus color might be different in these two important coronary syndromes.
In addition to its diagnostic importance, knowledge of thrombus composition has therapeutic implications. Jang et al38 have reported on the differential sensitivity of erythrocyte-rich and platelet-rich thrombi to lysis using tissue-type plasminogen activator. This difference may account for the apparent paradox of a clearly favorable response to attempted thrombolysis seen in acute myocardial infarction39 40 and the repeatedly poor response observed in unstable angina,12 41 despite the fact that both syndromes are presumably initiated by coronary thrombus formation. Foreknowledge of the relative sensitivity to drug-induced lysis of a given thrombus would be extremely useful to a cardiac interventionalist. A procedure complicated by acute thrombosis could then be most appropriately treated with intracoronary lytic agents or further mechanical intervention.
Interpretation of Thrombus Color
Despite the potentially important role of color in enhancing the diagnostic abilities of angioscopy, two major practical problems exist that could greatly limit its substantial theoretic utility. First, the color of an object as viewed on a video screen can differ markedly from its true color. The light source, the angioscopic imaging bundle, the white balance and gamma correction algorithms used by the camera, the format of video modulation, the linearity of the recording device, and the manual settings of the monitor color controls can all induce significant errors in color accuracy.36 Unfortunately, the impetus driving the development of video methodology and standards used in angioscopy has not been scientific concerns but the consumer market in which physically accurate color rendition may not be important or even desirable. The second practical problem involves the interpretation of color. Because of the complexities and individualities of color perception, each observer interprets and describes color in unique terms. This has led to many creative but imprecise angioscopic descriptions of thrombus color reported in the literature such as an appearance consistent with “tomato puree.”42 The magnitude of this limitation was recently highlighted by the European Working Group on Coronary Angioscopy.43 This group studied observer variability in the classification of findings derived from 30 clinical angioscopic procedures. Despite the elimination of color rendition errors discussed above, the κ values for chance corrected intraobserver and interobserver color variabilities were among the lowest recorded for any category (κ range, 0.29 to 0.82 and −0.04 to 0.52, respectively), including some interobserver values that were worse than predicted by chance alone. When the equipment-related color distortions occurring at multiple levels of the imaging chain are combined with the marked subjectivity of color perception, the visual interpretation of angioscopic color as currently practiced would appear to have limited clinical utility.
These limitations provided the impetus for the creation of a quantitative approach to color measurement. The precision of the quantitative colorimetric angioscopic analysis system developed in our laboratory has been validated previously.35 36
This study demonstrates that human thrombus models can be constructed in vitro to provide a cellular composition varying over a wide range in a predictable manner. These models appear to display a high degree of spatial homogeneity, as gauged by both histological evaluation and surface/interior color measurement. Sequential imaging and quantitative colorimetric measurement of the models provided repeatable values, with all absolute differences averaging <0.2%.
The C1 and C2 colorimetric values obtained were nearly independent of illumination intensity (SD <0.007). This observation is important if this technique is to be applied to clinical angioscopy. Illumination intensity is determined by the interaction of power output of the light source and the square of the object-to-catheter-tip distance. The former variable can be controlled. However, object distance during in vivo intraarterial imaging is impossible to fix and difficult to measure. Structures viewed at a distance visually appear to have a darker color than when viewed near the angioscope tip. Quantitative colorimetric analysis avoids this pitfall in color assessment. Unlike clinical imaging, our in vitro setup allowed us to set the imaging distance and hence accurately measure the color intensity of our thrombus models. As Fig 5⇑ shows, little discriminant information is provided by the accurate knowledge of intensity value; its exclusion from the final computational model is therefore of little consequence.
Colorimetry and Thrombus Composition
The relation of colorimetric values and thrombus model number is graphically depicted in Fig 4⇑. The plateau effect of C1 and C2 found at higher preclot concentrations of erythrocytes would at first glance appear to limit the usefulness of the technique in distinguishing between erythrocyte-rich thrombi. The explanation for this observation, though, appears to lie with differences in size of the formed thrombus models. Although each model was constructed from equal preclot volumes, Fig 9⇓ shows that the final clot mass generally decreased with increasing thrombus number. This phenomenon likely resulted from the relative effectiveness of platelet-induced clot retraction. In thrombi containing a large concentration of erythrocytes, the ability of platelets to coalesce and extrude entrapped serum could be hampered by the presence of the bulky cells, resulting in a larger thrombus. To correct for this effect, colorimetric values were compared with the number of erythrocytes contained within each nanogram of formed thrombus (Fig 10⇓). Not only was a significant curvilinear relation achieved (r>.99), but no plateau was observed. These findings support the utility of quantitative colorimetric analysis in distinguishing thrombi over the entire achievable range of erythrocyte thrombus concentrations.
Fig 10⇑ also includes the outlier, thrombus model 14. This thrombus retained a disproportionate volume of serum during its formation (Fig 9⇑). Although the resulting measured C1 value was 22% lower than expected from the relation shown in Fig 4⇑, this apparent error was completely eliminated by the use of thrombus erythrocyte concentration as the independent variable.
The colorimetric values obtained were also examined for thrombus erythrocyte content, thrombus hemoglobin content, thrombus hemoglobin concentration, and erythrocyte/platelet ratio. All these relations proved inferior to the thrombus erythrocyte concentration in the strength of their correlation to measured colorimetric values.
Interrelations of Colorimetric Parameters
A nearly linear relationship between C1 and C2 is evident in Fig 7⇑. This finding is most commonly noted in the field of colorimetry when two pure colors are mixed in progressive proportions. This physical concept appears quite analogous to our current model involving progressive erythrocyte concentrations.
In addition to its physical implications, the observed linear relationship in C1 and C2 provides the opportunity of simplifying the predictive model and enhancing clinical applicability. In analogy to the three types of cones in the human retina, tristimulus colorimetry can completely describe any color with the use of three variables, such as RGB or HSI. We have gone one step further in the C1/C2 chromaticity scale by eliminating the contribution of intensity in the description of color, a step that has few practical consequences in angioscopy because of the marked dependence of this parameter on illumination brightness. A two-parameter system of angioscopic color description is still awkward, though, because it requires the use of two numeric values for every color measurement, making comparisons particularly difficult. When quantitative colorimetric angioscopic analysis is used in the study of thrombus, our data suggest that transformation from a two-dimensional to a single-dimensional description of color using principal components analysis is possible with minimal (0.06%) loss of content. Thrombus can therefore be accurately characterized with a single parameter, the thrombus erythrocyte index.
It is important to note several possible limitations of our investigation. First, all the thrombus models were created under static conditions. In contrast, a sizable proportion of arterial thrombi form in vivo in continuity with flowing blood. This milieu frequently leads to thrombi composed of laminations of relatively erythrocyte-rich and platelet-rich material.44 Although dynamic models are available for the formation of layered thrombi, they were not used in the current investigation because the resulting heterogeneity of composition and color would preclude an accurate comparison of these two parameters.
Second, because of the desire to carefully control thrombus formation and composition, testing was not extended to other potential in vitro and in vivo thrombus models. Though unproven, extrapolation of our results to other types of thrombus appears rational, because colorimetry relies on physical rather than biochemical properties of the interrogated material.
Finally, although the focus of the current study was the relative contribution of erythrocytes, thrombus color may be potentially influenced by the relative amounts of platelets and fibrin as well. This issue is currently being explored in depth in our laboratory.
With increasing recognition of the importance of thrombus in heart disease comes the need for its improved identification and quantification. Although modern diagnostic techniques may be marginally useful in detecting thrombus, they cannot provide information on composition. Knowledge of composition may prove to be especially useful in assessing likely outcomes or establishing optimal treatment.
Our system of quantitative colorimetric angioscopic analysis provides an accurate and reproducible method of determining the concentration of erythrocytes in these thrombus models using a single derived colorimetric parameter. It is clearly superior to the eye of a trained observer in this task and avoids the variability in visual perception created by differing illuminating light intensities. This relatively inexpensive tool is available on-line for rapid analysis within seconds and can, for the first time, provide quantitative information during routine angioscopic imaging.
We gratefully acknowledge the assistance of Hans Schuurbiers and Frank van der Panne in technical areas, of Theo Steinen and Eric Boersma in data analysis, of Yukio Asaki in data collection, of Mimi Platt in manuscript preparation, and most importantly, of Hans van Daele in his superb and multiple contributions to the construction of the thrombus models.
- Received December 18, 1996.
- Revision received June 19, 1997.
- Accepted June 26, 1997.
- Copyright © 1997 by American Heart Association
Falk E. Unstable angina with fatal outcome: dynamic coronary thrombosis leading to infarction and/or sudden death. Circulation. 1985;71:699-708.
Fuchs J, Cannon CP, for the TIMI 7 Investigators. Hirulog in the treatment of unstable angina. Circulation. 1995;92:727-733.
The TIMI IIIB Investigators. Effects of tissue plasminogen activator and a comparison of early invasive and conservative strategies in unstable angina and non-q-wave myocardial infarction: results of the TIMI IIIB trial. Circulation. 1994;89:1545-1556.
Theroux P, Waters D, Qiu S, McCans J, deGuise P, Juneau M. Aspirin versus heparin to prevent myocardial infarction during the acute phase of unstable angina. Circulation. 1993;88:2045-2048.
Fuster V, Badimon L, Cohen M, Ambrose JA, Badimon JJ, Chesebro J. Insights into the pathogenesis of acute ischemic syndromes. Circulation. 1988;77:1213-1220.
Stratton JR, Ritchie JL. Indium-111 platelet imaging of left ventricular thrombi: predictive value for systemic emboli. Circulation. 1990;81:1182-1189.
Stratton JR. Thrombosis imaging with indium-111 labeled platelets. In: Schelbert HR, Skorton PJ, Wolf PJ, eds. Cardiac Imaging: Principles and Practice: A Companion to Braunwald’s Heart Disease Second Edition. Philadelphia, Pa: WB Saunders Co; 1996.
White CJ, Ramee SR, Collins TJ, Escobar A, Karsan A, Shaw D, Jain SP, Bass TA, Heuser RR, Teirstein PS, Bonan R, Walter PD, Smalling RW. Coronary thrombi increase PTCA risk: angioscopy as a clinical tool. Circulation. 1996;93:253-258.
Feld S, Ganim M, Carell ES, Kjellgren O, Kirkeeide RL, Vaughn WK, Kelly R, McGhie AI, Kramer N, Loyd D, Anderson HV, Schroth G, Smalling RW. Comparison of angioscopy, intravascular ultrasound imaging and quantitative coronary angiography in predicting clinical outcome after coronary intervention in high risk patients. J Am Coll Cardiol. 1996;28:97-105.
Bauters C, Lablanche JM, McFadden EP, Hamon M, Bertrand ME. Relation of coronary angioscopic findings at coronary angioplasty to angiographic restenosis. Circulation. 1995;92:2473-2479.
de Feyter PJ, Ozaki Y, Baptista J, Escaned J, Di Mario C, de Jaegere PPT, Serruys PW, Roelandt JR. Ischemia-related lesion characteristics on patients with stable or unstable angina: a study with intracoronary angioscopy and ultrasound. Circulation. 1995;92:1408-1413.
Nissen SE. Radiographic principles in cardiac catheterization. In: Roubin GS, Califf RM, O’Neill W, Phillips H, Stack R, eds. Interventional Cardiac Catheterization: Principles and Practice. New York, NY: Churchill Livingstone, Inc; 1993:409-425.
Michaelis L. Der acetat-veronal puffer. Biochemische Zeitschrift. 1931;234:139-141.
Lehmann KG, Oomen JA, Slager CJ, deFeyter PJ, Serruys PW. Chromatic distortion during angioscopy:assessment and correction by quantitative colorimetric angioscopic analysis. In press.
Cotran RS, V Kumar, SL Robbins. Robbins’ Pathologic Basis of Disease. Philadelphia, Pa: WB Saunders Co; 1989:99-120.
Jang IK, Gold HK, Ziskind AA, Fallon JT, Holt RE, Leinbach RC, May JW, Collen D. Differential sensitivity of erythrocyte-rich and platelet-rich arterial thrombi to lysis with recombinant tissue-type plasminogen activator: a possible explanation for resistance to coronary thrombolysis. Circulation. 1989;79:920-928.
Waters D, Lam JYT. Is thrombolytic therapy striking out in unstable angina? Circulation. 1992;86:1642-1644.
Inoue K, Kuwaki K, Ochiai K, Ueda K, Takano E, Minato H. Percutaneous transluminal coronary angioscopy as the guiding therapy for intracoronary thrombolysis and angioplasty. In: Vogel JHK, King SBI, eds. Interventional Cardiology: Future Directions. St Louis, Mo: CV Mosby Co; 1989:1-11.
den Heijer P, Foley DP, Hillege HL, Lablanche JM, van Dijk RB, Franzen D, Morice MC, Serra A, de Scheerder IK, Serruys PW, Lie KI. The “Ermenonville’ classification of observations at coronary angioscopy: evaluation of intra- and inter-observer agreement, European Working Group on Coronary Angioscopy. Eur Heart J. 1994;15:815-822.
Ritchie AC. Boyd’s Textbook of Pathology. Philadelphia, Pa: Lea and Febiger; 1990:117-132.