2.1.

Human Arterial Samples

This study was performed on 30 postmortem samples of coronary artery walls evaluated in parallel by CP OCT, histology, nonlinear microscopy, and AFM. The samples were obtained during autopsy in accordance with a scientific research protocol approved by the Research Ethics Board of the Nizhny Novgorod State Medical Academy, Russia. All arterial vessels were obtained no later than 24 h after death and delivered to the laboratory in gauze saturated with the phosphate buffer at 7°C.

CP OCT imaging was performed from the intimal side of the samples. After the OCT imaging, the OCT probe area (6 mm in the diameter) on the sample surface was marked with an ink circle and the sample was fixed (10% formalin for 48 h with the standard paraffin embedding process) and serially sliced in the central area, with the slice orientation matching the OCT imaging plane. Serial $10\text{-}\mu \mathrm{m}$-thick sections were used for nonlinear microscopy, $7\text{-}\mu \mathrm{m}$-thick sections were used for AFM and histological slides stained with hematoxylin and eosin (H&E). A total of 108 sets of CP OCT, histology, nonlinear microscopy, and AFM images were acquired and analyzed. Among these 108 sets, we found lesions of the following different degrees of atherosclerosis development: 19 thickened intima (stage I), 8 thickened intima with a lipid streak (stages II/III), 9 fibrous cap over a small lipid core (stage IV), 19 thin fibrotic cap over a big lipid core (stage Va, a most interesting one from the clinical viewpoint), 32 fibro-calcified plaque (stage Vb), and 21 very thick fibroatheromas (stage Vc).¹⁷

2.2.

Cross-Polarization Optical Coherence Tomography System

The spectral domain CP OCT system used in the study has been described in Refs. 11, 18, and 19. The light source was a superluminescent diode with a central wavelength of 1310 nm, radiation power of 20 mW, and a spectral width of 100 nm, resulting in the axial resolution of $15\mu \mathrm{m}$, scanning depth of 1.7 mm, scanning speed of 20,000 A-scans per second. The B-scans were formed from 512 consecutive A-scans with the frame rate of $40\text{frames}/\mathrm{s}$ and transverse scanning range of 2.7 mm for both images in the co- and cross-polarization channels. The spectrometer uses a prism-based linearization of the spectrum in the $k$-space.^18,20 Additional linearization was done with a signal processing as described in Ref. 19. As per common path interferometer configuration utilized here, there are no separate sample and reference arms. The reference radiation is created by a partial reflector in the tissue probe, and the primary Fizeau interferometer is created between this partial reflection and the actual tissue backscattering. The additional interferometer (upper part of Fig. 1) has its optical path difference approximately equal to the primary interferometer offset $L_0$. One arm of the additional interferometer contains an electrically controlled 45 deg Faraday rotator. For the copolarization channel, no driving voltage is applied, so polarization states in both arms are coaligned linear. Application of driving voltage rotates the polarization state by 90 deg and yields the orthogonal polarization states in the two arms, corresponding to the CP channel. The co- and cross-polarization channels are acquired sequentially in the left-to-right and right-to-left lateral scanning directions. The system had an active, closed loop control of the polarization at the fiber output, enabling maintenance of the circular polarization state emerging from the tissue probe; this is true even in the presence of fiber mechanical bends and resultant changes in the fiber birefringence.

Fig. 1

CP OCT schematic. LS, light source; BS, beam splitter; L, lens; and $L_0$, interferometer offset. Reprinted, with permission, from Gubarkova et al.¹¹

Note that the tissue optical anisotropy is related to its fiber orientation and the tissue eigen polarizations are linear. To observe the tissue polarization response, it is interrogated with the equal mix of the eigen linear polarizations. The circular polarization is best suited for this purpose, because it automatically provides this equal mix and does not require any alignments between the incident polarization and tissue anisotropy axes orientation. The system acquires the copolarization channel in odd-numbered (left-to-right) B-scans; then the Faraday rotator is activated to acquire CP channel in the even-numbered (right-to-left) B-scans.

2.3.

Quantitative Processing of the Cross-Polarization Optical Coherence Tomography Images

The quantitative processing of the CP OCT images involved calculation of the DF and effective birefringence ($\mathrm{\Delta }n$).

The calculation algorithm for DF was developed previously and applied for the evaluation of the collagen fibers’ state in a variety of tissues (bladder mucosa and aorta).^21,22 The DF is calculated as a ratio of OCT signals in the cross- and co-polarization channels according to the following equation:

Maps of effective birefringence ($\mathrm{\Delta }n$) were also built for CP OCT images as the in-depth derivative of the effective phase delay, as suggested in Ref. 6, similar to mapping the birefringence coefficient for PS OCT images and expressed according to the following equation:

This approach considers the phase delay and the resulting birefringence as effective values, because both processes—birefringence and CP backscatter—take part in the formation of the CP OCT signal and are hard to separate.

These values were also averaged over selected regions of interest (ROI) as the integral DF (IDF) and the average birefringence, respectively. As discussed earlier in Sec. 1, while it may be tempting to associate the IDF with the CP backscatter and effective $\mathrm{\Delta }n$ with the tissue birefringence, this should be attempted with great caution.

It should be noted that the terminology difference for the same spatial averaging (using the words “integral” and “average”) happens for historical reasons because the IDF was already used in the previous publications^21,22while the term “average birefringence” has been used in the literature for a long time.²³

2.4.

Average Isotropy Index for Nonlinear Microscopy Images

Visualization of collagen and elastic fibers in the vessels’ walls was conducted with an LSM 710 confocal laser-scanning microscope (Carl Zeiss, Germany). A pulsed Ti:sapphire laser (Chameleon Vision2, Coherent, Santa Clara) emitting 140 fs pulses with a central wavelength of 800 nm at a repetition rate of 80 MHz was used. The signals were detected simultaneously in two channels using filters in the wavelength ranges of 362 to 415 nm and 480 to 554 nm, when detecting the SHG signal from collagen fibers and the TPEF signal from elastin, respectively.²⁴ The images were obtained using a С-Apochromat water-immersion objective with a $40\times$ magnification and a numerical aperture of 1.2, providing a $212\times 212\mu \mathrm{m}$ field of view ($1024\times 1024\text{pixels}$).

The quantitative processing of the nonlinear microscopy images was performed based on Fourier analysis.^13,24 The two-dimensional Fourier spectrum of the images was binarized with the threshold level of 30% of the maximum pixel intensity. At the next step, the ratio of the smaller to the larger ellipse axes was calculated. This ratio is called the average isotropy index and could serve as a measure of the tissue’s anisotropy. Structures with the fiber orientation along a certain prevailing direction tend to have a smaller value of the isotropy index. Disorganization of fibers with a more isotropic distribution results in the isotropy index increasing toward 1.

2.5.

Flicker-Noise Spectroscopy Parameterization of Atomic Force Microscopy Images

The tissue sections on glass slides were imaged on air in the semicontact mode, using a Solver P47 AFM (NT-MDT, Russia), a Solver scanner ($14\times 14\times 2\mu \mathrm{m}$), and tapping etched silicon probes (Bruker, Billerica, Massachusetts) with a nominal spring constant of $42\mathrm{N}/\mathrm{m}$, nominal resonant frequency of 320 kHz, and a nominal tip radius of 8 nm. The acquisition of $6\times 6\mu \mathrm{m}$ AFM images was performed at a scan rate of 1 Hz, with a $512\times 512\text{pixels}$ resolution. The ROI for AFM was selected in accordance with the histological findings, with the use of an optical microscope combined with the AFM instrument.

The FNS technique was applied to quantitatively parameterize the morphology of the arterial wall’s extracellular matrix and its changes occurring with the progression of atherosclerosis. FNS is a general phenomenological approach for analyzing series of chaotic signals, via the search of correlations in a sequence of dynamic variables and ascribing an informational significance to such correlations. FNS, as a technique of statistical physics, introduces a number of nanometrology parameters, which may quantitatively characterize the surface texture.¹⁶ Application of FNS instead of standard metrological parameters (ISO 4287) is highly preferable in case of biological samples characterized by a considerable height heterogeneity. While the standard metrology parameters are nonspecific and may reflect the entire history of a sample preparation, which affects its surface (pretreatment, drying, and microtome sectioning), FNS parameterization generally extracts only correlations inherent for the sample’s structure, essentially filtering out the systematic artifacts brought in by the sample preparation.

The details of the FNS algorithm are described in Ref. 16. Briefly, when applying the FNS parameterization to the three-dimensional surface images obtained by AFM, the roughness profiles $h(x,y)$ are considered as the dynamic variables, and the extracted surface parameters are related to the autocorrelation function for every profile. Different ranges of spatial frequencies are considered in the $h(x,y)$ profiles: low-frequency (“resonant”) and high-frequency ranges, the latter represented by two basic types of irregularities against the low-frequency baseline—“jumps” (relatively moderate-height steps) and “spikes” (very narrow peaks of a relatively significant height).

The contribution of the low-frequency baseline (originating from the sample positioning and other manipulations) is excluded, so only high-frequency random steps and spikes remain for consideration. In general, several FNS parameters are introduced, but, as was shown earlier,¹⁵ the changes in the biological samples’ morphology are best characterized by the two basic FNS parameters reflecting contributions of the above-mentioned types of high-frequency irregularities: a “stepwiseness factor” $\sigma$ (nm) for the jump-like irregularities of the studied AFM-profile and a “spikiness factor” $S(L_0^{-1})$ (${\mathrm{nm}}^2\mu \mathrm{m}$) for the spike-like irregularities of the studied profile, where $L_0$ is the length of correlation for the high-frequency irregularities at the nanoscale (see in Ref. 16 for the corresponding calculation procedure).

A custom-made software was used for the FNS computations. The “stepwiseness factor” $\sigma$ and the “spikiness factor” $S(L_0^{-1})$ were used as the surface parameters for the comparison of the surface morphology at different atherosclerosis stages. It should be noted that, while FNS power spectra do demonstrate the intrinsic periodicity of collagenous structures (D-period), in general, FNS parameters have no direct reference to the measurable characteristics of extracellular matrix, such as the collagen content or fibrils’ dimensions. However, correlations in the chaotic surface profiles (revealed by the FNS analysis) appear rather sensitive to the structure ordering/disordering due to various processes,^15,16 and thus, the FNS parameters derived from such correlations may be used as markers for a comparative analysis of the extracellular matrix structure.

2.6.

Region of Interest Selection

The ROIs for the quantitative evaluation of CP OCT, SHG, and AFM images were selected manually. These ROIs had a specific size for each imaging modality: $2\times 0.3\mathrm{mm}$ for CP OCT, $0.2\times 0.1\mathrm{mm}$ for SHG, and $6\times 6\mu \mathrm{m}$ for AFM. The ROIs of CP OCT, SHG, and AFM were specifically chosen within the upper part of an atherosclerotic plaque, namely, in the fibrous cap or intima of different stages of atherosclerosis development: thickened intima (stage I), thickened intima with the deposition of lipids (stages II/III), thickened fibrous cap (stage IV), a thin fibrous cap (stage Va), a fibro-calcified cap (stage Vb), and a very thick fibrous cap (stage Vc).

The manual ROI selection was used with the understanding that if a significant difference between these areas was observed, the next research stage would be moving toward more automated processing to develop an automatic procedure of plaque differentiation in the future.

2.7.

Statistical Processing

The data distribution normality was verified using the Kolmogorov–Smirnov criterion. The Kolmogorov–Smirnov test quantifies the distance between the empirical cumulative distribution function of the sample and the cumulative distribution function of the reference distribution; in this case, the reference is the Gaussian normal distribution. The null hypothesis for the Kolmogorov–Smirnov test is that the distribution is different from normal, and it is rejected if the $p$ value (probability that the difference is just a random chance) is higher than 0.05 [in our case, it is found to be 0.2 for the IDF and average $\mathrm{\Delta }n$ values for two plaque stages (IV and Va)]. Therefore, parametric statistical tests were used. All the results were expressed as $\text{mean}\pm \mathrm{SD}$. The statistical significance of differences in effective $\mathrm{\Delta }n$, IDF, and isotropy indices between the stages IV and Va was calculated by the student’s $t$-test. The multiple comparisons with the Bonferroni correction were used, in particular to compare stage I with all the other stages. In all cases, the differences were considered statistically significant when $p<0.05$. The Spearman’s correlation coefficient ($r$) was calculated to determine the correlation among the isotropy index, effective birefringence, and DF.

The statistical data processing was done in MS Excel 2010 with a public domain software plugin for statistical analysis STATISTICA 10 (StatSoft, Inc., Tulsa, Oklahoma) and the Prism v6 software (GraphPad Software, La Jolla, California).

3.1.

Visual Assessment of Cross-Polarization Optical Coherence Tomography, Histology, Second-Harmonic Generation, and Atomic Force Microscopy Images

CP OCT, histology, SHG, and AFM were used previously to separately assess the content, spatial, and structural organization of collagen fibers in atherosclerotic plaques.^11,15 In this study, these methods are combined for a parallel comparative evaluation of the different stages of plaque development.

The comparison of CP OCT images of atherosclerotic plaques at different stages of atherosclerosis (I, II/III, IV, Va, Vb, and Vc stages) reveals a tendency for violation of the layered structure in the vessel wall [Figs. 2(a)–2(f)]. Second, the changes in anisotropic structures, such as collagen fibers, influence the CP channel brightness [the lower part of the image in Figs. 2(a)–2(f)]. It should be noted that the changes in packaging density, microstructural organization, and the geometric orientation of collagen fibers in the intima/fibrous cap are sensitive markers for the progression of atherosclerosis [Figs. 2(g)–2(l), yellow rectangle] and identification of vulnerable (prone to rupture) plaques [Fig. 2(j)]. For a detailed analysis of the fibrous patterns of the intima or fibrous caps, SHG and AFM images were obtained to compare them with the corresponding CP OCT data.

Fig. 2

Visualisation of different stages of a plaque development by CP OCT, histology, SHG, and AFM. (a–f) CP OCT images (upper panel = copolarization channel, lower panel = CP channel), (g–l) histological slides stained with H&E, (m–r) SHG images, and (s–x) AFM images. The rectangles in the CP OCT images indicate the area covered by histological slides and SHG images, respectively. The arrow points to accumulations of foam and inflammatory cells at the border of the fibrous cap and the lipid core. The square in the SHG images indicates the area studied by AFM.

CP OCT image in the CP channel at the stage I of a plaque demonstrated conservation of the vessel wall layered structure visible as thin horizontal layers with a higher signal level in the thickened intima [Fig. 2(a), rectangle]. The intima is represented predominantly by collagen fibers [Fig. 2(g)], which are well oriented along the vessel wall, demonstrate an intensive SHG signal [Fig. 2(m)], and have a typical banded structure of collagen fibrils, forming a basket-weave pattern at the AFM image [Fig. 2(s)].

With the development of atherosclerosis, the collagen fibrils’ packing is changed and they are covered with an amorphous material (protein–proteoglycan complexes) as can be seen at the AFM images [Figs. 2(t)–2(x)]. The corresponding SHG images demonstrate a gradual change of the packing density and organization of collagen fibers [Figs. 2(n)–2(r)]. It is especially important to note the strong disorganization of collagen fibers in the fibrous caps of vulnerable plaques (stage Va) as seen from histology [Fig. 2(j)], SHG [Fig. 2(p)], and AFM [Fig. 2(v)] images.

The corresponding CP OCT images can be analyzed for the detected changes in organization/disorganization of collagen fibers by the OCT signal level in the CP channel [Figs. 2(a)–2(f)]. The deposition of lipids on the surface or inside the bulk of the intima as the lipid strips (stages II/III) produces almost no CP signal at the top of the image, but then the CP brightness increases, possibly because of the fiber birefringence coexisting with the lipid [Fig. 2(b), rectangle]. In the case of stable plaques (stages IV, Vb, and Vc), a high CP brightness is observed near the surface from a fibrous cap over the lipid core [Figs. 2(c), 2(e), 2(f), yellow rectangle, respectively]. Specific for a vulnerable plaque (stage Va) is a highly heterogeneous OCT signal, quickly decreasing with the depth in the copolarization channel [the upper part of the image in Fig. 2(d)]. In the CP channel, this area shows a weak heterogeneous OCT signal. This finding indicates the prevalence of disrupted fibers in the fibrous cap and can serve as an indicator of plaque instability [Fig. 2(d), rectangle]. The cell clusters located between the lower part of the fibrous cap and the surface of the lipid core are visible in the CP channel as “bright spots” [arrow, lower panel, Fig. 2(d)].

Thus, one may note that the CP channel in CP OCT plaque images provides an additional contrast for the assessment of the collagen fibers’ condition.

To provide a more objective assessment of the main structural components of plaques at different development stages, a quantitative analysis of the CP OCT, SHG, and AFM images was proposed (Sec. 3.2).

3.2.

Quantitative Assessment of the Cross-Polarization Optical Coherence Tomography, Second-Harmonic Generation, and Atomic Force Microscopy Images

Quantitative characteristics derived from the different imaging modalities data for different plaque development stages are shown in Fig. 3. The IDF calculation [Fig. 3(a)] revealed a statistically significant difference between a plaque at the initial stage of development (stage I), a vulnerable plaque (stage Va), and a fibroatheroma with severe fibrosis (stage Vc). The IDF allows reliable differentiation between stable (stage IV) and vulnerable (stage Va) plaques ($0.46\pm 0.21$ against $0.09\pm 0.04$; $p<0.001$) origination from significant difference in spatial organization collagen, although the structure manifested by thin fibrous cap and a lipid core is similar. Both types of plaques have thin fibrous caps and a lipid core, but a different spatial organization of collagen in the fibrous cap above the lipid core, thus, they are hard to be distinguished by traditional diagnostic techniques. The IDF decrease at Va stage is consistent with the histologically known significant disorganization of the collagen fibers’ orientation.

Fig. 3

(a) IDF, (b) average $\mathrm{\Delta }n$, and (c) average isotropy index for different stages of plaque development. Asterisk (*) indicates a statistically significant difference between stages I and all the subsequent plaque development stages, $р<0.05$ (Bonferroni posthoc test), double asterisk (**) indicates a statistically significant differences between stable (IV) and vulnerable (Va) plaque stages, $p<0.001$ (student’s $t$-test).

The results of the average effective $\mathrm{\Delta }n$ calculation from CP OCT images of plaques at different stages are shown in Fig. 3(b). The average $\mathrm{\Delta }n$ value is significantly higher in the slightly thickened intima at the initial stage of a plaque (stage I) as compared to the following stages of plaque development ($p<0.05$). At the same time, the high values of the average $\mathrm{\Delta }n$ for stages I, IV, Vb, and Vc may indicate the presence of a large number of ordered anisotropic structures (collagen fibers) distributed homogeneously in the fibrous cap of a stable plaque. It was found that the average $\mathrm{\Delta }n$ of the fibrous cap of a stable plaque (stage IV) amounting $(0.47\pm 0.10)·{10}^{-3}$ is higher than that for the disorganized fibers of a vulnerable plaque (stage Va) reaching ${[(0.25\pm 0.07)·10}^{-3}$; $p<0.001$]. However, the Va plaque cap is too thin to have a measurable (by CP OCT) birefringence, so this is considered as an evidence-based observation rather than a statement about the CP OCT measurement of the real $\mathrm{\Delta }n$.

The low values of the IDF and average $\mathrm{\Delta }n$ at stage Va may indicate the predominance of disorganized collagen fibers in the fibrous cap. Another explanation for such a decrease of the CP channel brightness is that ROI included a lipid strip on the surface of the intima, and lipid (for example at the stages II/III) is not capable of providing CP backscattering.

The calculation of the average isotropy index from SHG images was used as an alternative method to analyze the collagen fibers order in the intima/fibrous cap. The calculated values of the average isotropy index are shown at the histogram in Fig. 3(c). As expected, the average isotropy index has lower values for stable (stages I, II/III, IV, and Vb) plaques than that for vulnerable (stage Va) plaques, demonstrating a higher alignment of collagen fibers in a stable plaque in contrast to a more isotropic distribution within a vulnerable plaque [Fig. 3(c)]. The low values of the isotropy index for stages II/III indicate a presence of organized collagen fibers and it appears difficult to determine a correlation with CP OCT showing the low value of the average $\mathrm{\Delta }n$ and IDF. This occurs due to the fact that the SHG signal in tissue is observed exclusively in densely packed, organized structures, such as collagen, which still retains organization in between clusters of lipids at this stage of plaque development.

A statistically significant difference ($p<0.05$) between the average isotropic indices of disorganized fibers of fibrous caps in a vulnerable plaque (stage Va) and of all the other stages of a plaque (stages I, II/III, IV, Vb, and Vc) was found. In addition, a statistically significant difference exists between the average value of isotropy index of stable (IV) and vulnerable (Va) plaque stages ($0.30\pm 0.10$ against $0.70\pm 0.08$; $p<0.001$), which supports the interpretation of the OCT signal decrease in CP OCT images of a vulnerable atherosclerotic plaque [Fig. 3(a), 3(b)].

The Spearman’s correlation coefficient shows a high negative correlation between the average isotropy index derived from SHG images and the average $\mathrm{\Delta }n$, calculated from CP OCT images: for a stable (stage IV) plaque $r=-0.81$ and for a vulnerable (stage Va) plaque $r=-0.94$ ($p<0.001$). The correlation coefficients between the average isotropy index and IDF are as follows: for a stable plaque $r=-0.83$ and for a vulnerable plaque $r=-0.89$ ($p<0.001$). This correlation is shown for the first time and demonstrates the high degree of concordance between the tissue’s ability to exhibit birefringence/CP backscattering. Consequently, basing on the CP OCT images, a structural organization of collagen fibers in tissues can be assessed. Moreover, by quantitative characterization of organization of collagen fibers through the IDF and average $\mathrm{\Delta }n$ analysis, higher values were revealed for stable plaques as compared to vulnerable plaques.

The total averaged values of the peak-to-valley roughness and FNS parameters in the AFM images are represented in Fig. 4. The gradual reduction of the peak-to-valley roughness $h_m$ with the severity of atherosclerosis stages indicates that the surface becomes smoother [Fig. 4(a)]. The high values of FNS [$\sigma$ and $S(L_0^{-1})$] parameters in the area of the norm (stage I) are likely associated with a wavy structure of collagen fibers in the intima, which is manifested as “holes” at the cross sections of a sample, making the tissue surface looser. For all the following pathological stages of plaque development (stages II/III, IV, Va, Vb, and Vc), both FNS parameters have lower values, but at the later plaque stages, the parameters’ values increase, which is associated with disordering of collagen fibrils [Figs. 4(b) and 4(c)]. The growth of the $S(L_0^{-1})$ peaks with the transition from the early stage to the later stages of atherosclerosis may be caused by the increased proportion of uncoated collagen fibrils and by gaps in the collagen network, resulting in more explicit differences in the heights of profiles [Fig. 4(c)]. At the stage Vc of atherosclerosis development, the FNS parameters again decrease, and this can be explained by the formation of a smoother surface due to the thick coverage of collagen fibrils and fibers with unstructured proteins. At the same time, the characteristic frequencies in the power spectra of the last stages are either weakened (stage Vc) or almost disappear (stage Va).¹⁵ It should be noted that there is a rather noticeable error in the measured values of the $S(L_0^{-1})$ parameter due to the high degree of surface heterogeneity of the studied samples.

Fig. 4

(a) Histograms of peak-to-valley roughness $h_m$ and FNS parameters of AFM images of different stages of plaque development, (b)“stepwiseness factor” $\sigma$, and (c) “spikiness factor” $S(L_0^{-1})$.

Thus, the FNS analysis of the surface topography, which indicates the disordering of collagen fibers with the development of atherosclerosis, also supports the reduction in the polarization properties at CP OCT images obtained for the fibrous cap of a vulnerable atherosclerotic plaque.