Author + information
- Received January 13, 2014
- Revision received June 25, 2014
- Accepted July 16, 2014
- Published online January 1, 2015.
- Jennifer E. Phipps, PhD∗,
- Deborah Vela, MD†,
- Taylor Hoyt, BA∗,
- David L. Halaney, BS∗,‡,
- J. Jacob Mancuso, MD∗,
- L. Maximilian Buja, MD†,
- Reto Asmis, PhD∗,
- Thomas E. Milner, PhD§ and
- Marc D. Feldman, MD∗,‡∗ ()
- ∗University of Texas Health Science Center San Antonio, San Antonio, Texas
- †Texas Heart Institute, Houston, Texas
- ‡Department of Veterans Affairs, South Texas Veterans Health Care System, San Antonio, Texas
- §University of Texas at Austin, Austin, Texas
- ↵∗Reprint requests and correspondence:
Dr. Marc D. Feldman, University of Texas Health Science Center San Antonio, 7703 Floyd Curl Drive, MSC 7872, San Antonio, Texas 78229.
Objectives This study hypothesized that bright spots in intravascular optical coherence tomography (IVOCT) images may originate by colocalization of plaque materials of differing indexes of refraction. To quantitatively identify bright spots, we developed an algorithm that accounts for factors including tissue depth, distance from light source, and signal-to-noise ratio. We used this algorithm to perform a bright spot analysis of IVOCT images and compared these results with histological examination of matching tissue sections.
Background Bright spots are thought to represent macrophages in IVOCT images, and studies of alternative etiologies have not been reported.
Methods Fresh human coronary arteries (n = 14 from 10 hearts) were imaged with IVOCT in a mock catheterization laboratory and then processed for histological analysis. The quantitative bright spot algorithm was applied to all images.
Results Results are reported for 1,599 IVOCT images co-registered with histology. Macrophages alone were responsible for only 23% of the bright spot–positive regions, although they were present in 57% of bright spot–positive regions (as determined by histology). Additional etiologies for bright spots included cellular fibrous tissue (8%), interfaces between calcium and fibrous tissue (10%), calcium and lipids (5%), and fibrous cap and lipid pool (3%). Additionally, we showed that large pools of macrophages in CD68+ histology sections corresponded to dark regions in comparative IVOCT images; this is due to the fact that a pool of lipid-rich macrophages will have the same index of refraction as a pool of lipid and thus will not cause bright spots.
Conclusions Bright spots in IVOCT images were correlated with a variety of plaque components that cause sharp changes in the index of refraction. Algorithms that incorporate these correlations may be developed to improve the identification of some types of vulnerable plaque and allow standardization of IVOCT image interpretation.
Intravascular optical coherence tomography (IVOCT) is the highest resolution technique available to image vulnerable plaque in coronary arteries (1). In 2003, Tearney et al. (2) published a quantitative study that demonstrated that bright spots in IVOCT images could represent macrophages. They showed that regions with an increased normalized standard deviation (NSD) correlated with areas in human aortic plaque that stained positively for macrophages in immunohistochemical studies. These findings were used to interpret subsequent clinical and animal IVOCT studies (3–7). However, the finding that macrophages cause increased NSD has not been validated histologically by other research groups (8).
Concerns have been raised about the specificity of bright spots in identifying macrophages, despite the use of shadowing behind the bright spots as a secondary confirmation of macrophage identification (1). For example, other plaque components appear to cause bright spots, including fibrin accumulations (9), neoatherosclerosis in previously placed stents (9), elastic lamina, cholesterol crystals, and microcalcifications (1,10). Furthermore, the NSD method was designed to be accurate only in fibrous caps and does not apply to the detection of macrophages in deeper arterial structures (2).
Tearney et al. (2) proposed that regions of high NSD represent areas in which the optical index of refraction (IR) has a higher heterogeneity; macrophages appear as bright spots because of the difference in IR between them and the surrounding fibrous tissue. Using this hypothesis, we proposed that bright spots can arise from constituents other than macrophages in which a sharp change in IR occurs at interfaces between plaque components. This sharp change in IR occurs in multiple scenarios: lipid (IR 1.33) mixed with calcified cores (IR 1.65), lipid mixed with fibrous tissue (IR 1.47), and cellular fibrous tissue rich in proteoglycans (IR 1.33)—each of which may generate bright spots in IVOCT images on the basis of this hypothesis.
We developed an algorithm that can be applied to the entire depth of the artery to enable quantitative identification of bright spots in IVOCT images of human atherosclerotic plaque. In the present study, we used this algorithm to analyze bright spots in images of human atherosclerotic plaque and compared these results with those of corresponding histological sections. In addition, we further examined the hypothesis that abrupt changes in plaque components of differing IR result in the generation of bright spots in IVOCT images.
Ten human hearts (from 3 women and 7 men) at autopsy within 24 h of death were examined. The average age at death was 65 ± 11 years. The cause of death was cardiac in 6 cases. We imaged 14 coronary arteries (10 left anterior descending arteries [LADs] and 4 right coronary arteries [RCAs]). The institutional review board at the University of Texas approved this study.
The human heart catheterization laboratory was recreated with a custom IVOCT system (Volcano Corporation, San Diego, California) to access the raw signal data. The IVOCT system has a 1,310-nm swept source laser (HSL-1000, Santec, Hackensack, New Jersey) and a bandwidth of 80 nm scanning at a repetition rate of 34 kHz. The measured free-space axial resolution was 20 μm with a 2.8-mm scan depth. The IVOCT signal was sampled with a linear k-space clock to allow real-time OCT image acquisition and display. A fluoroscopy system (GE Healthcare, Little Chalfont, United Kingdom) and a chamber designed to maintain the tissue at 37°C were used. Left and right coronary 6-F guide catheters were sewn into the coronary ostia, 0.014-inch guidewire access to the coronary arteries was gained under fluoroscopic guidance, and a stent was deployed 80 mm from the guide catheter tip as a fiduciary marker. IVOCT pullbacks were acquired from the stent to the guide catheter (80-mm total pullback length). The LAD and RCA were imaged. Following imaging, the RCA and LAD were perfusion-fixed with formalin at 100 mm Hg. The left circumflex artery was not imaged due to its tortuosity in the ex vivo heart.
The LADs and RCAs were perfusion-fixed with 10% neutral-buffered formalin, excised from each heart, individually radiographed on a Faxitron MX-20 (Faxitron Bioptics LLC, Tucson Arizona), and decalcified overnight with Cal-Rite (Richard Allen Scientific, Kalamazoo, Michigan), if necessary. The arterial segments were sliced into 2- to 3-mm–thick rings and further processed on a Tissue-Tek vacuum infiltration processor (Sakura Finetek USA, Torrance, California) for standard paraffin-embedded sections. An average of 25 rings was generated from each artery. Serial tissue sections (5-μm thick) were cut at 150-μm intervals and stained with hematoxylin and eosin, modified Movat pentachrome, and Von Kossa. Anti-CD68 (Dako North America, Inc., Carpinteria, California) and anti–alpha smooth muscle cell actin (Sigma-Aldrich, St. Louis, Missouri) antibodies were used in immunohistochemical studies to identify macrophages and smooth muscle cells, respectively.
IVOCT and histology co-registration
Each histological ring was matched to a respective IVOCT frame. Co-registration was performed between IVOCT images and histological sections on the basis of the following: 1) 2 fiducial landmarks—a stent deployed at the distal end of the pullback and the sewn-in guide catheter at the proximal edge—that were visible in IVOCT images, fluoroscopy, and radiography before histopathologic processing; and 2) the physical position of IVOCT images in the pullbacks measured against the estimated distance in microns from the fiducial landmarks in the tissue sections. Additionally, anatomic landmarks (e.g., arterial branches or calcification patterns when present) and luminal geometric features further aided co-registration. Two researchers independently co-registered the IVOCT images and histology, and discrepancies were discussed to find agreement between both co-registrations.
In IVOCT images, bright bands <65 μm thick that covered diffusely shadowed regions were identified as thin-cap fibroatheromas (TCFAs). Histological TCFAs were identified by fibrous caps <65 μm thick that covered lipid or necrotic cores.
Histological composition of bright spot–containing areas
Regions within the arterial wall that elicited bright spots after application of the algorithm were first categorized by whether macrophages were present or not. Next, each of these macrophage-positive or macrophage-negative bright spot sources were classified into the following 4 broad categories: 1) hypocellular or acellular collagen-rich fibrous tissue (mesh-like collagen-rich areas mixed with lipid or the fibrous cap of fibrocalcific plaques); 2) cellular fibrous tissue (as found in intimal thickening or early lesions with high smooth muscle and proteoglycan content); 3) cholesterol clefts within necrotic cores; and 4) areas of layering or interface (as observed in remodeled plaque ruptures, at the interfaces between calcium and surrounding tissue, between lipid and calcium in fibrocalcific plaques, at the interface of necrotic or lipid cores and the overlying fibrous cap, at neovascularization sites and the media, or at the elastic lamina intimal/medial or medial/adventitial interface).
Bright spot quantitative detection
The detection method is outlined in Figure 1. First, we measured the distance between the lumen edge and the catheter for each A-scan per frame. Next, the mean of those distances was calculated for each frame. To account for variations in signal intensity that occur as the catheter moves closer or further away from the lumen, we calculated 2 reference A-scans by averaging all A-scans that were less than or greater than the mean distance to the catheter. Then, to account for varying signal-to-noise ratio (SNR), the reference A-scans were normalized (divided by the difference between the maximum and minimum values of each frame). We compared each A-scan to the averaged and normalized reference A-scan that corresponded to whether its catheter to lumen edge distance was less than or greater than the mean; this provided a threshold to identify bright spots on the basis of tissue depth, distance from catheter, and SNR of the IVOCT system and catheter.
Four statistical analyses were performed: 1) interobserver and intraobserver variability between 2 expert IVOCT readers who evaluated the unprocessed IVOCT frames for identification of bright spots; 2) joint probability of agreement between 1 expert IVOCT reader and the bright spot algorithm; 3) sensitivity and specificity of 1 expert IVOCT reader for identifying macrophages compared with the gold standard of histology; and 4) sensitivity and specificity of the bright spot algorithm for identifying macrophages compared with the gold standard of histology. Sensitivity and specificity calculations were performed at 2 depths, superficial (<100 μm) and deep (>100 μm).
We imaged 14 coronary arteries (10 LADs and 4 RCAs) from 10 human hearts, generating 300 IVOCT images per vessel. After application of the algorithm, we observed 2,206 IVOCT frames with bright spots. A total of 1,111 of the IVOCT frames with bright spots were co-registered with histology, and 1,700 regions within these 1,111 frames caused distinct sources of bright spots. A total of 488 IVOCT frames without bright spots were co-registered to histology for negative control. See Online Table 1 for the morphological classification of the bright spot locations. When only the raw, unprocessed IVOCT frames were reviewed for identification of bright spots between 2 expert IVOCT readers, intraobserver variability was 88% and interobserver variability was 76%.
Macrophage detection by the bright spot algorithm
Using our quantitative algorithm and histological examination, we characterized the role of macrophages in the origin of bright spots in IVOCT images (Table 1, Figure 2). Macrophages alone were responsible for bright spots in 391 regions (23%). A combination of macrophages and other etiologies were the source of bright spots in an additional 574 regions (34%). In total, macrophages were present in 57% of bright spot–positive regions. See the Online Appendix for a description of how NSD images were generated and for a discussion of the NSD method compared with our bright spot method for detection of macrophages.
Large, dense CD68+ areas on histology corresponded to dark regions (Figures 2 and 3⇓) in the associated IVOCT image (125 of the CD68+ regions). Additionally, we identified 61 regions in which macrophages were located too deep in the tissue or too far from an eccentric catheter position to be visualized by IVOCT (Figure 2I). There were also 89 regions of CD68 positivity in histological sections that did not cause bright spots and thus were not identified by the bright spot algorithm. Lastly, in 186 regions, macrophages were depicted as bright spots that caused superficial shadowing. Of these 186, 115 (62%) were found in regions where calcium was colocalized with the macrophages (Figure 4D) and 4 were in regions where cholesterol crystals were colocalized with the macrophages.
The sensitivity and specificity of the bright spot algorithm compared with an expert IVOCT reader for identification of macrophages, with histology as a gold standard, are summarized in Table 2. The IVOCT reader used unprocessed IVOCT images for the analysis. The algorithm was more sensitive to bright spots (80%), implying that the presence of macrophages was more often correctly identified by the algorithm than by an expert reader; however, the algorithm was less specific (49%), as anticipated, because it detects sources of bright spots due to etiologies other than macrophages. This is also supported by the joint probability of agreement that was calculated between the algorithm and expert reader (53%), which reflects the fact that the algorithm identified regions of macrophages that the expert IVOCT reader would have missed.
Algorithm identified bright spots not colocalized with macrophages
Bright spots were also associated with fibrous tissue (cellular 8%; acellular 8%) (Table 3, Figure 5G), areas of plaque layering between old and new fibrous tissue in remodeled plaques (4%) (Figure 5A), calcified lipid cores with noncalcified remaining lipid (5%) (Figure 4A), and the fibrous cap and lipid pool interface (3%) (Figure 5D).
Algorithm identified bright spots colocalized with TCFAs
Bright spots occurred in 175 regions of IVOCT frames morphologically classified as TCFAs (Table 4). Of these, 165 regions were also colocalized with CD68 positivity. Figure 6 demonstrates bright spots originating in a TCFA; the bright spots in this case were caused by both macrophages in the fibrous cap and the fibrous cap interface with the lipid core. There were only 10 regions of bright spots in TCFA IVOCT frames that were not colocalized with macrophages (data not shown); of these regions, the source of bright spots was the fibrous cap and lipid or necrotic core interface (n = 4), the fibrous cap and calcium border (n = 3), and acellular fibrous tissue (n = 3).
We confirmed that IVOCT bright spots can be caused by macrophages; however, we also identified new alternative etiologies. Our findings suggest that IVOCT bright spots can be generated in areas characterized by sharp changes in IR. Our results support the principle that spatial gradients in IR are responsible for enhanced light scattering that results in bright regions in IVOCT images. Thus, most bright spots in IVOCT images are not caused by macrophages but originate from a mixture of atherosclerotic components that have maximal differences in optical IR. Of the sources of bright spots identified, all were found in regions with known sharp gradients of IR. In the case of layering between old and new fibrous tissue, the colocalization of different types of collagen fibers or the degree of maturation is responsible for the shift in IR (11). In particular, after a rupture has healed, collagen type III is replaced by type I and results in a band of high backscattering signal between the layers of tissue—this is likely due to the greater optical density of collagen type I in comparison to type III (1).
Some macrophages appear dark
Our finding that large pools of macrophages appear dark supports the observation that homogeneous material in plaque, even groups of macrophages as shown in this study, can have a homogeneous IR and thus would not be expected to cause bright spots. Moreover, this finding implies that a large pool of macrophages that have engulfed lipid (foam cells) will appear dark on IVOCT images, similar to lipid pools, and may not be easily identified by IVOCT (Figures 2 and 3). Thus, the juxtaposition of foam cells with a low IR next to the fibrous cap with a higher IR may have been the origin of bright spots identified in previous studies (2).
An alternative hypothesis for explaining why not all macrophages appear as bright spots in IVOCT images involves differences in macrophage subsets. M1 macrophages, considered the “classic” phenotype, are thought to be proinflammatory and engulf lipid to form foam cells, whereas M2 macrophages, considered anti-inflammatory, contain smaller vesicles of engulfed lipid (12) and a higher density of mitochondria (13). Thus, M1 macrophage foam cells may appear as shadows or dark regions because of the large amount of intracellular lipid, but M2 macrophages may appear bright because of a higher density of light-scattering mitochondria (14,15). Further study of how M1 and M2 macrophages appear in IVOCT images is needed. Additionally, combining IVOCT with other imaging techniques that have a higher specificity for lipid, fibrous tissue, and macrophages, such as two-photon luminescence (16) or fluorescence lifetime imaging (17), may provide enhanced contrast for distinguishing between the subtypes of macrophages.
Macrophages alone are responsible for few regions of brightness
Although 57% of all regions with algorithm-defined bright spots were CD68+, 34% of all regions that were CD68+ were also colocalized with other tissue components that caused bright spots in the absence of any macrophages (Table 3). It is uncertain whether bright spots in the presence of both macrophages and those other components were caused by the macrophages or by another etiology. Thus, only 23% of the bright spots were definitely caused by macrophages.
Mechanism of superficial shadowing caused by macrophages
In addition to appearing as bright spots in IVOCT images, macrophages can cause shadowing that may appear as a lipid pool (1,18). Although our algorithm for identifying bright spots does not directly search for shadowing behind the bright spots, we identified 186 regions of bright spots that caused shadows; 119 were located in regions characterized by the colocalization of microcalcification or cholesterol crystals and macrophages (Figures 4D to 4F). We believe that shadowing was actually caused by microcalcification and/or cholesterol crystals. Considering Mie scattering, which describes the way light scatters from symmetrical objects, smaller features with higher IRs will cause increased shadowing. Thus, small cholesterol crystals and microcalcifications, both of which have high IRs compared with the other plaque components, can cause shadows. In addition to ingesting lipid, macrophages can engulf microcalcifications (19) and cholesterol crystals (20). Macrophages alone do not have optical properties that would cause a shadow, unless they have engulfed a microcalcification or plaque component that has an IR substantially higher than that of lipid. The high IR of cholesterol clefts is also consistent with the bright spots observed within necrotic cores. Thus, we propose that the type and distribution of engulfed material affects the shadowing by macrophages.
Algorithm-identified bright spots in TCFAs
Most of the bright spot regions found in TCFA IVOCT frames (165 of 175 regions) were colocalized with macrophages. Interestingly, the majority were caused by either macrophage-rich areas with fibrous tissue or the fibrous cap/lipid pool interface, which is also where macrophages are often found. Thus, it can be difficult to distinguish when bright spots found in TCFAs are caused by macrophages or the fibrous cap/lipid pool interface.
Statistics regarding identification of macrophages
The algorithm was more sensitive at detecting the true presence of macrophages than an expert reader, demonstrating its value. This result is supported by the joint probability of agreement between the expert IVOCT reader and the algorithm—only 53%, which quantifies the finding that the algorithm identified regions of macrophages that the expert IVOCT reader missed. The reduced specificity of the algorithm was anticipated because it also detects causes of bright spots other than macrophages, due to differences in IR of comingled plaque components, the hypothesis of this paper. The increased specificity (76%) of the IVOCT reader showed that an expert reader can distinguish between bright spots caused by macrophages and bright spots caused by other sources. One way to increase the accuracy for identification of macrophages is to allow an expert reader to sort through the bright spot processed images and disregard images that have bright spots obviously not caused by vulnerable plaque morphologies. Future advances in multimodal or more advanced image processing methods could discard regions of brightness from nonvulnerable plaque types automatically. For example, the accuracy of identifying vulnerable plaque could be improved by combining IVOCT with a novel technique that provides biochemical specificity such as time-resolved fluorescence (21) or Raman spectroscopy (22).
Algorithm-identified bright spots at the lipid and calcium interface
We found that regions with lipid intermingled in fibrocalcific plaque can also generate bright spots. We believe the underlying reason for this finding is that fibrocalcific plaque shows rich signal heterogeneity within the calcified cores. Lipid cores initially develop microcalcifications, which have been associated with vulnerable plaque (23). These microcalcifications coalesce into larger calcifications; the intermingling of lipid with a low IR and microcalcifications with a high IR is responsible for the complexity and heterogeneity of some calcification sites. Once these cores become homogeneous calcified plates (24), the bright reflections may resolve. However, homogeneous calcified plates in the absence of residual pools of lipid are not frequently found in human atherosclerosis and are not commonly seen during OCT imaging.
Advantages of our bright spot algorithm
Because current methods for interpreting IVOCT images are qualitative, interpretation varies widely. This is especially true when identifying TCFA because distinguishing lipid from calcium can be difficult (25–27) and macrophages can cause shadowing that falsely appears as a lipid core (18). Developing algorithms to quantify plaque composition is critical if IVOCT is to be used for accurately identifying plaque composition. Furthermore, several optical properties must be considered when identifying IVOCT bright spots. Light attenuates through tissue at an exponential rate dependent upon both depth and tissue composition, and the intensity of light reflections varies with distance from the catheter. Finally, the SNR will vary between IVOCT images and pullbacks due to differences in the power of laser sources, manufacturing variability between catheters, and other clinical variables such as residual blood in the field. Thus, tissue depth, distance from the catheter, and the SNR are factors that should be considered when identifying IVOCT bright spots, and all are taken into account with the bright spot algorithm presented here.
Although our study used algorithm-defined bright spots within IVOCT images to focus our histology examination, the opposite was not performed. Second, as with all imaging studies, histological co-registration is complex, and the possibility of error exists. Additionally, the histological sections were 5 μm thick, whereas the IVOCT images were separated by 270 μm of tissue. However, serial sectioning throughout the ROIs greatly improved the accuracy of our co-registration. Lastly, distortions due to histological processing are always possible sources of error in co-registration and IVOCT image interpretation.
We developed a novel quantitative technique to identify bright spots in IVOCT images. Our findings indicated that not all bright spots are caused by macrophages; rather they can be generated by a combination of plaque components that create sharp changes in the IR. Moreover, we found that macrophage foam cells can be seen as dark regions on IVOCT images. Our study underscores the importance of developing more discerning algorithms. Software that incorporates our quantitative technique may improve the identification of some types of vulnerable plaque and may enable the standardization of IVOCT image interpretation.
This study was funded by the Veterans Health Administration merit grant I01 BX000397, Clayton Foundation, Janey and Dolph Briscoe Division of Cardiology at the University of Texas Health Science Center (NIH T32 HL007446), American Heart Association (13POST17080074), and the Biomedical Engineering Advancement fund at the University of Texas at Austin. The intravascular optical coherence tomography system used in this study was provided by Volcano Corporation. All authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- index of refraction
- intravascular optical coherence tomography
- left anterior descending artery
- normalized standard deviation
- right circumflex artery
- signal-to-noise ratio
- thin-cap fibroatheroma
- Received January 13, 2014.
- Revision received June 25, 2014.
- Accepted July 16, 2014.
- American College of Cardiology Foundation
- Tearney G.J.,
- Regar E.,
- Akasaka T.,
- et al.
- Tearney G.J.,
- Yabushita H.,
- Houser S.L.,
- et al.
- MacNeill B.D.,
- Jang I.K.,
- Bouma B.E.,
- et al.
- Tahara S.,
- Morooka T.,
- Wang Z.,
- et al.
- Raffel O.C.,
- Tearney G.J.,
- Gauthier D.D.,
- Halpern E.F.,
- Bouma B.E.,
- Jang I.K.
- Raffel O.C.,
- Merchant F.M.,
- Tearney G.J.,
- et al.
- Ali Z.A.,
- Roleder T.,
- Narula J.,
- et al.
- Falk E.,
- Nakano M.,
- Bentzon J.F.,
- Finn A.V.,
- Virmani R.
- Tavakoli S.,
- Zamora D.,
- Ullevig S.,
- Asmis R.
- van der Meer F.J.,
- Faber D.J.,
- Baraznji Sassoon D.M.,
- Aalders M.C.,
- Pasterkamp G.,
- van Leeuwen T.G.
- van Soest G.,
- Regar E.,
- Goderie T.P.,
- et al.
- Nadra I.,
- Mason J.C.,
- Philippidis P.,
- et al.
- Maldonado N.,
- Kelly-Arnold A.,
- Vengrenyuk Y.,
- et al.
- Rieber J.,
- Meissner O.,
- Babaryka G.,
- et al.