Determination of Human Coronary Artery Composition by Raman Spectroscopy
Background We present a method for in situ chemical analysis of human coronary artery using near-infrared Raman spectroscopy. It is rapid and accurate and does not require tissue removal; small volumes, ≈1 mm3, can be sampled. This methodology is likely to be useful as a tool for intravascular diagnosis of artery disease.
Methods and Results Human coronary artery segments were obtained from nine explanted recipient hearts within 1 hour of heart transplantation. Minces from one or more segments were obtained through grinding in a mortar and pestle containing liquid nitrogen. Artery segments and minces were excited with 830 nm near-infrared light, and Raman spectra were collected with a specially designed spectrometer. A model was developed to analyze the spectra and quantify the amounts of cholesterol, cholesterol esters, triglycerides and phospholipids, and calcium salts present. The model provided excellent fits to spectra from the artery segments, indicating its applicability to intact tissue. In addition, the minces were assayed chemically for lipid and calcium salt content, and the results were compared. The relative weights obtained using the Raman technique agreed with those of the standard assays within a few percentage points.
Conclusions The chemical composition of coronary artery can be quantified accurately with Raman spectroscopy. This opens the possibility of using histochemical analysis to predict acute events such as plaque rupture, to follow the progression of disease, and to select appropriate therapeutic interventions.
Ruptured atherosclerotic lesions run a high risk of thrombus formation at the rupture site, acutely changing the severity of the lesion and accelerating clinical symptoms. Lesion composition, rather than lesion area or volume, is the major determinant of whether a stable or slowly growing arterial lesion will rupture,1 2 3 but a technique to study lesion chemistry in living patients has not been available. Arteriography, which quantifies stenosis severity, does not reliably predict plaque instability because aside from recognizing dense calcification or thrombus, this technique gives little information about the chemical composition of an atherosclerotic plaque.4
Previous studies have shown that vascular tissue components can be identified with FT Raman spectroscopy.5 6 7 Although these experiments demonstrated proof of principle, the 30-minute signal collection times are not suitable for clinical applications. The advent of single-stage CCD/spectrometer systems, high-rejection long-pass filters (60 db/300 cm−1), and compact laser-diode systems has made Raman spectroscopy possible in hospital clinical settings, in which spectra can be collected within <1 second.8 9 10 This opens the possibility of developing Raman spectroscopy as an intravascular technique for rapidly assessing the chemical composition of arterial tissue in vivo during vascular surgery or cardiac catheterization. We are developing the scientific basis and technology for such intraoperative and percutaneous applications.5 7 8 9 10
In the present study, we report the development of a method of analyzing Raman spectra to quantify the amounts of FC, CE, TG and PL, and CS present in a small volume of arterial tissue. The relative weights calculated from a tissue spectrum agree with those obtained with traditional lipid chemical assays and calcium mineral assays conducted on the same spectroscopically examined sample. This chemical information, when obtained in a clinical setting via optical fiber–based catheters, may be useful in diagnosing atherosclerotic lesions and guiding medical intervention.
Human coronary artery segments were obtained from nine explanted recipient hearts within 1 hour of heart transplantation. The arteries were rinsed with phosphate-buffered saline s (pH 7.4), frozen in liquid nitrogen, and stored at −85°C until use. Before use, specimens were allowed to reach room temperature.
Raman spectra from the intact segments were collected using the spectrometer system that we describe. The adventitia was then removed. Finely ground minces were prepared from artery segments from each explanted heart by grinding them in a mortar and pestle containing liquid nitrogen. Segments harvested from the same heart and exhibiting different stages of atherosclerosis were selectively combined before grinding to produce minces with varying amounts of lipids, specific lipid classes, and calcium minerals. The adventitia was included in a few minces to increase TG content. Homogenized tissue was used to ensure that the spectroscopically examined volume was representative of the much larger volume to be assayed chemically, thus facilitating comparison of the results of Raman and chemical assays.
Each homogenized artery mince was examined spectroscopically at 10 different sites on the mince with a specially developed near-infrared CCD-based Raman spectrometer system, described in detail elsewhere (Fig 1⇓).9 Samples were irradiated in a 100-μm-diameter spot with 350 mW of 830 nm infrared light. The Raman light emitted from the sample was collected and imaged onto the entrance slit of a spectrograph/CCD system with a spectral resolution of 8 cm−1. In this spectral region, 850 to 1000 nm, the fluorescence background is 1 order of magnitude larger than the Raman spectral bands, and the S/N is primarily determined by the shot noise from this fluorescence. The spectra were collected in ≈60 seconds, which produced spectra with S/N sufficiently high that the subsequent analysis was not affected by the remaining noise.11
After each spectrum was frequency-calibrated and corrected for chromatic variation in spectrometer system detection, a fourth-order polynomial was fit to it by LSM9 and this polynomial was subtracted from the spectrum to derive a filtered Raman spectrum.12 Although filtering removes the DC and low-frequency spectral components and thus alters the original relative heights of Raman bands in a given spectrum, the filtered spectra contain the vibrational information required to extract compositional data. In the following, a spectrum so derived is called a Raman spectrum.
A model was developed to extract relative weight fractions of arterial constituents from the Raman spectra. The model treats a coronary artery Raman spectrum as a linear superposition of spectra of individual chemical compounds. Details of the model, including the component spectra used, are given in “Appendix 1.” We found that spectra of seven compounds (two delipidized tissues, three lipid classes, calcific plaque, and β-carotene) were needed to model accurately the spectra collected from the full range of intact coronary artery segments; this was established by computing the residuals obtained by subtracting the model spectra from the observed spectra. A linear combination of these seven component spectra, appropriately scaled, was used to estimate the fractional weight of each compound at a particular artery site. The accuracy of the model was established by comparing the relative amounts of chemical compounds in homogenized minces obtained from the Raman spectra with those measured with standard assay techniques, as discussed in “Appendix 2.”
The Raman spectra of intact coronary arteries exhibiting a wide range of atherosclerotic progression are well described using the seven spectral components, judging from the residuals of the fits to the observed spectra (Fig 2⇓). The noise level of these spectra is comparable to the thickness of the line used to plot the residuals, and the small spectral features in the residuals are likely due to imperfect wave-number calibration of the spectra, unmodeled compounds, or slight compositional variations in the seven compound groups at the different artery wall locations.
Fig 3⇓ (left) shows the amounts of lipids and CS in the artery minces, obtained from the Raman spectra, plotted against those measured with standard assay techniques. The quantities determined by standard assays are given as a percentage of the total dehydrated weight of each mince, whereas the quantities determined with Raman spectroscopy are given as the amount of each compound relative to the other six. The horizontal error bars (±1 SD) denote the variability in amounts measured by standard chemical assays from five fractions of each mince, and the vertical error bars indicate the corresponding variability of the 10 spectra taken from each mince. Linear regression was applied to each of the plots to yield the slope, M, the y axis intercept, y0, and the correlation coefficient, r. The values are listed in the Table⇓.
The extent of agreement between the Raman spectral model calculations and standard chemical assays can be assessed by plotting the differences between the results of the two methods against their means13 (see Fig 3⇑ [right]). The differences between the two methods can be summarized by calculating the bias, estimated by the mean difference, and the limits of agreement, ±2 ς, with ς the standard deviation of the differences.13 The bias and limits of agreement are plotted in Fig 3⇑ (right) as dashed and solid lines, respectively. Their values are listed in the Table⇑. The slope of the line fit to the CS comparison data is 1.5 (Fig 3f⇑, left), a significant departure from the unity line signifying perfect agreement, so in analyzing the CS measurement accuracy, the CS contents calculated with Raman spectra were first divided by 1.5.
If the differences are normally distributed, ≈95% of these data should fall between the limits of agreement. Inspection of the graphs in Fig 3⇑, right, indicates that this is the case and that in each plot the differences cluster tightly. This indicates that results obtained with Raman spectral analysis agree with those of standard assay to within a few percentage points of the dry weight of the coronary artery minces.
The excellent fits of the model to Raman spectra from coronary artery segments exhibiting a wide range of disease progression (Fig 2⇑) indicate its applicability to intact tissue segments.
The mince studies show that Raman spectral analysis can quantify the relative weights of FC, CE, TG, and PL present in a 1-mm3 volume of coronary artery tissue7 with a maximum error of a few percentage points. The Raman-determined CS content can also be made to agree with those of the chemical assay to within a few percentage points. The correlations between relative weights of chemical and spectral assay are all highly significant (P<.005).
For the total lipid, TG/PL measurements, the mean and SD values characterizing the differences between Raman and chemical assay techniques are quite small, ≈≤2.5%. The mean difference between the standard and Raman assay for TC content is ≈−2%, whereas those for the FC and CE comparisons are ≈−2% and ≈0%, respectively. This suggests that FC may be overestimated slightly by the spectral model, causing the calculation of TC content to be a little high.
We have not identified the source of the discrepancy between the Raman and chemical CS assays. As discussed in “Appendix 1,” the relative Raman cross section of CS in calcific plaque was inferred by assuming that the 960 cm−1 vibration was primarily caused by hydroxyapatite. Scaling factors, which could be inaccurate, were used to extrapolate the total CS weight from the hydroxyapatite content calculated from the Raman spectra. The chemical assay also used scaling factors to infer CS content from the phosphorous assay. Nevertheless, by rescaling the CS content determined by Raman assay, we could predict accurately (within 2%) the CS content estimated using the inorganic phosphorous assay. Therefore, assuming the chemical assay to be correct, the rescaled Raman assay can be used to accurately measure CS content.
Our technique for analyzing the Raman spectra was developed to quantify the lipid and CS content of arterial wall. Other compounds, such as proteins, glycosaminoglycans, and DNA are also present in arteries, and it may be possible to expand the current spectral model to quantify the concentrations of these compounds. In future work, it may also be possible to quantify macromolecules such as lipoproteins.
Spectroscopic techniques based on fluorescence have been used to classify atherosclerotic plaques.14 These techniques are not capable of providing quantitative chemical information because the fluorescence spectra of many arterial components are similar.15 We earlier developed a Raman spectral model to calculate the relative weight fractions of lipids in human aorta and obtained preliminary verification of the model using chemical assay.16 17 18 This earlier model used a set of arterial components that included dehydrated elastin and collagen (type I), which have higher-energy amide I vibrations than proteins in hydrated artery samples.19 To improve accuracy, the model of the present study used line shapes of compounds directly extracted from coronary artery.
Coronary artery minces were used in this study because the results from initial studies on intact (unground) coronary artery samples gave considerable spread to the Raman assay data, as measured against standard assay.11 We concluded that point-to-point spatial variations in the structure of arterial lesions made it difficult to compare the ≈1-mm3 region sampled with Raman spectroscopy with the much larger volume sampled by conventional assay. Therefore, in the ensuing work, comparisons were made with homogeneous minces, so the chemically assayed volume of tissue would correspond to that examined spectroscopically. Although in the present study we used tissue homogenates rather than whole-tissue samples, we have just completed a study on intact artery wall that demonstrates that the quantitative histochemical information provided can be used to accurately classify intact arteries with coronary artery disease. This work will be reported separately (T.J. Römer, J.F. Brennan, M.L. Feldstein, M. Fitzmaurice, J.R. Kramer, R.S. Lees, M.S. Feld, unpublished observations).
The present work demonstrates that Raman spectroscopy can be used to accurately quantify the chemical composition of coronary arteries. This study is part of a program to develop a catheter-based methodology for intravascular Raman spectroscopy. In such a system, infrared light is delivered to the arterial wall, and the return Raman signals are collected percutaneously via an optic-fiber catheter. The Raman spectra can be processed rapidly to provide histochemical and histopathological information. The technique is applicable to peripheral arteries (J.P. Salenius, J.F. Brennan, A. Miller, Y. Wang, T. Aretz, B. Sacks, R.R. Dasari, M.S. Feld, unpublished observations). Optic-fiber catheters for use in the catheterization laboratory to collect Raman spectra of arterial plaque in vivo are also under development.9
In vivo information about the chemical composition and histopathology of atherosclerotic lesions may provide a powerful new way to study the progression of atherosclerotic plaque. Intravascular Raman spectroscopy has the potential to predict acute events such as plaque rupture. It may be useful in monitoring drug therapy and should enable the study of the chemistry of disease progression in vivo.
Selected Abbreviations and Acronyms
|DA I||=||delipidized artery I|
|DA II||=||delipidized artery II|
|FC||=||unesterified (free) cholesterol|
Each coronary artery spectrum was modeled in the Raman shift range of 1820 to 685 cm−1 with seven spectral components (Fig 4⇓): two of delipidized tissue, three of lipids, one of calcific plaque, and one of β-carotene. The linear independence of these seven components was verified; the intensity values that compose each spectrum can be found in Ref. 11. The spectra from some coronary artery components, such as proteins and TG, are mixtures of closely related molecules that would be difficult to duplicate with commercially available chemicals. We approached this problem by extracting these compounds from arterial wall and obtaining spectra from them. The spectra of a given compound group, eg, TG or calcific plaque, do not appear to change significantly among patients or coronary artery wall locations within a patient. Four of the seven spectra were measured from components directly extracted from coronary artery samples (see below).
Previous studies with FT Raman spectra of human aorta (1.06-mm excitation wavelength) showed a linear relationship between infrared Raman signal intensity and concentrations of individual components.16 17 18 Those studies also demonstrated that a linear superposition of component spectra can be used to estimate chemical concentrations in mixtures. Because the tissue attenuation does not vary appreciably over the spectral range of interest, we could model the near-infrared Raman spectrum R(v), with Raman frequency shift v, as a linear superposition of spectral components: The density ρi=mi/Vo, with mi the mass of the ith component sampled per volume Vo. K is a constant that depends on signal collection geometry, laser excitation power, and other factors that may vary among measurements. The quantity li(v) is the Raman line shape of the ith component, weighted to account for its Raman cross section relative to that of the other components. As discussed below, relative cross sections are sufficient for extracting the fractional weights of the constituents; absolute Raman cross sections are not required.
The measured Raman spectrum is modeled as a linear combination of line shapes li(v), with the fit coefficients xi selected by the LSM routine: In a comparison of Eqs. 1, and 2, the relative weight fraction of a component is given by the following: which is independent of the volume sampled and the overall Raman signal strength. If the sampled volume is homogeneous, relative weight fractions are also independent of signal collection geometry and depth of penetration of the excitation light into the tissue. A spectrum measured from a volume containing equal weights of any two components will yield identical fitting parameters xi for these two components when weighted line shapes are used.
Spectral features of TG and cholesterol were identified in many coronary artery spectra, so spectra of these compounds were included in the LSM model. Both FC and CE can be present in the arterial wall,22 23 24 and spectra of both were required to model the spectral contribution of cholesterol.
TG were extracted from the artery. Lipids were extracted from adventitial adipose tissue with Folch’s method,23 and the extracted lipids, >95% TG as measured with TLC, were examined spectroscopically (Fig 4e⇑). The extracted fat line shape accounts for all noncholesterol components that contain fatty acids, which include TG, PL, and free fatty acids.
Because FC is a specific molecule (as opposed to a lipid subclass), a spectrum of commercially available purified FC (Sigma C-8667) was used to model FC spectral features in the artery (Fig 4c⇑). However, CE are a mixture of cholesterol esterified to any one of several fatty acids. In a simple approximation, the CE spectrum is the sum of an FC and a fatty acid spectrum, appropriately weighted. In reality, the ester linkage perturbs the original unesterified FC spectrum. To generate a corrected spectrum of the cholesterol moiety of CE, the spectra of FC (Fig 4c⇑) and linoleic acid were fitted to a measured cholesterol linoleate spectrum by LSM. The linoleic acid contribution to this CE spectrum was subtracted to yield a “CEC” spectrum (Fig 4d⇑).
The extracellular proteins of the arterial wall vary widely with the state of the wall. To model the spectrum of these proteins, we immersed several noncalcified arteries for >24 hours in CHCl3/CH3OH (2:1 vol/vol) to remove lipids, rinsed them several times with saline, and examined them spectroscopically. Chloroform and methanol spectral features were not visible in the measured spectra. Although mostly protein, the samples so treated contained multiple other compounds, including glycosaminoglycans and DNA. Because our objective was to subtract the contribution of all of these from those of lipids and calcium mineral in arterial wall, we grouped them together in our model. From an analysis of the entire set of delipidized artery spectra, two distinct spectral components were identified; these are referred as DA I and DA II in Fig. 4a⇑ and 4b⇑. With inclusion of the spectra from these two components in our model, the spectra from all our artery samples could be modeled quite well.
Calcium Mineral and Other Components
Calcific plaques exhibited distinct spectral features, and these did not differ greatly from specimen to specimen. Therefore, to account for the spectral contribution of CS, a region of calcific plaque was dissected from a highly diseased coronary artery and examined spectroscopically (Fig 4f⇑). The Raman spectrum from 900 to 1125 cm−1 was used in the model because only that portion exhibited contributions from calcified material. Carotenoid features were modeled by a spectrum of crystalline β-carotene (Fig 4g⇑). Water, which accounts for ≈80% of coronary artery weight, was not required in the model; its relative Raman cross section in this spectral region is ≈100 times smaller than that of FC, and thus its spectral contribution is negligible.11
Weighted Line Shapes
The relative Raman signal strength per unit mass was determined for each of the seven compounds. Because FC has a definable molecular weight and a known Raman cross section, its spectrum was used as a standard to which the other component spectra were normalized. The amplitude of the carbon-hydrogen vibration features at 1439 cm−1 was set equal to unity, and the amplitudes of the remaining six spectra were then adjusted to produce a set of weighted line shapes.
Elastin was used for convenience to relate the delipidized artery spectra to the normalized FC spectrum. Nine different mixtures of elastin (from bovine neck ligament) and FC were examined spectroscopically, along with pure FC and pure elastin, at four sites on each mixture. To model these mixtures, the spectrum of pure elastin and the normalized FC spectrum were used as LSM spectral components. The amplitude of the elastin spectrum was adjusted so that the relative weights calculated from the Raman spectra agreed with the measured weight fractions.11 We have found (unpublished data, 1995) the Raman cross sections (by weight) of the amide I vibration of elastin and collagen to be roughly equal (within 20%), in agreement with the findings of other researchers.16 17 With the assumptions that most protein substances have similar cross sections and the delipidized tissue is mostly protein, the amplitudes of the amide I vibration (≈1750 to 1600 cm−1) of DA I and DA II were equated to that of elastin.
Next, the CEC spectrum was normalized to that of FC by matching the unesterified features of the CEC spectra (Fig 4d⇑) to the FC spectrum. To weight a fatty acid line shape, a cholesterol oleate spectrum was modeled using CEC and oleic acid spectra as LSM spectral components. Because cholesterol oleate is ≈60% cholesterol and ≈40% oleic acid by weight, the amplitude of the oleic acid spectrum was adjusted to bring the relative weights calculated from the Raman spectra into agreement with these values. The amplitude of the spectrum of the almost pure TG extracted from adipose tissue was equated to that of the oleic acid spectrum because a single fatty acid provides approximately the same density of scattering centers as TG (although TG, which contain three fatty acids, occupy about three times more space than one fatty acid).
In CE, the fatty acid moiety weighs about two thirds that of the cholesterol moiety. Because the CEC line shape does not model the spectral contribution of fatty acids in CE spectra, the TG line shape was used to model these contributions in the observed Raman spectra. To correct for this excess TG content, the CEC fitting parameter (xce) was multiplied by two thirds and the product was subtracted from the fat-fitting parameter and added to xce before calculation of TG and CE weight fractions.
We used four hydroxyapatite/FC mixtures to scale the calcific plaque spectrum relative to the normalized FC spectrum. The amplitude of a hydroxyapatite spectrum was adjusted so that the relative weights calculated from the Raman spectra agreed with the measured weight fractions,11 in a manner similar to the method used to determine the lipid/protein correlation. The amplitude of the 960 cm−1 band of the calcific plaque spectrum was equated with that of the hydroxyapatite spectrum. A small satellite band in the calcific plaque spectrum (Fig. 4g⇑) at 1070 cm−1 is not well modeled by hydroxyapatite and probably arises from carbonated apatites.17
Because they possess a relatively large Raman cross section in comparison with the other six spectral components, carotenoids sometimes contribute spectral features to coronary artery spectra, although their weight contribution is well below 1% of the total artery weight. A β-carotene spectrum was used as a spectral component to fit the carotenoid features observed in artery spectra, but its LSM fitting parameter was not used in the calculation of relative weights because its weight contribution is so small.
Each of nine minces was divided into five fractions, creating 45 different minces that were individually assayed for the content of TG, CE, FC, PL, and inorganic phosphorous. The mean weight and SD for each component were calculated from the results for each set of five fractions. The mean values were taken as the content of each component in the original mince, and the SDs were taken as estimates of random assay error. Before being chemically assayed, each mince fraction was lyophilized for >12 hours and weighed.
To each dehydrated mince fraction, 0.01 μCi of [4-14C]FC (Amersham) was added as an internal standard to allow correction for the recovery of lipids during subsequent lipid extraction.23 The extracted lipids were dissolved in 1.0 mL of CHCl3/CH3OH (2:1 vol/vol), and the radioisotope content was determined to be 98.7±4.6% of the initial dose by counting aliquots in a β-scintillation spectrometer (Wallac Rac β-1217). Total lipid weight in each mince fraction was determined gravimetrically by microbalance (Cahn C-31), after solvent evaporation, by averaging three separate 100-μL aliquots from each sample.
The four major lipid classes were separated by spotting appropriate aliquots of the remaining lipid solution from each mince fraction on prewashed silica gel G TLC plates (Analtech), which were developed in a chamber with hexane/ethyl ether/:acetic acid (70:30:1 vol/vol/vol). Authentic standards of known mass were included on all plates, and the amounts of the standards and samples were selected to fall within the linear range of the densitometer used during subsequent analytical procedures. After air-drying, the developed plates were sprayed evenly with 18 N H2SO4 and charred on a hot-plate (≈220°C) until all of the H2SO4 fumes had dissipated. Each TLC lane was scanned, and the absorbance of the charred lipid bands was measured with a densitometer (Molecular Dynamics). The integrated optical density across each band was used to compute the percentage weight distribution among the four lipid classes. The band optical densities were correlated to the lipid weights via the authentic standards.24
The PL-free arterial tissue that remained in each mince fraction after delipidization was assayed for inorganic phosphorous content.25 Hydroxyapatite [Ca5(PO4)3(OH)] has a molecular weight of 502.3, of which 92.9 is due to phosphorous, so the measured phosphorous content was multiplied by 502.3/92.9 ≈ 5.4 to estimate the amount of hydroxyapatite originally present in each fraction. The hydroxyapatite content of three delipidized artery calcifications were estimated with the same assay to be ≈70% of the total weight, which is in agreement with the results of other researchers.26 Therefore, the hydroxyapatite content in each fraction was multiplied by 100/70 to estimate the calcium mineral content.
Financial support from the National Institutes of Health (NIH R01-HL-51265 and NIH P41-RR-02594) and the Netherlands Heart Foundation (R-93310) is gratefully acknowledged. This research is part of the doctoral thesis of James Brennan.11 We are indebted to Drs R. Dasari, R. Manoharan, M. Feldstein, and M. Goormastic for technical assistance and useful discussions. The research was conducted at the MIT Laser Biomedical Research Center.
- Received October 9, 1996.
- Revision received January 9, 1997.
- Accepted February 2, 1997.
- Copyright © 1997 by American Heart Association
Small DM. Progression and regression of atherosclerotic lesions: insight from lipid physical biochemistry. Arteriosclerosis.. 1988;8:103-129.
Libby P. Molecular bases of the acute coronary syndromes. Circulation. 1995;91:2844-2850.
Rava RP, Baraga JJ, Feld MS. Near infrared Fourier transform Raman spectroscopy of human artery. Spectrochim Acta. 1991;47A:509-512.
Liu CH, Sha Glassman WL, Zhu HR, Akins DL, Deckelbaum L I, Stetz ML, O’Brien K, Scott J, Alfano RR. Near-IR Fourier transform Raman spectroscopy of normal and atherosclerotic human aorta. Lasers Life Sci.. 1992;4:257-264.
Baraga JJ, Feld MS, Rava RP. In situ optical histochemistry of human artery using near-infrared Fourier transform Raman spectroscopy. Proc Natl Acad Sci U S A.. 1992;89:3473-3477.
Brennan JF, Wang Y, Dasari RR, Feld MS. Near infrared Raman spectrometer systems for human tissue studies. Appl Spectrosc. 1997; 51:201-208.
Kramer JR, Brennan JF, Römer T J, Wang Y, Dasari RR, Feld MS. Spectral diagnosis of human coronary artery: a clinical system for real time analysis. In: Johnson DE, Abela GS, Shapshay SM, eds. Proceedings of Lasers in Surgery: Advanced Characterization, Therapeutics, and Systems V. Bellingham, Wash; SPIE, 1995:376-382.
Brennan JF. Near infrared Raman spectroscopy for human artery histochemistry and histopathology. PhD thesis, Cambridge, Mass: Massachusetts Institute of Technology; 1995.
Press WH, Teukolsky SA, Vetterling WT, Flannery BP. Numerical Recipes in C, 2nd ed. New York, NY: Cambridge University Press, New York; 1992:671-681, sect. 15.4.
Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;8:307-310.
Baraga JJ. In situ analysis of biological tissue: vibrational Raman spectroscopy of human atherosclerosis. PhD thesis, Cambridge, Mass: Massachusetts Institute of Technology; 1992.
Manoharan R, Baraga JJ, Feld MS, Rava RP. Quantitative histochemical analysis of human artery using Raman spectroscopy. J Photochem Photobiol. 1992;16:211-233.
Campbell ID, Dwek RA. Biological Spectroscopy. Menlo Park, Calif: Benjamin/Cummings; 1984:48.
Böttcher CJF, Keppler JG, Ter Haar Romeny-Wachter CC, Boelsma-van Houte E, Gent CM. Analysis of lipids of the arterial wall. Lancet. 1958;1207-1209.
Böttcher CJF, Boelsma-van Houte E, Ter Haar Romeny-Wachter CC, Woodford FP, Gent CM. Lipid and fatty-acid composition of coronary and cerebral arteries at different stages of atherosclerosis. Lancet. 1960;2:1162-1166.
Folch J, Lees M, Sloane Stanley GH. A simple method for the isolation and purification of total lipids from animal tissues. J Biol Chem. 1957;226:497-509.
Bartlett GR. Phosphorous assay in column chromatography. J Biol Chem.. 1959;234:466-468.