2.1.

Atherosclerosis

In atherosclerosis, deposits forming atheromatous plaques develop in the artery wall, and these plaques can be more or less vulnerable to rupture constituting a risk for the patient. ^{28, 29, 30, 31} To correctly determine the diagnosis, prognosis, and the required treatment of advanced atherosclerosis, a modality capable of differentiating vulnerable plaques from stable plaques is required. However, the definition of a vulnerable plaque is a functional one and not based on anatomic appearance. Studies have indicated some common features of plaque vulnerability, such as a thin fibrous cap, a large lipid content, the presence of haemorrhagic areas, and degree of inflammation.²⁸ During the process of plaque formation, different stages can be identified, and this forms the basis for the current classification systems. A frequently used system was proposed by the American Heart Association (AHA), ^{30, 31} and is based on histological examination of arteries from autopsies. Plaques are divided into early lesions (grade I-II), intermediate lesions (grade III), and advanced lesions (grade IV-VI). The use of roman numerals indicates a successive development through all stages which is not necessarily the case, as pointed out by Virmani, who have proposed changes to the AHA-classification.³²

2.2.

Optical Properties

To detect and characterize atherosclerotic plaques it is necessary to identify the various chromophores and fluorophores of both healthy and pathologic tissue.

The most important chromophores in this application are different forms of hemoglobin, lipids/ carotenoids, connective tissues (collagen and elastin), and calcifications. Hemoglobin is a pronounced absorber with characteristic absorption spectra according to the oxygenation of the molecule, see Fig. 1. Deoxygenated hemoglobin has absorption maxima at 433, 554, and 760 nm, and appears dark red with a bluish tint. Oxygenated hemoglobin has absorption maxima at 414, 542, and 576 nm,³³ and appears bright red. Methemoglobin is a defect hemoglobin molecule (caused by oxidative stress) with no oxygen carrying capability, and appears dark brown with absorption peaks at 404, 508, and 635 nm (see Fig. 1). ^{34, 35}

Fig. 1

Molar extinction coefficients of oxygenated and deoxygenated hemoglobin (Ref. 33), methemoglobin (Ref. 34), and beta-carotene (Ref. 37).

Lipid rich plaques have a higher concentration of carotenoids, mainly betacarotene, than normal aortic tissue.³⁶ Beta-carotene has two characteristic absorption peaks at 450 and 480 nm, ^{37, 38} (see Fig. 1), and accumulates in fatty tissue causing the yellow color of such tissue. Beta-carotene is therefore an indicator of lipids. Likewise, superficial foam cells, which may be an indicator of plaque vulnerability, have been shown to have beta-carotene absorption, as such foam cells may contain lipids.⁴ Water absorption is low in the visible region, but has absorption peaks in the NIR and IR regions. The absorption peak at 980 nm can be detected with the camera setup used in this work. Scattering from biological tissue is a combination of Mie- and Rayleigh scattering, and generally falls off monotonously with increasing wavelength.³⁹ In soft tissues, scattering is most pronounced for shorter wavelengths where it is dominated by small fibrils such as collagen.

The major fluorophores expected to be present in normal aortic tissue and atherosclerotic plaques are collagen, elastin, and carotenoids. ^{7, 40} If multiple fluorophores are present, the pure spectra of each fluorophore are superpositioned, and if other absorbers are present the fluorescence spectra will be affected accordingly. Collagen is known to exhibit strong autofluorescence at approximately 450 nm.⁴¹

3.1.

Aortic Samples

Measurements were performed on sections from the aorta excised during autopsies. A total of 11 samples were obtained from males 65 to 85 years old, eight of the eleven subjects died from complications associated with the circulatory system/ cardiovascular diseases. Acquisition of the hyperspectral images were performed 1 to 4 days post mortem. The samples had various degrees of atherosclerosis. After excision, the aorta sections were opened longitudinally to provide samples of size approximately 6×8 cm. They were then rinsed with saline to remove superficial blood, wrapped in a moist towel (saline) and kept cool for storage, transport, and in between measurements. The samples were placed on a piece of black, nonfluorescent plastic to ensure even illumination and avoid (fluorescent) light originating from the background. Fluorescence images were obtained after white-light images had been recorded. Hyperspectral data from the lumen was recorded with a HSI-system using a white-light lamp and a UV-laser for illumination.

The project was approved by the Regional Committee for Medical and Health Research Ethics, Norway.

3.2.

Hyperspectral Camera Setup

The imaging system was based on a HySpex VNIR-1600 (Norsk Elektrooptikk AS, Oslo, Norway) push-broom hyperspectral camera. The camera utilizes a polarization independent grating and a Si-CCD detector (1600 spatial pixels) which yields 160 bands in the 410 to 1000-nm wavelength region (3.7-nm separation).⁴² The camera records one line at a time, hence scanning along one axis is necessary to construct a 2D-image. The camera was mounted on a rack facing downward (birds-eye view). The samples were placed on a table mounted on a translation stage, see Fig. 2. A 150W dc white-light source with a 12.5-cm wide fiber optic line light guide (DCR II, Fostec, Auburn, New York) provided illumination of the sample. Specular reflections dramatically deteriorate image quality as the reflected light affects the measured spectrum and/or saturates the detector, masking the true reflection spectrum of the sample. The light source was set at an angle that minimized the specular reflections from the sample surface. Furthermore, as the samples are not flat, a linear polarizer film (International Polarizer Inc, Marlboro, Massachusetts) was fitted to the light line, and a second polarizer (NT52-574, Edmund Optics, York, UK) was placed in front of the camera lens and rotated to minimize specular reflections. This effectively suppressed spectral reflections, but due to limited bandwidth of the polarizers this was only effective up to 750 nm. As a consequence, all spectra were truncated to 410 to 750 nm (95 bands) for analysis.

Fig. 2

Measurement setup.

For fluorescence imaging a pulsed, frequency tripled Nd:YAG-laser (Quanta Ray Lab-series L-190, Spectra Physics, Mountain View, California) provides UV-excitation (355 nm) of the sample. The laser output was shaped to a line using two cylindrical lenses (plano-concave and plano-convex, Newport Corp., Irvine, California) and projected by a mirror via a diffusor onto the sample, as presented in Fig. 2. The diffuser broadens the line to ensure uniform excitation light distribution over the camera's field of view as the samples were not perfectly flat. The irradiance on the sample was approximately 20 to 30 mW/cm².

In Fig. 3 the concept of a hyperspectral imaging cube is shown, where two spatial axes (x, y) represent the 2D-image, and the z-axis represents the spectral information (amplitude) at different wavelength bands.

Fig. 3

Hyperspectral imaging principle.

3.3.

Image Acquisition

The samples were scanned along the x-axis to construct a 2D-hyperspectral image. A full scan takes approximately 30 s to complete, depending on the scan settings such as scan (translation) length, exposure time, and averaging.

For white-light reflectance measurements an optical reflection standard (Spectralon WS-1, Ocean Optics, Duiven, The Netherlands) was placed beside the sample and recorded in the image. The reflection standard allows for calibration of the raw radiance image versus the illumination lamp spectrum to yield the true reflectance image. A radiometric calibration was performed automatically by the camera software to compensate for the detector response and dark current noise, producing true radiance images.

Fluorescence images were obtained directly after visible images had been recorded. The camera polarizer was removed before fluorescence measurements to improve the signal-to-noise ratio as the polarizer attenuates the light. The positioning of the sample and spatial scan settings were maintained during both measurements to have co-registered reflectance and fluorescence-images.

A preliminary analysis of the hyperspectral images was performed directly after the images were acquired, and three to six areas of interest were identified on each sample. The selection of these sites was based on both macroscopic determination of plaques and lesions, and also on areas with abnormal spectra found in the preliminary analysis. Plaques assumed to be advanced lesions were preferred in the selection. These areas of interest were later excised for further histological examination. In total there were 42 biopsies from the eleven samples.

3.4.

Histological Examination and Lesion Classification

The areas of interest were excised for biopsies (10×8 mm), bisected, and inkmarked; one-half was snap-frozen in liquid nitrogen and the other half fixed in 10% buffered formalin and later embedded in paraffin. The biopsies were then sliced and stained, the frozen halves with Oil red O and the paraffin embedded with hematoxylin-eosin-saffron (HES), Van Gieson, Gordon/Sweet and elastin stains.

The stained sections from the aorta samples were examined using a standard light microscope at three magnifications (ob. 5×, 10×, and 20×), and photographed. The three layers of the arterial wall (intima, media, and adventitia) were identified and the thickness of the intima was measured and defined as the distance from the media to the lumen surface. Lesions and plaques were classified according to the AHA-classification scheme as type I-VI. ^{30, 31} Of the advanced lesions, IV and Va are lipid laden lesions considered the most vulnerable plaque types. Type Vb represent calcified lesions, and type Vc are fibrotic lesions without calcifications or a lipid core. Type VI-lesions are ulcerated (ruptured) plaques. The classification was based on plaque features such as the presence of calcium deposits, foam cells, lipid core, cholesterol clefts, macrophage infiltration, fibrous cap, media influence, and the size of these features such as cap thickness and total plaque thickness, when present. The size was found by using a ruler calibrated to an image of a microscope measurement gauge taken at the same magnification. If there was a large discrepancy of the thickness of these features across the length of the section, both the minimum and the maximum thickness were measured.

3.5.

Data Analysis

The hyperspectral camera records 160 continuous bands in the 400 to 1000 nm range for each image element, and consequently generates a large amount of data in need of further processing and analysis (approx. 1 Gb per image). Commercially available software was used for the analysis of the acquired images, Environment for Visualizing Images (ENVI and IDL, ITT Visual Information Solutions, Bracknell, Berkshire, UK). The HySpex acquisition software exports images to a standard format read by ENVI. Raw image data were converted to reflectance data in ENVI by performing a flat field calibration routine in reference to the Spectralon-tile.

4.1.

Macroscopic Evaluation of Aorta Samples

The aorta tissue samples were examined macroscopically. A photograph of a representative sample is shown in Fig. 5a. Areas of apparently normal vessel wall, as well as areas with macroscopically visible atheromatous changes, were present in all samples. The normal vessel wall had a gross appearance with smooth, pale yellow colored lining and a soft texture, whereas areas with atherosclerotic plaques demonstrated elevation of the vessel lining and had a darker yellow, red or brown color. Furthermore, the intimal surface was irregular and partially ruptured or ulcerated and the texture was firm or even hard. The plaques were of different sizes and varied in shape from round to oval or irregular. Particularly in the larger and most severe atherosclerotic areas, the appearance of one single plaque could differ considerably between different regions of the plaque, and some lesions seemed to have merged by the expansion of adjacent plaques.

Fig. 5

(a) Pseudo-RGB images of an aortic sample; white-light reflectance image and (b) UV-excitation autofluorescence image.

4.2.

Microscopic Evaluation of Aorta Tissue Samples

The atherosclerotic plaques were evident as areas of the intima with increased thickness (215 ± 167 μm), whereas in the biopsies without evident plaques (n = 4), the intima thickness was 120 ± 25 μm. It is important to notice that areas of the arterial wall without plaques also show some degree of morphological changes such as increased intima thickness, increased amount of collagen fibers in the intima, and decreased amount of elastin fibers in the media. These changes are normal with age (subjects age 72.3 ± 6.5 years, range 64 to 85), and these samples can be considered age matched controls. Most of the plaques had a cap consisting of collagen rich fibrous tissue overlying a necrotic center mainly containing lipids, partly in the form of cholesterol clefts and foam cells. In some samples, nodules of calcium were present in the center of the plaque, and in others larger areas of calcification were evident [calcified areas are usually traumatized during the section preparation, and appear as “empty” areas with surrounding remains of dark purple stained tissue as seen in Fig. 4b].

Fig. 4

(a) Photomicrographs of a grade Va-plaque and (b) a grade Vb-plaque classified according to the AHA-classification.

The plaques demonstrated a various amount of foam cells characterized by the presence of vacuoles (containing lipid) in the cytoplasm. The foam cells were, in many instances, located at the peripheral regions, the shoulder, of the plaque. However, in some cases a considerable number of foam cells were seen in the cap.

The intima thickness was measured in each sample as the distance from the lumen to the media defined as the presence of elastin fibers visible as the eosinic base-layer in Figs. 4a and 4b.

4.3.

Frequency of Population Lesion Types Based on Histology

The majority of plaques were found to be advanced lesions; 38 of the 42 areas of interest were classified as grade IV and Va (20), Vb (11), Vc (5), and VI (2) lesions. The remaining lesions consisted of three grade I- and one grade II lesion. No completely healthy aortic wall tissue (grade 0) were among the sampled biopsies, as even the most healthy regions showed intimal thickening (grade I-II), not to be unexpected in the elderly demographic. No grade III lesions were found.

4.4.

Lesion Characterization Based on Hyperspectral Image Analysis

Figure 5a shows a pseudo-RGB image of a tissue sample (from 77 y.o. male) as acquired by the HySpex-camera under white-light illumination (bands R = 608 nm, G = 557 nm, and B = 451 nm, all RGB-images presented use these bands). Figure 5b shows the same tissue sample acquired under UV-excitation. As these images clearly demonstrate, there are various regions which can easily be distinguished by the naked eye based on their visual appearance. Note how the leftmost region of the sample appears red under white-light illumination in Fig. 5a, and almost completely dark under UV-excitation as shown in Fig. 5b, while the central region appears white and bright white correspondingly. More subtle differences can be found by examining the corresponding wavelength spectra of the hyperspectral image pixels, and allows for differentiating and possibly classifying the various regions of the samples accordingly. Figure 6 shows an advanced plaque (type Vb, from 85 y.o. male) under white-light illumination, Fig. 6a, and the UV-excitation autofluorescence image of the same plaque in Fig. 6c. The corresponding K-means classification images are shown in Figs. 6b and 6d based on the reflection- and fluorescence image, respectively. Note how in both classification images there is a central region surrounded by one or more regions of different spectral characteristics determined by the K-means classification, suggesting a heterogeneous plaque. Figure 7 shows spectra from regions of the same aorta sample, however, these regions were histologically classified as different grade lesions. The spectra in Fig. 7a shows the apparent absorption spectra [-log(Reflectance)] of grade I, Va, Vb, and VI plaques and Fig. 7b show the corresponding fluorescence spectra. The spectra of the grade VI plaque differs significantly from the others as it has a much higher apparent absorption, and also expresses low fluorescence as compared to the other grade lesions. Note also that the largest absorption peak has redshifted to 433 nm as compared to 418 nm for the other spectra. The other absorption spectra appear more homogeneous, and a small local maximum can be observed around 480 nm. The fluorescence spectra however, differ significantly. The Vb-grade (calcified) plaque exhibits a very strong fluorescent signal compared to the others.

Fig. 6

(a) White-light reflectance image and UV-excited autofluorescence image (c) of an advanced plaque. The corresponding K-means classification overlays are shown in (b) (white-light) and (d) (fluorescence).

Fig. 7

(a) Apparent absorption spectra and (b) fluorescence spectra (not normalized) of grade I, Va, Vb and VI-plaques from the same aorta sample.

4.4.1.

Reflectance and fluorescence images

Figure 8 shows a HES-stained histological section of a grade Va-lesion. The plaque is the thickened, lighter part (p) on top of the darker (eosin) stained media (m). The arrow on the left indicates a shoulder-region of the plaque with foam cells and cholesterol crystallization with early stages of calcification. There are visible dense collagen fibers in the cap layer. As one moves rightward along the plaque (C and D) these collagen fibers become more separated in regions where other necrotic materials are entwined between these fibers. The underlying core (D) contains foam cells and evidence of cholesterol and necrotic material. Toward the other end of the section (E), foam cells are evident near the surface and at the shoulder (arrow to the right). Figures 9 and 10 show reflectance and fluorescence images of the same HES-stained plaque shown in Fig. 8 (from 74 y.o. male). The spectra [Figs. 9b and 10b] are harvested from sites along a straight line (indicated by letters A, B, C, D, and E), also corresponding to the direction of the cut for the histology section (Fig. 8). The letters C, D, and E along the section in Fig. 8 correspond to positions in the hyperspectral reflectance and fluorescence images shown in Figs. 9 and 10.

Fig. 8

Histological section of grade Va-plaque, letters corresponding to positions as indicated in Figs. 9 and 10.

Fig. 9

(a) Atherosclerotic plaque under white-light illumination and (b) corresponding spectra.

Fig. 10

(a) Atherosclerotic plaque under UV-excitation and (b) corresponding spectra.

Site A from the hyperspectral reflectance image in Fig. 9 is from a lesion free area of the aorta. The corresponding apparent absorption spectrum shows no distinct features other than absorption peaks in the 540 to 580 nm range, plus the slightly redshifted absorption peak in the Soret-band. Site B appears lighter than site A, and the corresponding spectrum demonstrates lower absorption, and a small local maximum can be observed around 480 nm. Site C appears even lighter and is, as established from Fig. 8, a region of the plaque with separated collagen fibers over foam cells. The apparent absorption spectrum shows lower absorption than site B, and the spectrum is quite similar to the spectrum of site D which is from a part of the plaque with a thicker lipid core. The spectrum of site E shows a large absorption in the 450 to 490 nm region.

Figure 10 shows the co-registered hyperspectral fluorescence image of the same sample. The normal tissue (site A) exhibits a pronounced fluorescence with a main peak around 510 nm, and the right (red) side fluorescence tail quenched in the 540 to 580 nm region corresponding to hemoglobin absorption. Sites B and C fluorescence spectra are quite similar in shape and had the overall lowest fluorescence of the sites. A local minimum centered around 480 nm is visible. Site D has the highest overall fluorescence with a maximum around 450 nm, and a small crease around 480 nm. Site E has a main peak above 500 nm and the spectra also display a substantial re-absorption of light around 480 nm.

Generally, autofluorescence could be detected from the entire surface of the samples, also from the regions classified as normal by histology. Calcified regions show the strongest overall fluorescence centered around 470–500 nm; the absorption effect of hemoglobin and beta-carotene could be readily seen near the circumference of calcified regions in the spectra, but to a lesser degree in the center of calcified regions. The lowest overall autofluorescence levels were observed in regions with high hemoglobin and beta-carotene absorption, classified as grade VI.

4.4.2.

PCA and statistical analysis

For reflectance data, the first and most significant PCA eigenvector reveal maxima/ minima at 420, 515, 537, 558, and 573 nm. The second PCA eigenvector has extrema at 432, 486, 541, 557, and 574 nm. The third eigenvector has maxima/ minima at 415, 464, and 603 nm. PCA of the corresponding fluorescence data revealed maxima/minima at 482, 577, and 596 nm for the most significant eigenvector; 433, 511, 541, 555, 573, and 599 nm for the second eigenvector; and 482, 537, 559, 573, and 610 nm for the third eigenvector. Based on analysis of the cumulative variance for the PCA components, it was found that the first three PCA components were significant. Analysis of the reflectance- and fluorescence spectra as two separate batches revealed only slight differences between the histology groups, although the groups were found to be robust with regard to the contribution from the individual samples (leave-one-out or Jack-knife validation). The means of each group were not statistically different due to large standard deviation. Separate PCA analysis of the histology groups show that the groups were separable with respect to the first PCA component, but not to the second or third component, see Fig. 11. The mean of the PCA coefficients was found to be roughly independent on the contribution from individual samples. The largest intergroup variance was found in the grade IV-Va group.

Fig. 11

(a) PCA based on reflectance data and (b) PCA based on fluorescence data (normalized to peak intensity).