In the past decade, drug-eluting stents (DES) have led to a dramatic improvement in the percutaneous treatment of coronary atherosclerotic disease, where the incidence of in-stent restenosis has been significantly reduced to <10% at 9- to 12-month follow-up (1- 2). Despite the major advancements, there remains an infrequent, but real risk for life-threatening complications arising from late stent thrombosis (LST), primarily attributed to incomplete endothelial coverage of stent struts (3).

Both human autopsy (4- 5) and clinical imaging studies (6- 7) have emphasized the importance of delayed arterial healing and underlying plaque morphology as contributing factors to LST, where the best classifier of endothelial coverage is the ratio of uncovered to total stent struts (RUTSS) (8). It is therefore possible that improvements in imaging modalities would serve as an identifier of healing with DES and serve as a predictor of late complications.

Intravascular imaging devices that measure backscattered light or optical echoes derived from an infrared light source directed at the arterial wall are receiving much attention because of their high resolution (i.e., 10 to 20 μm). In pre-clinical animal models, Suzuki et al. (9) and Murata et al. (10) examined the accuracy and reproducibility of optical coherence tomography (OCT) for detecting neointimal coverage of stent struts. Moreover, Templin et al. (11), using optical frequency domain imaging (OFDI), a second-generation OCT device with better image acquisition by removing the depth degeneracy (12), recently showed an excellent correlation of stent strut coverage between OFDI and histology by scanning electron microscopy in a porcine model. The accuracy of OFDI has not been previously examined. In the present study, we sought to evaluate the capability of OFDI to recognize critical morphological features that may be associated with long-term stent performance by ex vivo interrogation of human stented coronary arteries at autopsy.

Unfixed or formalin-fixed hearts or vessels with intracoronary stents received in diagnostic consultation were studied; stents implanted <1 month at the time of autopsy were excluded from the analysis (Table 1).

Initially, a 0.014-in guidewire was introduced into the vessel followed by an intravascular ultrasound (IVUS) catheter (2.6-F). IVUS images of the entire stent and adjacent 5-mm proximal and distal nonstented segments were acquired at a pullback rate of 0.5 mm/s (30 frame/s). Following IVUS interrogation, the catheter was removed, and an OFDI catheter (2.4-F) was similarly introduced; sequential images were then acquired at a pullback rate of 15 mm/s (120 frame/s). Image acquisition of stented vascular segments was performed using IVUS (frequency 40 MHz, VISIWAVE, Terumo Corporation, Tokyo, Japan), and Terumo OFDI system (Terumo Corporation). Acquired images were stored on the system hard drive for the later offline analysis. When unfixed hearts were received (cases 3 and 9 in Table 1), imaging was repeatedly performed before and after formalin fixation. Following artery interrogation, the stented vessels were sent to histological processing (for the details, see Online Appendix).

IVUS and OFDI images were systematically co-registered to histological sections by 2 separate investigators (M.N. and M.V.), beginning at the most distal nonstented segment with consideration of the distance along the longitudinal axis, pullback speed, and frame rate. Adjustments were made using the stent and luminal configuration or anatomical landmarks such as vessel branches or presence of calcification, thus improving the accuracy of registration. Twenty-seven histological sections were excluded on the basis of cutting artifacts, co-registration difficulty, and large branch or existence of thrombi. IVUS and OFDI images acquired from fixed coronary segments were utilized for morphometry and tissue characterization where a total of 134 pairs of co-registered (matching) images including 1,656 struts (histology), 1,098 struts (IVUS), and 1,322 struts (OFDI) were available for the main analysis (5).

Cross-sectional areas (stent and lumen) at each level were measured in co-registered images from histology, IVUS, and OFDI using digital morphometry (IPLab, Scanalytics, Rockville, Maryland). Area measurements were used to calculate vessel layer areas with the following formulas; Neointimal Area = Stented Area − Lumen Area, % Stenosis = 100 × (1 − [Lumen Area/Stented Area]).

Neointimal thickness on histology and IVUS was defined as the distance from the inner surface of each stent strut to the luminal border, whereas for OFDI, the distance between the axial and lateral center of stent strut reflection and the luminal border was used, as previously reported (10). The mean, minimum, and maximum values of neointimal thickness for each section was calculated from each co-registered image.

The number of uncovered struts and total struts per section were counted together with the ratio of uncovered to total struts per section (RUTSS) as previously described (8). Recognition of uncovered struts in IVUS and OFDI was based on the visual determination that struts were detached (free) from the luminal surface or that struts were apposed to the arterial wall with an apparent discontinuity of neointimal tissue over and/or near the strut edge (Figure 1). Uncovered struts by histology were defined by the lack of neointimal coverage and/or surface endothelium (8).

To further evaluate the utility of OFDI to recognize covered versus uncovered struts, a detailed analysis was performed by 2 investigators blinded to histology (F.O. and M.T.) on a subset of histological sections with at least 1 uncovered strut (n = 13 sections), where a total of 97 covered struts and 43 uncovered struts were identified by histology. The stent struts of the remaining 121 sections were fully covered with neointima and were excluded so that only the sections with minimal neointima were examined. In this analysis, co-registered and adjacent frames in sequential OFDI images were deliberately investigated to identify individual struts that matched with histology (5).

Histological slides from stented arteries were reviewed for fibrin accumulation, foamy macrophages, and in-stent atherosclerosis including lipid/necrotic core, calcification, excessive inflammation (hypersensitivity), intimal disruption, organized thrombus, and angiogenesis; and the OFDI appearance of those features was descriptively characterized. For further evaluation, OFDI signals from tissues facing the lumen were quantitatively analyzed by a modified method from Kume et al. (13). The OFDI signal intensity was calculated from the luminal surface to 0.3- to 0.4-mm depth, and fitted to an approximate exponential formula (y = A × exp^−Bx) by the least-square method, where index A represents a “peak intensity,” and index B reflects an “attenuation rate” (i.e., reduction of signal intensity according to depth). For each section, 3 regions of interest were selected from the area consisting of the aforementioned in-stent features. The values of peak intensity and attenuation rate were measured from 3 regions of interest, and averaged per section. The OFDI signal data were also acquired from normal neointima of stents implanted ≤6 months or >1 year. Normal neointima was defined as a smooth muscle cell (SMC)-rich tissue with extracellular matrix (ECM) containing collagens and proteoglycans but had an absence of fibrin accumulation, foamy macrophages, lipid core/necrotic core, calcification, excessive inflammation (hypersensitivity), intimal disruption, and organized thrombus. The areas with nonuniform rotational distortion and/or other artifacts were excluded from this analysis. Prior to this evaluation, all OFDI images were adjusted to the same default setting, and the analysis was performed by ImageJ software (National Institutes of Health, Rockville, Maryland). Data from unfixed vessels in 2 cases were used to assess the effect of formalin fixation on OFDI signal reflection, and no significant difference was identified between unfixed and fixed tissues (for the details, see Online Appendix).

The values were expressed as mean ± SD for continuous valuables and as median (interquartile range [IQR]) for discrete valuables unless otherwise designated. One-way analysis of variance with post hoc Tukey-Kramer test was used to calculate the significance of differences between normally distributed data. Non-normally distributed data were compared using a Wilcoxon Kruskal-Wallis test. The correlation of the OFDI or IVUS measurements versus histology was calculated by Spearman rank correlation coefficient. The agreement of measurements was assessed by Bland-Altman analysis (14). In the per-strut analysis of neointimal coverage, sensitivity and specificity for detection of covered/uncovered struts by OFDI were calculated for each reviewer without consideration of correlation within individuals. The degree of agreement between 2 reviewers was quantified by the Cohen's κ test for concordance. A value of p < 0.05 was considered statistically significant.

Area measurements of the stent size (mm²), lumen (mm²), neointima (mm²), and % stenosis derived from IVUS and OFDI were similar to those derived from histology (Table 2). Bland-Altman analysis confirmed good agreements of neointimal area measurements by OFDI and IVUS versus histology with a marginally narrower range of “limits of agreement” in OFDI (1.09 mm² [95% confidence interval (CI): 0.92 to 1.25 mm²]) than in IVUS (1.34 mm² [95% CI: 1.14 to 1.54 mm²]) (Figure 2A). The mean values of subtraction OFDI and IVUS versus histology were −0.10 mm² [95% CI: −0.20 to −0.01 mm²] and −0.19 mm² [95% CI: −0.31 to −0.01 mm²], respectively.

Mean and maximum neointimal thickness (mm) values for each section showed a good correlation for both OFDI and IVUS (Table 2). However, the correlation with minimum neointimal thickness per section for IVUS (R² = 0.386, p < 0.001) was not as high as that of OFDI (R² = 0.667, p < 0.001), where the “limits of agreement” was lower with OFDI (0.105 mm [95% CI: 0.089 to 0.121 mm]) than with IVUS (0.174 mm [95% CI: 0.148 to 0.200 mm]), whereas the mean value was higher for OFDI (0.035 mm [95% CI: 0.026 to 0.044 mm]) than for IVUS (0.001 mm [95% CI: −0.014 to 0.016 mm]). For further evaluation, we performed a subanalyses based on minimum values of neointima thickness > or <80 μm as described in a previous animal study (10) (Figure 2B). In a subset analysis of sections with neointima thickness <80 μm (n = 82 sections), OFDI showed a poor correlation with histology (R² = 0.201, p < 0.001), whereas for neointima thickness >80 μm (n = 52 sections), the correlation was much greater (R² = 0.554, p < 0.001).

Although OFDI was superior than IVUS at recognizing the total number of struts per section, the values were still lower when compared with histology (OFDI: median = 10 [IQR: 7 to 12], IVUS: 8 [7 to 10], histology: 12 [9 to 15]; p < 0.001). In RUTSS calculations, OFDI provided a good correlation with histology (R² = 0.656, p < 0.001), whereas IVUS failed to detect struts with relatively minimal neointimal coverage (R² = 0.241, p < 0.001), thus resulting in a greater number of false positives (Table 2).

In another comparison, histological sections with documented covered (n = 97) and uncovered struts (n = 43) were intentionally matched with the appropriate OFDI images. Uncovered struts in histology were identified by 2 OFDI reviewers with a sensitivity of 79.1% (95% CI: 70.3% to 85.6%) and 76.7% (95% CI: 67.5% to 82.0%) and specificity of 96.9% (95% CI: 93.0% to 98.9%) and 95.9% (95% CI: 91.8% to 98.2%), respectively. The interobserver agreement between 2 reviewers was excellent, achieving a Cohen's κ value of 0.85 (95% CI: 0.72 to 0.93) (Table 3).

Strut coverage confirmed by histology but misclassified by OFDI included 5 of 97 covered struts in addition to 12 of 43 uncovered struts by either of the 2 OFDI reviewers. The misclassification of covered struts was mainly attributed to a limited ability of OFDI to resolve unhealed struts characterized by collections of fibrin and inflammatory cells (10 of 12 struts), which were deemed uncovered by histology (Figure 3). Moreover, thin layers of neointima or a single layer of endothelium over a strut could also potentially be misclassified as uncovered by OFDI (3 of 5 struts) (Figure 3). The other reasons of misclassification were not determined.

In the current study, recognition of histological features included fibrin accumulation (n = 13 sections), excessive inflammation (hypersensitivity) (n = 4), organized thrombus (n = 8), foamy macrophages (n = 16), in-stent atherosclerosis including lipid/necrotic core (n = 11), intimal disruption (n = 4), angiogenesis (n = 6), and calcified neointima (n = 8). The descriptions of these features as recognized by OFDI are listed in (Table 4),(Figures 3, 4, 5), and (5). Of these, the presence of accumulated foamy macrophages on luminal surface or within the neointima exhibited a distinct appearance by a remarkably bright signal with a trailing shadow. OFDI appearances of calcification, angiogenesis, and intimal disruption within neointima were similar to those of native coronary arteries as described in the previous reports (15- 16).

In-stent atherosclerosis with lipid core/necrotic core appeared as dark areas with indiscriminant borders as previously reported in native arteries, and these were frequently co-localized with focal strong “bright” signals, suggesting the presence of foamy macrophages and/or cholesterol crystals (9 of 11 sections = 82%) (5). Fibrin accumulation, mostly identified around stent struts by histology, had a relatively dark appearance without clear margins in OFDI. Moreover, areas of fibrin were not accompanied by strong signals from macrophages or cholesterol crystals (0 of 13 sections = 0%) (Figure 4). When fibrin was accumulated on the surface, it was associated with the luminal irregularity in 7 of 8 sections (88%). Organized thrombus, diffusely or focally localized in neointima, also appeared dark by OFDI, and frequently was accompanied by black holes that indicated angiogenesis (6 of 8 sections = 75%) (Figure 4). Focal collections of lymphocytes or eosinophils (hypersensitivity reaction) also appeared dark; however, the absence of signal did not necessarily reflect the extent of the density of inflammatory cells (Figure 4). Additionally, OFDI signal analysis was performed in the superficial neointima inclusive of normal neointima >1 year (n = 26 sections), normal neointima ≤6 months (n = 16), hypersensitivity (n = 4), fibrin accumulation (n = 8), organized thrombus (n = 7), and foamy macrophages accumulated on the luminal surface (n = 9) (Table 4). The distance between the light source to the luminal surface of the regions of interest ranged within 0.43 to 2.09 mm. The highest values of “peak intensity” achieved were for superficial foamy macrophages (198.2 ± 5.2), followed by normal neointima >1 year (183.4 ± 6.8), normal neointima <6 months (173.2 ± 6.1), excessive inflammation (hypersensitivity) (165.1 ± 2.7), fibrin deposition (154.2 ± 5.1), and least in organized thrombi (134.0 ± 4.0). Further analysis showed that the “attenuation rate” was also prominent in areas of foamy macrophages (2.56 ± 0.32), followed by fibrin (1.46 ± 0.14), excessive inflammation (hypersensitivity) (1.05 ± 0.06), normal neointima <6 months (0.52 ± 0.11), and organized thrombi and normal neointima >1 year, which had very low indices of 0.40 ± 0.07 and 0.33 ± 0.09, respectively. As shown in the OFDI signal analysis plot in (Figure 5), macrophage foam cells and organized thrombi constituted distinct groups, independent of other histological features.

In the current study, ex vivo OFDI of human coronary stents provided a good correlation with histology relative to conventional IVUS, not only for estimation of stent/lumen area, but also for neointimal thickness, regardless of the complexity of the underlying atherosclerotic lesion, although even by OFDI, there was a discrepancy in the measurement of neointimal thickness <80 μm because of the so-called “blooming artifact” (10). Factoring in the efforts to detect uncovered struts in the recent clinical studies (17- 19), we further assessed individual strut coverage using co-registered OFDI and histology, which achieved a high sensitivity and an excellent specificity for the detection of uncovered struts. On the contrary, IVUS was unable to accurately resolved strut coverage, particularly for struts nearest the luminal surface. Notably, however, there were a few, but non-negligible misclassifications by OFDI, particularly for falsely positive covered struts where the appearance of fibrin over and around stent struts with or without inflammatory cells is a likely contributing factor. In a recent study in normal swine coronary arteries interrogated by OFDI at various time points up to 28 days, accumulated fibrin above struts proven to be distinguishable from normal neointima by a signal intensity ratio standardized by intensity from an adjacent strut (fibrin: median = 0.395 [IQR: 0.35 to 0.43] vs. neointima: 0.53 [0.47 to 0.57]; p < 0.001) (12). Similarly, our study showed a clear difference regarding OFDI signal peak intensity between fibrin (154.2 ± 5.1) and normal neointima (>1 year: 183.4 ± 6.8). Moreover, we found the signal attenuation rate was significantly greater for fibrin (1.46 ± 0.14) than for normal neointima (>1 year: 0.33 ± 0.09), which is helpful for identifying malapposed struts accompanied by extensive fibrin accumulation as often observed in the LST, especially with paclitaxel-eluting stents (20).

Stent-related hypersensitivity, although of rare occurrence, is also associated with a thrombogenic milieu that can lead to stent thrombosis (21). Virmani et al. (22) histologically characterized hypersensitivity as extensive inflammation predominantly consisting of eosinophils and T lymphocytes accompanied by vessel enlargement. Although there are no previous data relating hypersensitivity reactions accessed by OCT/OFDI with histology, a few clinical reports suspected of hypersensitivity are available along with serial angiography, in which the authors described that IVUS and OCT documented findings of vessel enlargement and late stent malapposition or multiple cavities over time (23- 24). Similarly, our case of hypersensitivity (61 months after Cypher implant [Cordis, Bridgewater, New Jersey]) identified luminal surface irregularity and cavity formation around struts, which were clearly delineated by OFDI. In addition, the recognition of relatively dark tissue with moderate signal attenuation rate in the areas adjacent to the stent struts, as demonstrated in the OFDI signal analysis of this study, may help confirm hypersensitivity and future risk of thrombosis though prospective data is obviously needed to validate this claim.

Using collagen-SMC gels studied in vitro, Levitz et al. (25- 26) recently showed that OCT signal reflection was potentially regulated by modifications of ECM through active matrix metalloproteinases, rather than cell or collagen density. Moreover, other studies involving animals, human atherectomy, and autopsy specimens of coronary stents also demonstrated changes in neointimal ECM over time (27- 29). In the case of human coronary stents, Farb et al. (29) elucidated the process of healing following bare-metal stent (BMS) implant, where the early neointima consisting of SMCs and ECM rich in hyaluronan, the proteoglycan versican, and type III collagen, is gradually replaced by the proteoglycan decorin and type I collagen over 18 months to 2 years, reminiscent of biological wound healing, although markedly delayed relative to normal healing of dermal injury. Similarly, the present study demonstrated signal differences over time in normal neointima ≤6 months and >1 year (peak intensity: 173.2 ± 6.1 vs. 183.4 ± 6.8, attenuation rate: 0.52 ± 0.11 vs. 0.33 ± 0.09, respectively). Thus, signal analysis of OCT/OFDI may also be useful to unveil the component or process of neointimal healing after stent implants.

Although the emphasis has been placed on delayed arterial healing in DES as a causative role in LST, the atherogenic propensity of the neointima may also be involved. A recent IVUS study enrolling 30 consecutive patients with very late stent thrombosis involving 23 DES and 7 BMS revealed different etiologies of stent thrombosis, where the cause in BMS patients was exclusively attributed to plaque rupture within the neointima as opposed to only 43.5% of DES, where delayed healing and in-stent atherosclerosis both play a role (6). In another study, Higo et al. (30) reported by angioscopy the yellow appearance of neointima in DES to be more commonly associated with thrombogenecity, further suggesting the involvement of atherosclerotic plaque.

Although several studies have examined the utility of OCT/OFDI for histological characterization of native human atherosclerotic arteries in an ex vivo setting (31- 32), to date, there are no published data regarding stents despite the relevance to clinical outcome. Recently, Takano et al. (33) assessed neointimal changes in patients with BMS implants for ≤6 months and greater than 5 years by OCT, where a lipid-laden neointima, intimal disruption, and thrombosis were more frequently in later phases, albeit without histological validation. In this regard, the present study illustrated, not only lipid, but also accompanying pathological features such as macrophages and cholesterol crystals, which may be helpful for detecting “neointimal instability” by comparing OFDI directly with histology.

Although OFDI has a good potential for identification of in-stent atherosclerosis, the dark appearance of fibrin accumulation, organized thrombus, excessive inflammation (hypersensitivity), and mixture with fibrous tissue create a heterogeneous or layered appearance and might impede direct discrimination of these tissues when they exist within neointima. In those cases, supervening findings such as angiogenesis or cavity formation may be helpful for the differentiation. Another obstacle for the accurate diagnosis is the accumulation of foamy macrophages on the luminal surface which prohibits the visualization of tissues behind. Further investigation will be needed to overcome those limitations of OFDI.

All stented lesions were interrogated in a blood-free environment, and the main analysis were performed utilizing images acquired following formalin fixation, which is markedly different from less ideal conditions in the catheterization lab. However, we found that there is no significant difference in OFDI signal reflection regardless of formalin fixation, albeit in limited cases. Although a multistep approach was used to achieve ideal co-registration of OFDI and IVUS images with histology, outside of longitudinal alignment and visual inspection, there were no other indicators or computational methods to pinpoint the precise location of histological section to the catheter-based signatures. Our methodology for signal analysis is applicable only for tissues facing the lumen, not for the tissues deep within the neointima. Finally, the study was performed in a limited number of samples, and therefore, further studies are required to confirm the significance and application of our findings in a larger cohort of cases.

Ex vivo OFDI of stented human coronary arteries showed superiority to IVUS for estimation of minimal neointimal thickness with a greater ability towards detecting covered and uncovered struts. Moreover, our study demonstrated the potential capability of OFDI for recognizing tissue pathology such as accumulated fibrin, excessive inflammation (hypersensitivity), organized thrombus, angiogenesis, calcified neointima, and in-stent atherosclerosis inclusive of foamy macrophages and/or cholesterol crystals. A comprehensive understanding of the histopathological changes that occur within stents together with their assessment by OFDI as described in this study may extend the utility of OFDI to gauge stent performance, but this will require clinical studies powered to examine hard endpoints based upon OFDI findings.

The authors acknowledge Isao Mori from Terumo Corporation for the adjustment of OFDI raw data.

For expanded Methods and Results/Discussion sections, please see the online version of this paper.

Ex Vivo Assessment of Vascular Response to Coronary Stents in Humans by High-Resolution Optical Frequency Domain Imaging