1.1.

Background

Cardiovascular disease, especially coronary artery disease (CAD), is the leading cause of death in the world.¹ Atherosclerotic plaque that develops on the basis of lipid accumulation within the vascular wall is the most important pathophysiology in CAD. Progression of atherosclerotic plaques narrows the coronary arteries, hindering blood flow to the myocardium and leading to increased risk of myocardial infarction and heart failure.² Furthermore, rupture of vulnerable plaque and subsequent thrombogenesis is the main cause of acute myocardial infarction.³ Previous studies have reported that the vulnerability of atherosclerotic plaque is highly associated with the thickness of the fibrous cap, size of the lipid pool, and extent of macrophage infiltration.^4–6 In this context, accurate characterization of plaque composition is expected to hold great promise in detecting coronary vulnerable plaques.

Optical coherence tomography (OCT)⁷ is an interferometric imaging technique that is widely used in many medical fields including ophthalmology,^8,9 dermatology,^10,11 and cardiology.^12–14 Due to the unprecedented high spatial resolution of OCT (axial resolution of 10 to $20\mu \mathrm{m}$ and lateral resolution of 10 to $30\mu \mathrm{m}$) and rapid image acquisition speed ($>100\text{frame}/\sec$), this imaging technique can be used to analyze biological tissues by providing three-dimensional (3-D) imaging in vivo.¹⁵ Especially, intravascular OCT (IV-OCT), a catheter-based OCT imaging¹⁶ in cardiology, has been widely applied to visualize the microscopic structures of arterial inner walls and implanted stents in the coronary artery during percutaneous coronary intervention.^12–14

Atherosclerotic plaques are known to have three major components: lipids with cholesterol, calcium deposit, and fibrous tissue.¹⁷ IV-OCT can be used to visualize atherosclerotic plaques with high resolution, providing qualitative and quantitative information for these tissue components.¹⁸ The comparison study of IV-OCT images with corresponding histological sections by Yabushita et al.¹⁹ demonstrated the association of the tissue composition with OCT imaging such as lipid-rich plaques showing diffuse borders with signal-poor regions, fibrocalcific plaques showing signal-poor regions with sharply delineated borders along the lateral direction, and fibrous plaques showing homogeneous and signal-rich regions. Additionally, Xu et al.²⁰ quantified the attenuation and backscattering coefficients of these plaque components from an ex vivo histopathologic correlation study, demonstrating that lipid-rich plaques have higher attenuation and backscattering coefficients than fibrocalcific and fibrous plaques.

Several previous studies have characterized atherosclerotic plaques based on qualitative and quantitative tissue properties. Specifically, Van Soest et al.²¹ described a method to automatically classify atherosclerotic plaque constituents, namely, lipids, calcium, and fibrous tissues, based on the optical attenuation coefficient of the tissue measured in gray-scale OCT images. This method for measuring the optical attenuation coefficient has been successfully demonstrated in vitro. Furthermore, Ughi et al.²² developed an automated algorithm for characterization of IV-OCT atherosclerotic tissues using a supervised classifier, called random forests,²³ trained by the optical attenuation coefficient and textural features. However, characterization of plaques, especially lipid-rich plaques, using gray-scale OCT images can lead to misclassification with low accuracy and high ambiguity due to various IV-OCT intrinsic artifacts such as shadowing,¹⁸ tangential signal dropout,²⁴ and negative contrast generated by the high attenuation properties of lipids.¹⁸

To complement the limitations of tissue characterization based on gray-scale IV-OCT images, various spectroscopic imaging systems have been introduced to provide biochemical or compositional information about atherosclerotic plaques, including near–infrared spectroscopy^25,26 and spectroscopic intravascular photoacoustic imaging.^27,28 These techniques enable one to distinguish the composition of atherosclerotic plaque by analyzing distinct patterns in spectroscopic signal. The additional biochemical or compositional information has offered an opportunity for better diagnosis and characterization of atherosclerotic lesions.

1.2.

Spectroscopic Optical Coherence Tomography (S-OCT)

Spectroscopic OCT (S-OCT) is an extension of conventional OCT that allows depth-dependent spectroscopic analysis of samples.^29,30 S-OCT is accomplished using short-time Fourier transform (STFT), a method for time-frequency analysis of interferometric signals. S-OCT can be compared with conventional Fourier-domain OCT (FD-OCT), in which the image is acquired by Fourier transform of the signal (Fig. 1). Although axial resolution is sacrificed as a result of the uncertainty principle for STFT, this postprocessing technique provides backscattering spectra of all pixels in the OCT image. Importantly, S-OCT uses the same signal as conventional FD-OCT and thus does not require any additional devices to access the spectroscopic information, which makes this technique more attractive.

Spectroscopic analysis using S-OCT can be divided into global and local approaches.³¹ The global approach of S-OCT focuses on identifying global changes in a sample, i.e., substance identification, by extracting wavelength-dependent optical absorption and scattering coefficients from the spectroscopic data.^32,33 On the other hand, the local approach can be used to measure the sizes of scatters that are less than the coherence length of the source by investigating S-OCT spectra modulated by interference of light within a scatterer or from adjacent scatterers.^34–36

S-OCT has been widely used for functional medical imaging due to its ability to provide additional spectroscopic analysis of a sample with OCT images. Applications of S-OCT include precancerous lesion detection,^37,38 measurement of blood oxygen saturation level,^39–41 and burn analysis.^42,43 In addition, Fleming et al.⁴⁴ developed an automatic lipid detection algorithm with IV-OCT images using S-OCT by extracting the wavelength-dependent optical attenuation coefficient. However, this study had limitations in that the results were derived from ex vivo mimicking phantoms with a bench-top OCT system, which does not correspond to the practical setting.

In continued research on substance identification using S-OCT, quantitative measurement of wavelength-dependent attenuation properties has been used to characterize tissue types.^42–44 However, such quantitative measurements along a full axial range are limited to applications with highly scattering biological tissues.⁴⁵ To overcome this limitation, Jaedicke et al.⁴⁵ proposed metrics to analyze and visualize the spectroscopic information extracted by S-OCT. Specifically, the S-OCT metric analysis can be used to reduce the entire multidimensional S-OCT spectrum to between one- and 3-D metrics for effective visualization of spectroscopic properties of a sample. Thus, this method is suitable for characterization of highly scattering media such as biological tissues.

1.3.

Objective

The objective of this study was to develop a method for characterization and identification of lipid-rich plaques based on S-OCT metrics analysis. Using a catheter-based IV-OCT system, we validated the proposed method using a lipid phantom model, and then evaluated the method in vivo in an atherosclerotic rabbit model. The proposed method will be useful in alleviating the problems of ambiguity during characterization of gray-scale OCT images and is expected to enhance the accuracy of characterization.

2.1.

Intravascular Optical Coherence Tomography Imaging System

The IV-OCT imaging was performed with a custom-built OCT system and an imaging catheter recently developed by our group.⁴⁶ The OCT system was based on a wavelength-swept laser with a center wavelength of 1290 nm, a bandwidth of 116 nm, and a sweeping rate of 120 kHz. The OCT system was operated with a custom-made catheter with an outer diameter and length of 0.87 mm and 1.5 m, respectively. The maximum frame rate of the system was $100\text{frames}/\mathrm{s}$, and the pullback speed of the catheter was $20\mathrm{mm}/\mathrm{s}$.

2.2.

Lipid Phantom Model

First, lipid phantoms were created to demonstrate how the proposed method works ideally and theoretically. To mimic the lipid composition of atherosclerotic plaque, we chose mayonnaise (Heinz, Pittsburgh, Pennsylvania) typically composed of soybean oil ($\sim 80\%$), whose composition is similar to that of the lipid-rich plaque, where cholesteryl linoleate and oleate take a large proportion of the cholesterol ester group in the lipid ($\sim 75\%$).^47,48 We prepared three phantoms with different concentrations of mayonnaise and a phantom without mayonnaise. Specifically, mayonnaise was diluted with Tris-buffered saline with 0.05% of Tween 20 (TBST) solution to final concentrations of 28.6%, 16.7%, and 9.1%. A Phantom without mayonnaise was made with ketchup (Heinz, Pittsburgh, Pennsylvania) diluted with TBST solution to a final concentration of 28.6%, due to its scattering property and the absence of cholesterol. Each phantom solution was filled in the space between an outer plastic tube (ID 10.7 mm, OD 11 mm) and an inner fluorinated ethylene propylene tube (ID 4 mm, OD 5 mm, Fluorotherm Polymers, Parsippany-Troy Hills, New Jersey) to mimic the vessel luminal structure. Then, each phantom tube was imaged using the system described earlier.

2.3.

Atherosclerotic Rabbit Model

We developed a rabbit model of inflamed lipid-rich plaque (lipid-rich rabbit) and inflamed fibrous plaque (fibrous rabbit). Inflamed lipid-rich atherosclerotic plaques in a New Zealand white rabbit (3.0 to 3.5 kg, male; Dooyeol Biotech, Seoul, Korea) were induced with balloon denudation on a left iliac artery (LIA) and hypercholesterolemic diet (HCD). For the inflammatory fibrous rabbit model, neointimal thickening was induced by balloon denudation as described above, however the animal was placed on a normal chow diet instead. Detailed protocol has been described in a previous paper.⁴⁹ The fibrous rabbit model and an intact right iliac artery (RIA) of a lipid-rich atherosclerotic model served as control groups.

The animal experiments were approved by the Institutional Animal Care and Use Committee of Korea University (KUIACUC-2013–88) and were performed in accordance with the recommendations of the national and institutional guidelines. Under fluoroscopic guidance, the imaging catheter was inserted to the target iliac artery, and IV-OCT image sets were acquired while flushing the artery with contrast media. The arterial segments identified by anatomical landmarks from distal to proximal iliac arteries were imaged with a rotational speed of 100 rps and pullback speed of $20\mathrm{mm}/\sec$. Raw OCT signals containing 1024 A-lines per frame were used for further characterization processing. All processing for data sets of both the phantom and the rabbit was implemented in MATLAB R2015a (The MathWorks, Natick, Massachusetts).

2.4.

Spectroscopic Optical Coherence Tomography Metric Analysis for Lipids

In the bandwidth of our OCT system, fibrous or normal tissue exhibited a rather flat absorbance through the system band (purple and blue spectra in Fig. 2). On the other hand, the lipid with cholesterol had higher absorption properties at lower wavelength, i.e., a higher wavenumber ($k=2\pi /\lambda$) (yellow spectrum in Fig. 2).⁴⁴ Therefore, in regions of lipid-rich plaque, the intensity at lower $k$ was dominant compared with the higher band. Importantly, this spectroscopic difference between substances can be used to distinguish lipid-rich plaque from other tissues.^44,50

Fig. 1

Concept of S-OCT and comparison with conventional FD-OCT. In the FD-OCT system, an OCT intensity image and a set of depth profiles (A-lines) are acquired by Fourier transform of interferometric signals and used to determine the morphology of the sample. However, if STFT is applied to the interferometric signal instead of Fourier transform, additional spectroscopic information, namely the depth of each point, can be determined. This postprocessing technique is called S-OCT and enables depth-resolved spectroscopic analysis of a sample.

Fig. 2

Near–infrared absorbance spectra of cholesterol, collagen, and calcium.⁴⁴

In previous S-OCT researches,^30,45 the center of mass (COM) of the S-OCT spectrum has been used as a representative metric to visualize the property of the spectrum. However, the COM metric is likely to be distorted and measured with a large deviation since an S-OCT spectrum contains an oscillation pattern due to speckle noise. In other words, the calculation of the COM metric is affected by the phase and the amplitude of the oscillations, which are not related to the substance identity. Filtering the oscillation pattern of the spectrum is a possible solution to this problem,⁴⁴ but it is difficult to identify an appropriate cut-off frequency for such filtering. Thus, the mean parameter ($b$) of a Gaussian fitting of the S-OCT spectrum, which we named the Gaussian center of mass (GCOM) metric, was introduced as a robust metric to avoid this problem.

Figure 3 shows how the GCOM metric is determined using real S-OCT spectra of each axial point. To represent the lipid distribution in arterial tissues of IV-OCT images, GCOM metrics were computed from the spectra of all pixels within a tissue region of interest (ROI), defined as the region enclosed by the vessel lumen contour (yellow curve in Fig. 3) and the perivascular contour (ocean blue curve in Fig 3). The vessel lumen and perivascular contours were automatically segmented on gray-scale OCT images using an algorithm described in a previous study.⁵¹ Considering the absorption properties of lipids, we anticipated that the GCOM of an S-OCT spectrum of a pixel with lipid-rich plaque would be located at the lower band of $k$ compared with other tissues as described in Fig. 3. Thus, if lipids are distributed along the axial direction (the same direction as the illumination of the light), the GCOM metrics along the direction will experience a negative shift in the wavenumber domain (black dashed arrows in Fig. 3).

Fig. 3

Representative example of the extraction of the GCOM metric function along the axial direction, $b(z)$, in an IV-OCT image of the rabbit artery with a lipid-rich plaque. From spectra at axial points indicated by green arrows, GCOM metrics are extracted by calculating the mean of Gaussian fitting (blue curve) of S-OCT spectra (red curve).

As a result of attenuation being the cumulative effect along the axial direction, the intensity of an S-OCT spectrum at a specific axial point ($z$), assuming constant scattering in a medium, is generally represented by the integrated form of Beer’s law as follows:

Depth-resolved spectra were acquired by STFT of a calibrated interferometric signal with a window size of $58{\mathrm{cm}}^{-1}$, resulting in a spectral resolution of 9.6 nm and a degraded axial resolution of $76\mu \mathrm{m}$. In order to reduce noise in S-OCT,⁴⁵ 20 depth-resolved spectra adjacent to both sides of the pixel (a total of 41) were averaged, and the averaged depth-resolved spectra were again smoothed by a two-dimensional kernel with a pixel size of 3 by 7 ($13\mu \mathrm{m}$ by 67 nm) in the axial and spectral directions, respectively. Finally, we extracted the GCOM metrics, $b(z)$, from the processed S-OCT spectra of all pixels in the tissue ROI and applied differential operations to calculate the lipid distribution function, $d_{\text{lipid}}(z)$.

In this method, characterization images are represented in the hue, saturation, and intensity space to simultaneously describe the compositional information as color (hue) and morphological features of the artery as brightness (intensity). Therefore, $d_{\text{lipid}}(z)$ can be directly mapped to the hue space of the corresponding A-line of the image in order to visualize the lipid-rich plaque. After $d_{\text{lipid}}(z)$ mapping to the hue space, we transformed the images of the results to Cartesian coordinates for visualization.

2.5.

Histological Evaluation

Experimental rabbits were euthanized with intravenous KCl ($2\mathrm{mmol}/\mathrm{kg}$) under anesthesia, and both iliac arteries were dissected. A bright light image of resected vessels was captured using a digital camera to document the branch points of the arteries, which were used as reference points for OCT/histology matching. Then, 3-mm vascular rings were frozen in optimum cutting temperature compound (Sakura Finetek, Torrance, California) for histopathological analysis. A total of nine rings of the tissue from LIA with balloon denudation and one ring from RIA with no injury were generated. Serial cryosections ($10\mu \mathrm{m}$) were cut at $250\text{-}\mu \mathrm{m}$ intervals and stained with Oil Red O (ORO). ORO was used to histologically visualize fat cells on frozen sections.

2.6.

Optical Coherence Tomography/Histology Co-Localization

Arterial branches and arterial morphology were used for colocalization between OCT frames and histological sections. For colocalization, the locations of arterial branches in the OCT pullback were confirmed in the angiographic image, and the branches were longitudinally matched to the arterial tissue. Then, the iliac arterial tissue was cut at a thickness of $10\mu \mathrm{m}$ on a cryostat, and the cryosectioned tissue slides containing a branch were paired with the corresponding OCT frames. In cases without a branch, the distance from the nearest branching point, plaque shape, and neovessel formation were used for colocalization.

2.7.

Statistical Analysis

Statistical analysis was carried out to validate the methods for characterization of lipid-rich plaque based on ROI analysis between lipid-mapping results, mapped images by the values of $d_{\text{lipid}}$ of all pixels, and the corresponding histological sections with ORO stain, which are considered the gold standard for lipid characterization. In OCT/histology-matched sections prepared as described in Sec. 2.6, corresponding paired ROIs were determined on both the gray-scale OCT images and the gray-scale histological sections to consider the morphological features of images. A total of five fixed-size paired ROIs per each matched section were defined. In sections with a bifurcation, we determined a total of four paired ROIs on the matched sections. Next, the paired ROIs determined on the gray-scaled images were drawn on both the lipid-mapping results and the corresponding ORO-stained sections. The ROIs of the lipid-mapping and the corresponding ORO were directly compared for statistical analysis. A receiver operating characteristic (ROC) curve was constructed from sensitivity and specificity data determined by comparison of lipid-mapping ROIs with corresponding ORO ROIs. In order to evaluate sensitivity and specificity, a lipid-mapping ROI with $>50\%$ of pixels in which the signal was larger than the $d_{\text{lipid}}$ threshold value was considered lipid-positive. Similarly, an ORO ROI in which $>50\%$ of pixels stained red was considered lipid-positive. The best $d_{\text{lipid}}$ threshold value was determined from the constructed ROC curve. In addition, linear regression was performed between the percentage of lipid-positive pixels in the lipid-mapping ROIs and the percentage of pixels stained red in the corresponding ORO ROIs. Finally, the peak values of the lipid-mapping ROIs with lipid-rich plaque, fibrous plaque, and no lesion were statistically compared based on $\text{mean}\pm \mathrm{SD}$. Unpaired one-tailed Student’s $t$-test was employed to determine the statistical difference among the peak values of each group.

3.1.

Phantom Models of Lipid-Rich Plaque

Figure 4 shows the representative GCOM and $d_{\text{lipid}}$ of lipid phantom models as a function of depth. As shown in Fig. 4(a), the relative GCOM values, defined as $b(z)-b(0)$, in mayonnaise phantoms are linearly shifted toward the negative direction as expected. Moreover, the higher concentration of mayonnaise generated the larger negative shift of GCOM and the higher value of $d_{\text{lipid}}$. The $\text{mean}\pm \mathrm{SD}$ of $d_{\text{lipid}}$ is computed as $2.82\times {10}^{-4}\pm 0.29\times {10}^{-4}$, $4.92\times {10}^{-4}\pm 0.28\times {10}^{-4}$, and $7.74\times {10}^{-4}\pm 0.45\times {10}^{-4}$ for the phantoms with the mayonnaise concentration of 9.1%, 16.7%, and 28.6%, respectively. Figure 4(b) shows that the proposed method provides consistent values of $d_{\text{lipid}}$ in a homogeneous lipid phantom. As shown in Fig. 4(c), the mean values of $d_{\text{lipid}}$ have a linear relationship with lipid concentrations. In contrast, the phantom without mayonnaise showed positive GCOM shift [black curve in Fig. 4(a)] and negative value of $d_{\text{lipid}}$ [black curve in Fig. 4(b)] due to its high water content,⁵² which has higher absorption properties at higher wavelengths.⁴⁴

Fig. 4

Result of lipid phantom analysis. (a) Plots of relative GCOM and (b) $d_{\text{lipid}}$ of each phantom. The legend in (a) denotes the concentration of mayonnaise of each phantom. (c) The linear relationship between the concentrations of phantom and the mean values of $d_{\text{lipid}}$. Error bars, SD.

In order to compare the robustness of GCOM and COM metrics, we acquired $d_{\text{lipid}}$ using both metrics from A-lines of the 16.7% mayonnaise phantom with different level of averaging (21, 41, 61, and 81 depth-resolved spectra). As shown in Fig. 5, values of $d_{\text{lipid}}$ from the GCOM metric has a lower standard deviation regardless of the averaging level, demonstrating superior robustness of the proposed GCOM as compared with the COM.

Fig. 5

Results of comparison between the GCOM and COM metric by the different level of averaging.

3.2.

In vivo Intravascular Imaging of Atherosclerotic Rabbit Arteries

To validate the feasibility of the proposed methods in clinical application, we successfully performed in vivo imaging of the LIAs of both the lipid-rich rabbits and fibrous rabbits with the IV-OCT system. In addition, images with no lesion were acquired from the RIA of the lipid-rich rabbit model that was not injured by balloon denudation. With the recorded angiography, we successfully acquired IV-OCT images sets from 45 mm segments of iliac arteries, including the LIA of lipid-rich/fibrous rabbits and RIA of lipid-rich rabbits. After in vivo imaging, rabbits were euthanized with intravenous injection of KCl, and the right and left iliac arterial segments were resected. We identified 22 OCT/histology-matched paired sections (lipid-rich: $n=12$, fibrous: $n=6$, no lesion: $n=4$) from histological sectioning of tissue (LIA of lipid-rich rabbit, LIA of fibrous rabbit, and RIA of lipid-rich rabbit).

3.3.

Mapping Results

Figure 6 shows representative results of mapping of the lipid distribution function to characterize lipid-rich plaques, which were compared with corresponding histological sections stained with ORO for visualization of fat cells and neutral fat. The first [Figs. 6(a)–6(c)], second [Figs. 6(d)–6(f)], and third [Figs. 6(g)–6(i)] rows show the results of the LIAs of the lipid rabbit model (lipid-rich plaque), fibrous rabbit model (fibrous plaque), and the RIA of the lipid rabbit model (no lesion), respectively. The first column [Figs. 6(a), 6(d), and 6(g)] represents gray-scale OCT images of each type of plaque, the second column [Figs. 6(b), 6(e), and 6(h)] represents the lipid distribution function mapping results. The third column [Figs. 6(c), 6(f), and 6(i)] represents the corresponding histological ORO-stained sections. With respect to the lipid-rich plaque, the region with a high value of $d_{\text{lipid}}$ [yellow arrowheads in Fig. 6(b)] corresponded to the area stained red with ORO [yellow arrowheads in Fig. 6(c)]. However, low values of $d_{\text{lipid}}$ were observed in the fibrous plaque and area with no lesion, except the region indicated by the yellow arrow in Fig. 6(h). This finding was attributed to the fact that images with no lesion were acquired from the RIA of the lipid rabbit without injury but with an HCD, which could have contributed to natural lipid deposition in the artery.

Fig. 6

Representative results of mapping the lipid distribution function and comparison with the corresponding histology sections. The first, second, and third rows show the results for lipid-rich plaque (a)–(c), fibrous plaque (d)–(f), and no lesion (g)–(i), respectively. Gray-scale OCT images (a), (d), (g) only provide morphological information, whereas the mapping results of the lipid distribution function derived from S-OCT (b), (e), (h) provide information about lipid composition, which corresponded well with ORO-stained histological sections (c), (f), (i). Scale bars, $500\mu \mathrm{m}$.

3.4.

En Face Visualization

To visualize the lipid distribution of the entire pullback, averaged values of the lipid distribution function along the axial direction within the tissue ROI were calculated at each rotation angle for each longitudinal position as follows:

As shown in Fig. 7(a), large values of $D_{\text{lipid}}$ were distributed throughout the pullback, whereas no significant value was observed in the pullbacks with fibrous plaque [Fig. 7(b)] or no lesion [Fig. 7(c)]. This en face visualization enables observation of the lipid distribution through the pullback.

Fig. 7

Representative en face visualization of the averaged lipid distribution along the axial direction in the pullbacks, $D_{\text{lipid}}$, with (a) lipid-rich plaque, (b) fibrous plaque, and (c) no lesion, respectively. The green dashed lines in (a), (b), and (c) correspond to the first, second, and third rows in Fig. 6. D; distal, P; proximal. Scale bar, 10 mm.

3.5.

Statistical Validation

Figure 8 shows the results of the statistical analysis for characterization of lipid-rich plaques. We acquired 97 ROIs from 22 OCT/histology-matched paired sections. From the ROC curve shown in Fig. 8(a), the area under the curve was calculated as 0.8814, and the best $d_{\text{lipid}}$ threshold value to distinguish lipid-rich plaque from nonlipid tissues was also determined as 0.004 in the arbitrary unit. Using this threshold value, we achieved a 94.3% sensitivity, 76.7% specificity, and a high correlation coefficient [$R=0.8572$, $p<0.0001$, Fig. 8(b)] between the lipid-mapping method and the ORO staining. In addition, we found that the peak values of the lipid distribution function in lipid-rich plaques were significantly larger than those of the fibrous plaque ($p<0.0001$) and in those with no lesion ($p<0.0001$), as shown in Fig. 8(c). Mean and SD of $d_{\text{lipid}}$ in no lesion were slightly larger than those in fibrous plaque, which was not statistically significant ($p=0.4039$). The insignificant difference between the two groups is likely due to a few regions with lipid found in the RIA of the lipid rabbit fed high cholesterol diet without injury.

Fig. 8

Results of statistical analysis for the characterization of lipid-rich plaques. (a) ROC curve of the performance of the method with an AUC of 0.8814. (b) Correlation between the percentage of lipid pixels in the lipid-mapping ROIs and that of ORO-stained regions in the corresponding ORO ROIs ($R=0.8572$, $p<0.0001$). (c) Comparison of the peak values for lipid-rich plaques, fibrous plaques, and areas with no lesion. Error bars; SD, asterisk; significant difference, $p<0.0001$.