Sources of absorption and scattering contrast for near-infrared optical mammography.

Rationale and Objectives. Near-infrared (NIR) diffuse optical spectroscopy and imaging may enhance existing technologies for breast cancer screening, diagnosis, and treatment. NIR techniques are based on sensitive, quantitative measurements of functional contrast between healthy and diseased tissue. In this study, the authors quantiﬁed the origins of this contrast in healthy breasts. Materials and Methods. A seven-wavelength frequency-domain photon migration probe was used to perform noninvasive NIR measurements in the breasts of 28 healthy women, both pre-and postmenopausal, aged 18–64 years. A diffusive model of light transport quantiﬁed oxygenated and deoxygenated hemoglobin, water, and lipid by their absorption signatures. Changes in the measured light-scattering spectra were quantiﬁed by means of a “scatter power” parameter. Results. Substantial quantitative differences were observed in both absorption and scattering spectra of breast as a function of subject age. These physiologic changes were consistent with long-term hormone-dependent transformations that occur in breast. Instrument response was not adversely affected by subject age or menopausal status. Conclusion. These measurements provide new insight into endogenous optical absorption and scattering contrast mechanisms and have important implications for the development of optical mammography. NIR spectroscopy yields quantitative functional information that cannot be obtained with other noninvasive radiologic techniques.

This article describes the application of a new class of near-infrared (NIR) spectroscopy tools for noninvasive, quantitative characterization of breast tissue physiology.Diagnostic methods currently in use, such as mammography, magnetic resonance imaging, and ultrasound (US), offer excellent anatomic lesion-detection capabilities but generally cannot provide quantitative information regarding tissue function and composition (1).Positron emission tomography shows great promise in evaluating the metabolic demands of tissue but requires exogenous radionuclides and is insensitive to tissue hemodynamics.NIR optical spectroscopy is intrinsically sensitive to the principal components of breast: blood, water, and adipose, as well as epithelial and connective tissues.Results of preliminary studies suggest that the fractional contribution of each component to NIR signals depends strongly on factors such as age, menopausal status, and the progression of disease.Thus, NIR optical spectroscopy provides an opportunity for revealing physiologic information that cannot otherwise be obtained noninvasively.As optical diagnostic methods become more clinically applicable, there is a growing need to identify and quantify sources of endogenous contrast in breast (2)(3)(4)(5)(6)(7)(8)(9)(10)(11).This article presents the results of quantitative NIR spectroscopic studies in 28 subjects ranging in age from 18 to 64 years.We provide evidence that NIR tissue spectroscopy is inherently sensitive to long-term age and hormone-dependent dynamics.Similar studies have been initiated (12)(13)(14) but in only a few subjects or with limited sensitivity to physiologic parameters.These data are expected to play an important role in characterizing the sensitivity of NIR optical spectroscopy and imaging for applications such as tumor detection, therapeutic monitoring, and risk assessment.
NIR photons (600 -1,000 nm) are nonionizing and weakly absorbed by tissue.Many investigators have used them to explore brain, breast, and muscle tissue, exploiting their enhanced penetrance relative to that of visible or ultraviolet radiation (1).In breast tissue, the principal NIR absorbers are assumed to be reduced hemoglobin (Hb-R), oxygenated hemoglobin (Hb-O 2 ), water (H 2 O), and lipids (15).NIR tissue spectroscopy offers a safe way to quantify these components and to view unique functional information with low-cost "point-of-care" devices.Multiple light scattering complicates quantitative measurements of light absorption in the NIR range.The process of using frequency-domain photon migration (FDPM) techniques to separate the effects of absorption from scattering has been described in detail (16,17).FDPM enables the use of quantitative spectroscopic analysis tools to determine the composition and structure of tissue noninvasively (18).
The application of FDPM techniques to NIR tissue spectroscopy differs substantially from conventional bulk NIR methods, such as diaphanography (19).FDPM employs computational models that allow the quantitative separation of the effects of absorption and scattering (20).Thus, within the limits of applicability of the light transport models to the breast, all parameters measured in this study are absolute values.

Patient Selection
All volunteers enrolled in this study competently provided informed consent to participate in one of two trials (nos.95-563 and 99-2183) under the guidelines of an Institutional Review Board.The 28 women ranged in age from 18 to 64 years.Fifteen of them were premenopausal (average age, 28 years Ϯ 9), and seven were postmenopausal (average age, 56 years Ϯ 2).The remaining six women (average age, 56 years Ϯ 5) were taking some form of hormone replacement therapy (HRT); three of the six classified themselves as perimenopausal.None of the women had any known cancerous lesions or other known forms of breast disease.

Measurement Technique
Each subject rested in a supine position during the measurement.The instrument probe, slightly larger than a US probe (Fig 1 ), was the only item placed in contact with the subject.It consisted of a sealed APD module and an optical fiber placed 22 mm from the APD.All measurements were performed in a reflection-style geometry.The probe was placed on the breast with minimal pressure and only the force of gravity; no compression was used.In this configuration, we estimate that the mean sampling depth was approximately 1 cm below the skin.We report data only on measurements performed in the center of the left upper outer quadrant (measurements on the right upper quadrant yielded similar results).Error bars in the figures and tables represent the standard deviations (SDs) of repeated measurements.

Instruments
A schematic of our 1-GHz, portable FDPM device is shown in Figure 1.This FDPM instrument has been described in detail elsewhere (21).It employs multiple diode lasers that provide visible and NIR light at seven wavelengths (672, 800, 806, 852, 896, 913, and 978 nm).A hand-held probe has been designed to house an APD that records the modulated diffuse light signals after propagation through the tissue.This probe has a plastic attachment on the casing to position a source optical fiber a fixed distance from the APD.A 100-m-diameter graded index optical fiber positioned 22 mm from the APD detector was used to deliver the diode laser outputs to the tissue surface.A network analyzer (8753c; Agilent, Palo Alto, Calif), measured the phase and amplitude of the detected electronic signal from the APD over a range of source modulation frequencies.A steady-state current source was mixed with radiofrequency power provided by the network analyzer in a bias network.This bias network serially distributed power to each laser diode and produced intensity-modulated light.An optical switch delivered light serially from each diode to the tissue via the single optical fiber described above.The optical powerlaunched into the tissue ranged from 5 to 25 mW for each wavelength.
A sweep over all seven wavelengths took approximately 35-60 seconds.The system acquired data in less than 3 sec-onds per wavelength, with a 2-second delay between each laser diode in the system because of switching considerations.The system was wheeled into a medical clinic for each measurement.Instrumental artifacts were removed by means of calibration on a tissue-simulating phantom with known absorption and scattering properties.

Measured Parameters and Data Analysis
The amplitude and phase of a NIR diffusive light wave demodulates and retards, respectively, as the wave propa-gates through multiple-scattering media such as tissue.The real and imaginary parts of this diffuse wave were fit simultaneously to a light-diffusion model (the P 1 approximation to the Boltzmann transport equation) (22,23) by minimizing the 2 surface with a Marquardt-Levenberg algorithm.This fit determined the absolute optical absorption coefficient a and the absolute optical reduced scat- tering coefficient Ј s at each wavelength.Typical a and Ј s uncertainties, determined from the 2 distribution of the fits, are 3% to 1% of their mean values.Precision errors in actual tissue measurements were determined by performing multiple network analyzer sweeps and multiple probe placements on a given location and were found to be less than 10% for a and 5% for Ј s .
When the optical properties Ј s and a are recovered for the seven wavelengths, the spectral dependence of the absorption may be combined with known values of molecular extinction coefficients to calculate physiologically relevant parameters.We assume that bulk breast tissue is composed principally of four NIR absorbers: Hb-R, Hb-O 2 , H 2 O, and lipids.The concentrations of these absorbers (c) were quantified by solving the equation ជ a ϭ Ec ជ, where E is a 7 ϫ 4 matrix that contains the molar extinction coefficients of the four chromophores at the seven source wavelengths (24 -26).
For each measurement, we report four hemoglobin parameters: [Hb-R], [Hb-O 2 ], the total hemoglobin concentration (THC ϭ [Hb-R] ϩ [Hb-O 2 ]), and the tissue hemoglobin saturation (S t O 2 ϭ [Hb-O 2 ]/THC ϫ 100%), where the characters in square brackets denote concentration in moles per liter.Values for water and lipid content are reported as percentages.The water percentage is the concentration of measured tissue water divided by the concentration of pure water (55.6 mol/L).The lipid percentage indicates the kilograms per liter of lipid measured relative to an assumed "pure" lipid density of 0.9 kg/L.Thus, the water and lipid percentages we report are relative figures of merit compared with pure solutions of the substance and are not strict volume or mass fractions.
The scattering properties of the tissue also yield important physiologic information.NIR scattering in tissue has the following dependence: Ј s ϭ A ϪSP , where A is a constant, is the wavelength (in nanometers), and SP is the scatter power.Scatter power is related to the scattering center size (d) compared with the optical wavelength.As an example, scatter power is 4 in the case of Rayleigh scattering (d Ӷ ) and is approximately 1 for large Mie-like scatterers (d ϳ ).

Sensitivity: Pre-versus Postmenopausal Subjects
Both Figures 2 and 3 present a typical series of measurements of seven absorption and scattering coefficients, respectively.The points represent an average of several measurements in the center of the left upper outer breast quadrants in two subjects, a 32-year-old premenopausal woman and a 54-year-old postmenopausal woman.Error bars indicate the SDs of repeated measurements.There are vast absorption and scattering differences between pre-and postmenopausal breast tissue.The solid lines in Figure 2 represent a weighted least-squares fit of the seven absorption coefficients, with published extinction coefficients used for Hb-R, Hb-O 2 , H 2 O, and lipids (24 -26).Lines between the measured points have been interpolated.The solid lines of Figure 3 1.

Figure 3.
Measured scattering spectra for the same subjects as in Figure 2. The lines represent a fit to Ј s ϭ A ϪSP , where Ј s is the reduced scattering coefficient (per millimeter), A is a constant, is the wavelength (in nanometers) and SP is the scatter power.Results of the fit from this measurement are listed in Table 1.Error bars show the SDs of repeated measurements.
These optical spectra provide insight into breast tissue composition.There are higher concentrations of hemoglobin (ie, both Hb-R and Hb-O 2 ) in the premenopausal subject, as evidenced by the overall higher absorption in the 670 -850-nm range.There is also more water relative to lipids in the premenopausal subject, as revealed by the large water absorption peak at 980 nm.Recovery of the absorption spectra allows calculations of the tissue concentrations of Hb-R, Hb-O 2 , H 2 O, and lipids.In addition, the light scattering intensity is substantially lower in the postmenopausal breast.Table 1 summarizes the fitted physiologic properties for the two subjects represented in Figures 2 and 3.

Quantities versus Age
The next three figures present FDPM-measured quantities as a function of subject age. Figure 4 shows the THC, Figure 5 shows the tissue water percentage (referenced to the concentration of pure water), and Figure 6 shows the scatter power.Premenopausal subjects (ie, aged younger than 50 years) display a variety of values in all three plots.This spread is the result of intersubject variations, including, but not limited to, menstrual cycle differences and gynecologic age.The THC, and to a lesser extent the water, seemed to increase in premenopausal subjects (aged 18 -39 years), perhaps reaching a peak value near age 30 years.Between the ages of 40 and 49 years, the THC appears to level off, while water and scatter power seem to decline.After age 50 years (predominantly postmenopausal subjects), there is a general decrease with age in THC, water, and scatter power The late decrease in THC and water correlates well with previous histologic studies showing both atrophy in well-vascularized lobular tissue and an increase in the fat-to-collagen ratio after menopause (15).This decrease is consistent with compositional analysis data showing lower blood and water content for fat versus glandular tissue (27).Scatter power also declines with age after age 50 years.Compared with fat, collagen and glandular tissue scatter light with a higher intensity and a steeper spectral dependence (28).Thus, smaller scatter powers are expected in fatty tissue.Exact knowledge of the weighted biologic contributions to optical tissue scattering is a matter of debate.

Correlation between Absorption and Scattering Information
Figure 7 displays a correlation plot of the measured water and lipid percentages versus scatter power.The lipid percentage is the mass density of lipids in the tissue referenced to the mass density of soybean oil.Note that neither the water nor the lipid percentage represents a strict mass or volume fraction for the tissue (ie, not all components are normalized to 100%).Lines represent fits of water (solid line) and lipids (dashed line) as functions of scatter power.Tissue scattering correlates well with the parenchymal composition measured via absorption.As scatter power increases, the lipid percentage decreases (R 2 ϭ Ϫ0.841, P Ͻ .001) and the water percentage increases (R 2 ϭ ϩ0.850, P Ͻ .001).The absorption spectra of water and lipids overlap, which can lead to errors in measuring these chromophores.Figure 7, however, clearly shows that both scattering and absorption provide complementary information regarding breast composition, indicating that such errors do not substantially distort our findings.

Tissue Metabolism and Menopause
The S t O 2 is the percentage fraction of arterial and venous Hb-O 2 that contributes to the tissue THC, and it should not be confused with arterial saturation, which is measured with pulse oximetry.Table 2 presents the measured S t O 2 in all subjects with a body-mass index less than 40.The average S t O 2 was approximately 76% for both the premenopausal subjects and those receiving HRT; their means were not significantly different.The S t O 2 values for the postmenopausal women not receiving HRT, however, were significantly higher, averaging 81.9% Ϯ 8.8 (P ϭ .062).
Decreases in S t O 2 are associated with increased metabolism (ie, extraction of O 2 from the blood).Tumors, for example, have been shown in vivo to have lower S t O 2 values than nontumor tissue (6,18,29).The postmenopausal increase in S t O 2 relative to the premenopausal women and those receiving HRT is plausible, because general tissue metabolism in the breast should decrease   after menopause as the tissue begins to atrophy.The similarity between S t O 2 values in premenopausal and HRT women suggests that their metabolisms are similar because of the effects of both endogenous and exogenous hormones.

DISCUSSION
Although it is difficult to validate noninvasive in vivo measurements directly, our initial results indicate that the sensitivity of NIR spectroscopy is a reasonable reflection of long-term hormone-controlled breast remodeling.Additional work in the field has demonstrated that short-term menstrual cycle changes are detectible with a similar NIR technique (30).Our results indirectly validate the general accuracy of NIR breast spectroscopy.This quantitative physiologic information is not obtainable with any other noninvasive radiologic method.
An important feature of NIR methods is the ability to characterize quantitatively the breast tissue of women regardless of age, hormonal status, or mammographic density.Increased mammographic density contributes to a 22% false-negative rate and a high false-positive rate (56.2% cumulative risk after 10 examinations) in women less than 50 years of age (31,32).A recent study found that routine initial mammography was not clinically advantageous for women less than 35 years of age (33).Furthermore, the use of HRT in postmenopausal women is known to increase mammographic density (34) and has been recently shown to impede the efficacy of mammographic screening (35,36).The sensitivity of NIR spectroscopy to known biologic processes suggests that optical methods may reveal important new information complementing that provided by conventional diagnostic techniques, particularly in the case of radiographically dense breast tissue.
In general, premenopausal breast tissue is more optically attenuating than postmenopausal breast tissue, in both absorption and scattering.It is important to note that this trend, particularly in scattering, is also true for mammographic density.Radiographically dense breast tissue is due to differing amounts of fat, collagen, epithelium, and water (37).Although x rays are inherently sensitive to dense electron structures in tissue, such as microcalcifications, NIR photons are naturally sensitive to tissue constituents.Thus, NIR spectroscopy has a unique potential for quantifying the elements of breast tissue that contribute to mammographic density.This observation may be important in screening, since it may identify breast tissue at physiologic risk for malignant transformation, and the technique can be performed easily in all women.
In addition to scattering contrast, we report age-dependent variations in THC, H 2 O, and lipids, as well as changes in S t O 2 for pre-and postmenopausal subjects.Knowledge of the "normal" values of these chromophores will help in evaluating the usefulness of optical methods for detecting and characterizing lesions in the breast.Several investigators have reported a two-to fourfold THC contrast between normal and tumor structures.In vivo tumor S t O 2 values are also typically lower than those of normal tissue (6,18,29).Thus, detailed studies of normal tissue are essential for determining the sensitivity required of optical instrumentation for detecting lesions in women of varying age and hormonal status.Furthermore, as baseline levels are characterized, data on an individual's absorption and scattering variations could provide important insight into disease appearance and progression.When applied to patients receiving chemotherapy and/or HRT, this information could also be used to generate feedback that would permit customized treatment planning based on individual physiologic response.

Figure 1 .
Figure 1.Schematic drawing of FDPM instrument, hand-held probe, and measurement map.The components of the instrument are the diode lasers, avalanche photodiode (APD), network analyzer, steady-state current source (DC), bias network, and optical switch.Measurements were performed with a hand-held probe (close-up at bottom of Figure) that contained the APD and a source optical fiber.The measurement location was in the center of the left upper outer quadrant of the breast, as indicated.
represent a fit of the scattering coefficients to the equation Ј s ϭ A ϪSP .

Figure 2 .
Figure 2. Measured absorption spectra for a 32-year-old premenopausal subject (s, solid line) and a 54-year-old postmenopausal subject (OE, dashed line).Points represent the average of several measurements, and lines represent a least-squares fit (extrapolated to all wavelengths) based on the assumption that breast absorption is due only to Hb-R, Hb-O 2 , H 2 O, and lipids.Error bars show the SDs of repeated measurements.Concentrations of chromophores from this measurement are listed in Table1.
Measurements are given as individual means Ϯ SDs for the two subjects represented in Figures2 and 3. * Measured concentration relative to pure water.†Measured mass density relative to pure lipid.

Figure 4 .
Figure 4. Plot of the THC versus subject age.Premenopausal subjects (Ͻ50 years, n ϭ 15) show considerable variation with age because of intersubject variations, such as menstrual cycle variations and differences in overall hormone production.Postmenopausal subjects (Ͼ50 years, n ϭ 13) appear to show a decrease in THC.Error bars show the SDs of repeated measurements.

Figure 5 .
Figure 5. Plot of the measured tissue water percentage versus subject age.Premenopausal subjects show considerable variation over age because of intersubject variations such as menstrual cycle variations and differences in overall hormone production.Postmenopausal and perimenopausal subjects appear to show a decrease in water content.The percentages are water concentrations normalized to the concentration of pure water.Error bars show the SDs of repeated measurements.

Figure 6 .
Figure 6.Plot of the measured optical scatter power (in arbitrary units) versus subject age.Optical scatter power refers to the magnitude of the slope of the scattering versus wavelength curve.Premenopausal subjects show considerably higher scatter power than postmenopausal subjects.This general pattern is the same for mammographic density.Error bars show the SDs of repeated measurements.

Figure 7 .
Figure 7. Correlation plots for water percentage (ᮀ, solid line) and lipid percentage (OE, dashed line) versus scatter power.Percentages are referenced to the concentration of pure water and the mass density of pure fat.Scatter power correlates with water (R 2 ϭ ϩ0.850, P Ͻ .001) and correlates negatively with lipids (R 2 ϭ Ϫ0.841, P Ͻ .001).Error bars show the SDs of repeated measurements.