BJR
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
First published online May 30, 2007
British Journal of Radiology (2007) 80, 545-556
© 2007 British Institute of Radiology
doi: 10.1259/bjr/26858614
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Blackmore, K M
Right arrow Articles by Lilge, L
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Blackmore, K M
Right arrow Articles by Lilge, L

Full paper

Assessing breast tissue density by transillumination breast spectroscopy (TIBS): an intermediate indicator of cancer risk

K M Blackmore, MSc 1 J A Knight, PhD 2 R Jong, MD 3 and L Lilge, PhD 1,4

1 Ontario Cancer Institute, University Health Network, Toronto, Ontario, Canada M5G 2M9, 2 Samuel Lunenfeld Research Institute, Mount Sinai Hospital, Toronto, Ontario, Canada M5G 1X5, 3 Sunnybrook and Women's College Health Science Centre, Toronto, Ontario, Canada M4N 3M5, 4 Department of Medical Biophysics, University of Toronto, Toronto, Ontario, Canada M5G 2M9

Correspondence: Dr Lothar Lilge, Biophysics and Bioimaging, Ontario Cancer Institute, 610 University Avenue, Toronto, Ontario M5G 2M9, Canada. E-mail: llilge{at}uhnres.utoronto.ca


   Abstract
Top
 Abstract
Introduction
 Methods and materials
 Results
 Discussion
 Summary and conclusions
 References
 
Risk assessment by parenchymal density pattern, a strong physical indicator of future breast cancer risk, is available with the onset of mammographic screening programmes. However, due to the use of ionizing radiation, mammography is not recommended for use in younger women, thereby rendering risk assessment unattainable at an earlier age. Visible and near infrared light was used on 292 women with radiologically normal mammograms to determine whether transillumination breast spectroscopy (TIBS) can identify women with a high parenchymal density pattern as an intermediate indicator of breast cancer risk. Principal component analysis (PCA) was used to reduce the spectral data and generate density scores for each woman. To assess the accuracy of TIBS, logistic regression was used to calculate crude and adjusted odds ratios (OR) and 95% confidence intervals (CI) for each score. Receiver operator characteristic (ROC) curves and area under the curve (AUC) were also calculated for the crude and adjusted logistic models. Optical information relating to tissue chromophores, such as water, lipid and haemoglobin content, was sufficient to identify women with high parenchymal density. The resulting AUC for the final and most parsimonious multivariate logistic model was 0.922 (95% CI 0.878–0.967). TIBS provides information correlating to high parenchymal density and is a promising tool for risk assessment, particularly for younger women.


   Introduction
Top
 Abstract
 Introduction
Methods and materials
 Results
 Discussion
 Summary and conclusions
 References
 
Breast cancer is the most commonly occurring cancer in women [1, 2]. While screening programmes have resulted in decreased mortality rates as a result of the detection of early stage cancers, the overall incidence of breast cancer is still increasing [1–3]. Consequently, within the field of preventive oncology, intervention (i.e. risk-reducing) strategies are being considered with the primary goal of decreasing breast cancer incidence rates and thereby maintaining the health of the female population at a high level [4–8]. However, to attain such a goal, accurate and generalizable techniques are required to identify all women at greatest risk who would benefit from risk-reducing intervention programmes [4–8]. Therefore, the most useful risk assessment technique would also rely on an identifier demonstrating a large relative risk and be applicable to all women, unlike, for example, genetic markers such as BRCA1 and BRCA2 mutations, which occur only in a very small proportion of the female population [9].

Research into breast cancer aetiology has shown that the development of the disease is a slow process following initial transformation of the breast tissue, and that events early in life may be important in establishing future breast cancer risk [10, 11]. Moreover, initiating risk assessment in young women may have the added benefit that potentially less drastic intervention strategies (i.e. diet, exercise and lifestyle changes) acting over a longer period of time could be sufficient to achieve an adequate reduction in risk [12, 13].

One physical method of evaluating breast tissue risk that is relatively well established is the X-ray dense tissue content of the breast as obtained by standard mammography. Parenchymal density patterns assessed using quantitative methods have been consistently related to breast cancer risk [14, 15]. Several studies have shown that women with dense tissue occupying 75% or more of the total breast area compared with those with low density (< 25%) are four to six times more likely to develop breast cancer in the next decade [16–26]. Given that the extent of mammographic density is influenced by hormonal exposure (such as during the menstrual cycle and pregnancy), Boyd et al [14] have argued that mammographic density is a marker of susceptibility to breast cancer consistent with the concept of rate of breast tissue ageing introduced by Pike et al [10]. The main drawback to the use of mammography as an indicator of breast cancer risk is the required exposure to ionizing radiation, and hence there are concerns regarding its use in young women for regular screening (less than 40 years of age in the USA and younger than 50 years in Canada and the UK) [27, 28]. The technique also requires compression of the breast, which causes discomfort in some women.

Alternative methods that have been proposed for assessing breast tissue density include MRI and ultrasound. Preliminary work has demonstrated a correlation between relative water content derived with MRI and mammographic density [29, 30] and the extent of echogenic areas on ultrasound with mammographic parameters [31, 32]. Although MRI does not employ ionizing radiation, the technique is expensive and in high demand for other clinical applications, and being in an enclosed space can cause discomfort in some individuals. A more recently proposed method of assessing the state of breast tissue is to directly measure biomarkers in nipple aspirate fluid. There is some evidence that epithelial hyperplasia and atypical hyperplasia detected in this fluid are associated with increased breast cancer risk [33]. However, it is an invasive procedure, and nipple aspirate fluid was unobtainable in 40% of the cohort in whom breast cancer risk was assessed [33].

Near infrared (NIR) transillumination breast spectroscopy (TIBS) is a non-imaging, non-invasive technique that provides information about bulk tissue properties through the spectral dependency of photons that have passed through the breast tissue [34–36]. In contrast to mammography, TIBS uses non-ionizing visible and NIR light and can therefore be used on younger women, theoretically beginning at puberty. Plate compression of the breast tissue is not necessary, and each measurement takes no more than a few seconds. Additionally, no special training is required for its use, and the device and examination can be made available at costs at least an order of magnitude lower than mammography. Breast tissue is a highly scattering medium with relatively low absorption in the red and NIR wavelength ranges, permitting sufficient light penetration to detect signals through up to 7 cm of tissue, while maintaining the incidence power below government guidelines for light exposure to skin [37].

Multiple studies using NIR technologies to examine breast tissue have been published, and the findings for healthy breast tissue composition are fairly consistent [38–47]. In the NIR spectrum, the main absorbers of photons (i.e. chromophores) are water, lipids, collagen and haemoglobins (oxy- and deoxy-). In breast tissue, fibroglandular tissue results in increased water and concomitant decreased lipid-associated absorption. Stromal tissue, which contributes to high parenchymal density patterns, further suggests high collagen content. Larger total haemoglobin content and a trend towards lower oxygen saturation are also anticipated in high-density tissue compared with fatty tissue because of increased tissue vascularization and cellular proliferation, and thus increased metabolism. Variations in tissue scattering particle density (i.e. cells and intracellular organelles, collagen) affecting optical path length also have an impact on the wavelength-dependent light attenuation by the tissue. Specifically, dense cellular tissue volumes, such as ductal and fibroglandular tissue, will result in increased scattering and, conversely, in increased attenuation.

The study presented here is an extension of earlier publications by our group [34–36] and includes an analysis of the complete spectral data set obtained from a cross-sectional study comprising 292 pre- and post-menopausal women without radiologically suspicious lesions. The primary aim of the study was to evaluate the feasibility of identifying all women with very high breast tissue density (≥ 75% dense tissue) in vivo using TIBS. The underlying hypothesis is that TIBS provides information consistent with conventional mammography in identifying breast tissue density and hence is a potential intermediate indicator of breast cancer risk. Mammographic density was selected as the gold standard because it is a physical assessment of breast tissue applicable to the entire female population over 40 years of age, and carries the highest odds ratio (OR; 4–6 with respect to other known non-genetic-based risk factors.


   Methods and materials
Top
 Abstract
 Introduction
 Methods and materials
Results
 Discussion
 Summary and conclusions
 References
 
Study population
Three hundred participants were recruited among women having a standard X-ray mammography at the Marvelle Koffler Breast Centre in Mount Sinai Hospital, Toronto, Ontario (Table 1Go). Eligible women had a screening mammogram (233 film and 67 digital) within 12 months prior to recruitment (between 1 March 2000 and 30 September 2004) showing no radiologically suspicious lesions. Women did not have previous surgery to the breast tissue, including reduction or augmentation. Women displaying variations in parenchymal density between breasts were excluded as their correlation with cancer risk is weaker. Information concerning participants' age, menopausal status (pre-menopausal vs post-menopausal), height and weight were collected by means of a self-administered questionnaire. Participants were reimbursed for travel expenses only. Post-menopausal status was defined as having had no menstrual period for at least 12 months. Height and weight were used to calculate body mass index (BMI) defined as weight in kilograms divided by the square of the height in metres. This study was approved by the Institutional Review Boards (IRBs) of the University of Toronto, Mount Sinai Hospital and the University Health Network.


View this table:
[in this window]
[in a new window]

  		Table 1. Mammographic, demographic, reproductive and anthropometric information of participants included in the data analysis

 
Quantification of parenchymal density from mammograms
All usable mammograms (300) were classified on an ordinal scale by an expert radiologist (RJ) into low (< 25%), medium (25% to < 75%) or high (≥ 75% dense tissue area) density categories for analysis using ordinal data. Boyd et al [24] and Jong et al [48] have demonstrated a high level of inter-radiologist agreement for density classification (intraclass correlation coefficients of 0.94 and 0.89, respectively), indicating that the validity of the study was not negatively affected by using one radiologist. Furthermore, the radiologist used in the present study was very experienced and has participated in previous studies requiring classification of mammograms on an ordinal scale [24, 26, 48].

Optical set-up and procedure
All optical measurements were collected prior to quantification of the participant's tissue density class. The instrumentation used to gather transillumination spectra has been described in detail previously [34–36]. A 50 W halogen lamp served as the broadband light source. Ultraviolet, part of the visible spectrum and mid-infrared radiation were eliminated using a cut-on ({lambda} > 550 nm) and a heat rejection filter, respectively. The remaining light was coupled into a 5 mm diameter liquid light guide (Fibre Guide, Bridgeport, CT) placed in contact with the skin on top of the breast. A total power of 0.25 W, covering the 550–1300 nm bandwidth, was delivered to the skin. Transmitted light was collected via a 7 mm diameter optical fibre bundle (140 Si/Si fibres, 200 µm core diameter, numerical aperture: 0.36; P & P Optica, Kitchener, Canada). The source and detector fibre bundles (optodes) were held coaxially, pointing towards each other, by a calliper attached to the resting platform, also providing the interoptode distance. The source fibre was placed against the skin on the top surface of the breast with minimal compression. Wavelength-dependent detection in the visible and NIR was achieved using a spectrophotometer (Kaiser, CA) with holographic transillumination grating (15.7 rules mm–1 blazed at 850 nm) and a two-dimensional cryogenically cooled silicon charge-coupled device (CCD; Photometrics, NJ) at a spectral resolution of better than 3 nm between 625 and 1060 nm, achieved by using a 0.5 mm entrance slit to the spectrophotometer. Total data acquisition times for all required measurements ranged from 160 to 200 s. Considering typical tissue optical properties and tissue thickness ranging from 2.5 to 7 cm, ovoid-shaped tissue volumes of 12–54 cm3 were interrogated per spectral measurement. Hospital Grade Canada Standards Association (CSA) certification and Health Canada Investigational New Device Class II approval was obtained.

All measurements were taken in the dark, with the participant seated comfortably in an upright position and the breast resting on the support platform. A total of eight measurements in craniocaudal projection were taken per individual, four per breast (centre, midline close to the pectoral muscle; medial, 2 cm from the inner edge; distal, 2 cm behind the nipple; and lateral, 2 cm from the outer edge), resulting in optical interrogation of different anatomical regions of the breast [34–36]. Temporal and spatial reproducibility of the optical measurement is good, as addressed previously [34, 35].

Preparation of spectra for data analysis
Spectra were corrected for daily variations in the wavelength-dependent signal transfer function of the optical system (< 1% day by day) and the thickness of the interrogated tissue, given by the interoptode distance. To correct for the signal transfer function, spectra were referenced to a transmission standard made of 1 cm thick ultrahigh-density polyurethane (Gigahertz Optics, Munich, Germany). The optical properties (OD cm–1 ~ 1.8–2.3 over the wavelength range of interest) of the polyurethane standard were measured in a separate experiment using an integrating sphere diffuse reflectance set-up [49]. Hence, spectra used in further data processing are independent of the instrument and the interoptode distance, and are expressed in units of optical density per centimetre (OD cm–1), calculated using the negative log of the raw data spectrum, the reference spectrum of the polyurethane standard and the interoptode distance [34–36].

Principal component analysis (PCA)
To establish a correlation between the obtained transillumination spectra, here considered vectors (OD cm–1 vs wavelengths from 625 to 1060 nm) and a target (breast tissue density on an ordinal scale), PCA was used [50, 51]. PCA is a commonly used data analysis technique in the field of chemometrics and for spectroscopic analysis in medical applications [51, 52]. For PCA implementation, spectra from all usable high, medium and low tissue density women (Table 1Go; 300 women x 8 measurements, n = 2400) were employed. PCA was executed using Matlab 12.1 (The MathWorks Inc., MI). All other statistical analyses were carried out using SPSS (Statistical Packages for the Social Sciences, SPSS Inc., Chicago, IL), version 11.0 and SAS (Statistical Analysis Systems; SAS Institutes Inc., USA), version 9.1.

Prior to PCA implementation, the mean spectrum (Smacr;) of all spectra was calculated and subtracted from each individual spectrum (Si), resulting in Si = Si–Smacr;, in order to derive mutually orthogonal principal components, and hence independent principal component scores. PCA derives a minimum number of representative spectra, the principal components (pn), accounting for the majority of the variance seen in the entire mean-centred spectral data set. All principal component spectra contain a varying amount of metabolic and structural information from the interrogated tissue (tissue scattering by cellular or structural components, lipid and water content, deoxy- and oxyhaemoglobin content) [34–36]. Once the principal components were derived, scalar coefficients (i.e. scores, tin) were assigned to each individual mean-centred spectrum (Si) measured at each position for each woman, indicating the contribution of each principal component, and hence the optically interrogated chromophores, to that spectrum. In the present study, each individual spectrum is a linear combination of four principal component spectra (pn) multiplied by the respective scalar coefficient or score (tin), such that: Si = ti1p1 + ti2p2 + ti3p3–ti4p4 + isin, where isin represents the residual error. (Note: while t1 to t3 showed an inverse relationship with mammographic density, t4 showed a direct relationship.)

Bilateral symmetry in the spectra at corresponding quadrants was demonstrated previously [34, 35] in women without variations in parenchymal pattern. Consequently, the resulting principal component scores (tin), herein referred to as "density" scores, were averaged over all measurement positions on both breasts for each woman prior to subsequent statistical analysis. Averaging scores post PCA results in a global optical assessment of the tissue, similar to mammographic density class assignment, while permitting assessment of intraperson variability. Descriptive statistics (median ± interquartile range (IQR)) were determined for the derived density scores (tin) for the high (≥ 75%) and combined low and medium density categories (< 75%) for all women and by menopausal status (pre-menopausal vs post-menopausal). The analysis of high vs combined low and medium density best approximates the clinical decision-making process, requiring high sensitivity and specificity for the identification of only women at risk from among the entire female population. As the scores were generally not normally distributed, score differences between < 75% and ≥ 75% tissue densities were tested by non-parametric methods using the Mann–Whitney U-test.

For all analyses, p-values ≤ 0.05 were considered to be statistically significant. As eight women were missing information on height and weight, their BMI could not be calculated. For consistency, these women were excluded from further statistical analyses, and the data presented here are for n = 292 women with complete spectral and demographic information.

Logistic regression using PCA scores
To measure the association between the derived density scores (tin) and breast tissue density, univariate and multivariate logistic regression was used to estimate both crude (i.e. unadjusted) and adjusted odds ratios (OR) and 95% confidence intervals (CI), respectively, with mammographic density as the outcome. For the analysis presented here, breast density was treated as a dichotomous variable, by comparing women with very dense breasts (≥ 75% dense tissue) with those with < 75% dense tissue. All models were fitted with the density scores treated on a continuous scale.

To identify potentially confounding variables for inclusion in the multivariate models, univariate logistic regression analysis was also used to estimate crude OR and 95% CI for age, BMI and menopausal status with high mammographic density (≥ 75% dense tissue) as the outcome. For this analysis, age and BMI were treated on a continuous scale, while menopausal status was treated as a dichotomous variable (pre- vs post-menopausal status). The strength of the association of each density score with mammographic density as a function of menopausal status was assessed by creating an interaction term between the variable menopausal status and each density score and including these terms in the logistic models examined. For all logistic models, covariates with p-values ≤ 0.05 were considered to be statistically significant. Furthermore, for both the univariate and the multivariate logistic models, the calculated individual probabilities of having density ≥ 75% were used to derive receiver operator characteristic (ROC) curves and area under the curve (AUC) in order to evaluate how well each model predicts the outcome.

Correlation analysis was also executed between the scores (tin), age and BMI to examine the relationship among the dependent variables.


   Results
Top
 Abstract
 Introduction
 Methods and materials
 Results
Discussion
 Summary and conclusions
 References
 
Study population
Table 1Go lists the mammographic, demographic, reproductive and anthropometric information for all eligible women used in the data analysis (n = 292). The overall study population proportions are comparable to those seen in the Canadian National Breast Screening Study (CNBSS age range 40–59 years) [24].

PCA of tissue density categories
The first four principal components resulting from PCA (Figure 1Go) capture 96.41%, 1.84%, 1.06% and 0.35% of the variance, respectively, yielding a combined total of 99.66% for the data set comprising spectra from all three tissue density classes. Principal component p1 carries information on the overall light attenuation of the interrogated tissue due to differential optical path length resulting from light scattering and losses at the boundary, while components p2 to p4 contain information about the water, lipid and the oxy- and deoxyhaemoglobin content of the tissue [34–36].


Figure 1
View larger version (15K):
[in this window]
[in a new window]

  		Figure 1. Principal component spectrap1 to p4 (a to d) of system transfer and thickness-corrected spectra. (a) p1 showing average light attenuation for < 75% (grey) and ≥ 75% (black) dense tissue; (b–d) p2 to p4, respectively, showing contributions from lipids, water and haemoglobins (oxy and deoxy). Areas above the dashed line represent spectral regions of interest for low density tissue, while areas below the dashed line represent spectral regions of interest for high density tissue.

 
Box plots of the derived density scores t1 to t4 according to risk classification (low < 75% vs high ≥ 75% dense tissue) are shown in Figure 2Go for all women (n = 292). The median values and interquartile range (IQR) for each of the first four PCA scores t1 to t4 for women with very dense tissue (≥ 75%) vs women with < 75% dense tissue are shown in Table 2Go for all women and by menopausal status. A Mann–Whitney U-test demonstrated that t1, t3 and t4 were each significantly different between < 75% and ≥ 75% tissue densities among all women and post-menopausal women, while only t1 and t3 retained significance among the pre-menopausal group. Density score t2 was not significantly different between high and low risk tissue density for all women, or when examined by menopausal status (Table 2Go).


Figure 2
View larger version (16K):
[in this window]
[in a new window]

  		Figure 2. (a–d) Box plots of derived scores t1 to t4 by tissue density classification (< 75% vs ≥ 75 %) for all women (n = 292).

 

View this table:
[in this window]
[in a new window]

  		Table 2. Median, interquartile range(IQR) and results of Mann–Whitney U-test for derived scores t1 to t4 for all women (n = 292) and by menopausal status

 
Logistic regression using density scores
Table 3Go presents the unadjusted OR and accompanying 95% CI for the OR for each density score t1 to t4, age, BMI and menopausal status (pre- vs post-menopausal). The OR and accompanying 95% CI for the OR are expressed over the IQR of the respective density score tin, age or BMI for n = 292 and represent the odds of high breast tissue density (≥ 75%) vs the odds of < 75% dense tissue for an increase in the respective density score, age or BMI over the IQR (i.e. from the first quartile (Q1) to the third quartile (Q3)).


View this table:
[in this window]
[in a new window]

  		Table 3. Results of univariate logistic regression analysis for each density scoret1 to t4, age, body mass index (BMI) and menopausal status (n = 292)

 
The odds of having ≥ 75% tissue density were significantly and inversely associated with the values of t1, t3 and t4 while, for density score t2, the OR was not significant. BMI was strongly and inversely associated with high breast tissue density, while the odds of ≥ 75% dense tissue among pre-menopausal women was 2.25 times the odds among post-menopausal women. The association between age and high breast tissue density was not significant. Among the univariate logistic models, the calculated AUC was highest for density score t3.

Correlation analysis demonstrated a strong positive association between t1 and BMI (r = 0.663, p < 0.001), and between t1, t2 and t4 and age (r = 0.248, r = 0.298 and r = 0.391, respectively; all at p < 0.001).

Table 4Go shows OR and accompanying 95% CI for the OR for each density score t1 to t4 adjusted for age, BMI and menopausal status (models I to IV). In the adjusted models, density score t1 was no longer significantly associated with high breast tissue density. As in previous analyses, t2 was not significantly associated with mammographic density, while the effect of t3 did not change (Table 3Go vs Table 4Go). However, unlike the other density scores, the strength of the inverse association of density score t4 with high breast tissue density varied according to menopausal status (model IV, Table 4Go). Figure 3Go displays the estimated odds of high breast tissue density as a function of t4 separately for pre- and post-menopausal women while adjusting for mean age (50.9 years) and mean BMI (25.8). Although the odds of high breast tissue density were higher among pre-menopausal women, this varied only slightly across values of t4, while among post-menopausal women, the odds showed a strong inverse association as t4 increased.


View this table:
[in this window]
[in a new window]

  		Table 4. Results of multivariate logistic regression analysis for each density scoret1 to t4 adjusted for age, body mass index (BMI) and menopausal status (n = 292)

 

Figure 3
View larger version (9K):
[in this window]
[in a new window]

  		Figure 3. Odds of high breast tissue density(≥75%) as a function of mean centred density score t4 for pre-menopausal (dashed line) vs post-menopausal women (solid line). Note: Odds defined as probability of high density (p≥75%) divided by 1–(p≥75%); odds adjusted for mean age (50.9 years) and mean body mass index (BMI; 25.8).

 
Table 5Go displays the OR and 95% CI for a model including density scores t3 and t4, menopausal status, BMI and the interaction between t4 and menopausal status. Because t1, t2 and age were not significantly associated with high breast tissue density, either independently (Table 3Go) or once adjusted for other covariates (Table 4Go), they were excluded from further analyses in order to attain the most parsimonious model. As can be seen from Table 5Go, the inclusion of both t3 and t4 in a single model resulted in a decrease in the OR (i.e. stronger) for t3 for all women and for t4 among post-menopausal women, and both remained significantly and inversely associated with high breast tissue density. BMI also remained strongly and inversely associated with the outcome. The AUC for the final model was 0.922 (95% CI 0.876–0.968) (Figure 4Go).


View this table:
[in this window]
[in a new window]

  		Table 5. Final logistic regression model including density scorest3 and t4, menopausal status, body mass index (BMI) and the interaction between density score t4 and menopausal status

 

Figure 4
View larger version (12K):
[in this window]
[in a new window]

  		Figure 4. Receiver operating characteristic(ROC) curve for final multivariate logistic regression model for all women (n = 292).

 

   Discussion
Top
 Abstract
 Introduction
 Methods and materials
 Results
 Discussion
Summary and conclusions
 References
 
This study extends earlier publications by our group correlating TIBS with density classification on an ordinal scale [35, 36] to include the complete spectral data set of 292 pre- and post-menopausal women with radiologically normal breast tissue. While previous studies covered only high (≥ 75%) vs low (< 25%) mammographic tissue density, the main purpose of the analysis presented here was to determine the ability of TIBS to discriminate women with ≥ 75% dense tissue from the remainder of the total study population in order to establish the utility of TIBS as a potential assessment tool to identify those women at greatest risk for the development of breast cancer and who would benefit most from risk-reducing interventions. Because risk assessment is not commonly available to women until midlife (i.e. 40 years or older), valuable years are lost for risk reduction interventions to exert their influence. Therefore, the ability to initiate risk assessment using a non-invasive pre-screening technique such as TIBS at a younger age (i.e. twenties to early thirties or even during puberty), when breast tissue is more vulnerable to hormone exposure and carcinogenic insults [10, 11], is very desirable.

Characterization of tissue density by TIBS
In previous work by our group [34–36], PCA was executed on non-mean-centred spectra, and t1 was negative for all tissue density classes, thus representing the average attenuation for the tissue. In the current analysis using mean-centred spectra, information about the average attenuation is removed, thereby augmenting differences between tissue density groups (i.e. negative vs positive scores).

Similar to previous work, the obtained principal component spectra identify some metabolic and structural properties of breast tissue that are important for tissue density classification, including absorption by water, lipids and the haemoglobins (oxy- and deoxy-), as well as changes in photon path length due to light scattering and boundary losses [34–36]. In the analysis presented here, the sign and the magnitude of the density scores t1, t3 and t4, associated with principal component spectra p1, p3 and p4, are important in differentiating spectra as originating from women with high tissue densities (≥ 75%) vs those with < 75% dense tissue. Alternatively, as the spectrum of p2 is flat between 62 nm and 875 nm (the primary region of haemoglobin absorption), and the only notable feature is a reduced lipid peak at 925 nm (Figure 1bGo), the scores t2 associated with principal component p2 were not significantly different between women with < 75% vs women with ≥ 75% dense tissue.

Principal component p1 carries information on the light scattering properties of the interrogated tissue, which varies in part with wavelength [34–36]. Scattering increases the path length travelled by light, resulting in a greater probability of absorption and thus a smaller probability of traversing the tissue. Additionally, it also increases the losses due to diffuse reflection at the tissue surface. Fibroglandular tissue has a larger scattering coefficient than fatty tissue because of increased cellular content [39, 40] and structural support tissues, such as the collagen matrix [53, 54].

In the analysis presented here, the median value for t1 was positive for < 75% dense tissue and negative for women with ≥ 75% tissue densities (Table 2Go). As depicted in Figure 1a, aGo smaller (i.e. negative) t1 suggests more attenuation relative to a larger (i.e. positive) density score, as would be expected for dense tissue. This finding was confirmed by other groups who demonstrated a similar relationship between increased tissue density and increased light scattering and hence attenuation [39–47].

Adipose tissue predominantly characterizes lower density tissue [29] by its absorption maximum at 925 nm, while high water content with absorption at 970–975 nm was shown to correlate with higher density tissue [29, 30]. Larger total haemoglobin content and a trend towards lower oxygen saturation (indicated by more deoxy- vs oxyhaemoglobin) are anticipated in higher density tissue compared with fatty tissue on account of increased cellular proliferation and metabolism [43] and tissue vascularization [47]. These spectral features are also identifiable in the principal component spectra of p3 and p4 (Figure 1cGo,dGo), which contain information about the water and lipid content of the tissue, as well as the deoxy- and oxyhaemoglobin content of the tissue [34–36].

In ≥ 75% dense tissue, smaller (i.e. negative) t3 and t4 (Table 2Go) indicate greater contributions from water at 970–975 nm (p3 and p4) and smaller contributions between 775 and 875 nm (p4) (Figure 1cGo,dGo). Underlying contributions from deoxyhaemoglobin between 625 and 725 nm (p3) (Figure 1cGo) and from oxyhaemoglobin from 750 to 900 nm (p4) are also present (Figure 1dGo). Conversely, in < 75% density tissue, larger t3 and t4 (Table 2Go) suggest contributions from lipids at 925 nm (p3 and p4) and smaller contributions at 825 nm (p3) (Figure 1cGo); contributions from deoxyhaemoglobin between 625 and 725 nm (p4) (Figure 1dGo) are also evident. The combined contribution (OD cm–1) of both oxy- and deoxyhaemoglobin (i.e. THC) was greater in ≥ 75% density tissue compared with < 75% dense tissue [34–36]. Furthermore, among high density tissue, the relative contribution of deoxyhaemoglobin to oxyhaemoglobin was larger [34–36].

The information captured in each principal component spectrum p3 to p4 with respect to tissue density is, as anticipated, according to the known anatomical and physiological properties of healthy, potentially at risk, breast tissue [34–36, 38–47].

Discrimination of high breast tissue density by TIBS
The observed inverse associations for each density score t1, t3 and t4 in the logistic regression models with high breast tissue density are consistent with the spectral features captured by each principal component spectrum (GoGoTables 3–5Go). As in other analyses (see Table 2Go), t2 was not associated with high breast tissue density. In the crude models (Table 3Go), density score t3 carried the strongest inverse association and highest level of accuracy, as indicated by the individual OR and AUC, followed by t1, then t4. The finding that density score t1, a measure of overall light attenuation caused by the same structural components contributing to the parenchymal density pattern of the breast (i.e. relative amounts of connective and epithelial tissue and fat), resulted in lower AUC compared with t3 suggests that TIBS through t3 provides additional information to X-ray derived mammographic density. Specifically, information about the water to lipid ratio and deoxyhaemoglobin content of the tissue captured by p3 is more accurate in discriminating between < 75% and ≥ 75% density tissue.

Similar to previous studies, BMI showed a significant inverse association with high mammographic density, and the risk of high density parenchymal patterns was higher among pre-menopausal women (Table 3Go) [55, 56]. However, in contrast to other studies, age at TIBS was not significantly associated with high mammographic density. BMI and age also showed significant positive correlations with some of the density scores, namely t1 (BMI and age) and t2 (age only) and t4 (age only). There was also some suggestion that discrimination of high mammographic density using density scores differed between pre- and post-menopausal women (Table 2Go). Based on these findings, all three demographic variables were retained in the multivariate logistic models as potential confounders in the association between the derived scores and mammographic density.

After adjusting for age, BMI and menopausal status (Table 4Go), the association of t1 with high breast tissue density was no longer significant. This reduced significance is most probably explained by the inclusion of BMI in the same model as t1 as both variables were strongly and positively correlated (r = 0.663). This is expected as a higher BMI means more adipose tissue overall, more fatty replacement in the breast [55], and a larger t1 (i.e. more positive) reflects reduced scattering and less attenuation due to smaller amounts of connective (collagen) and epithelial tissue relative to fatty tissue [34–36]. However, the fact that BMI retained significance after adjustment for t1 indicates that, of the two measures, BMI is the more relevant independent predictor of high breast tissue density in this study. Consequently, BMI was retained in further logistic models in lieu of t1. In terms of developing a predictive model, BMI is an easily and readily obtainable demographic variable.

In contrast to t1, density score t3 remained significantly and independently associated with high mammographic density among all women even after adjustment for BMI and other demographic covariates. The OR associated with density score t4 also retained significance after adjustment for other indicators of mammographic density; however, this association was only significant among post-menopausal women. This latter finding suggests that the additional optical information captured by component spectrum p4 relative to p3, namely contributions from deoxyhaemoglobin between 625 and 725 nm in < 75% dense tissue, and from oxyhaemoglobin and water between 775 and 875 nm and 750 and 900 nm in ≥ 75% dense tissue, respectively, are necessary to discriminate high from low density tissue among post-menopausal women, but not among pre-menopausal women. The use of menopause-specific PCA models should be explored; however, because of the limited number of women, and hence spectra, in each group in the current study, menopause-specific PCA was not feasible here.

In the final multivariate model, which included only those variables significantly associated with high mammographic density in both the crude and the adjusted analyses, the inclusion of both t3 and t4 in a single model resulted in a stronger inverse association of each score with mammographic density (post-menopausal women only for t4) (Table 4Go vs Table 5Go). A change in the OR associated with each score is likely, as all derived density scores were originally based on perpendicular mutually exclusive principal components and, even though the scores were independent of one another (i.e. not correlated), each optical spectrum is a linear combination of the mean spectrum of all four components p1 to p4 and all four density scores t1 to t4. However, the fact that, of the four density scores, only t3 and t4 (post-menopausal women only) remained significantly associated with high tissue density indicates that optical information relating to water, lipid and haemoglobin content is sufficient to identify women with high parenchymal density, after adjusting for BMI and menopausal status.

The high AUC associated with the final multivariate logistic model suggests that TIBS is a potential pre-screening tool for preventive oncology. However, as all 292 women were considered in obtaining the AUC, the authors acknowledge that the value of 0.922 is likely to be an overestimate of the technique's accuracy in identifying women with high breast tissue density (≥ 75%). Future work will focus on validating the predictive ability of TIBS using a group of women soon to be enrolled in a recently funded study for which both mammographic density and TIBS measurements will be available.


   Summary and conclusions
Top
 Abstract
 Introduction
 Methods and materials
 Results
 Discussion
 Summary and conclusions
References
 
This study demonstrated that in vivo TIBS is a physical method of assessing breast tissue composition and is thus a promising tool for breast cancer risk assessment. Specifically, the derived component scores (tin) are analogous to "density scores" assigned to each woman, which identify her breast tissue as either low or high density and hence, by proxy, as being at low or high risk of future breast cancer, respectively.

Unlike X-ray mammography, TIBS does not exploit the atomic composition of the breast tissue, but rather some of its biomolecular markers, and does not carry a dose penalty. Specifically, contributions from water, lipids and haemoglobins were sufficient to discriminate between women with low and very high tissue density. This suggests that TIBS provides complementary information to tissue X-ray attenuation. An added advantage of TIBS over current imaging modalities is the fact that the results are derived from mathematical models; hence highly trained personnel are not required for image interpretation or assessment. The inherent safety (i.e. use of non-ionizing visible and NIR light) and comfort of this method will also permit risk assessment in women outside the recommended age range for X-ray mammography (< 40 years of age), in whom risk assessment is currently not obtainable, as well as in those individuals who are non-compliant with standard screening.

Future investigation of menopause-specific PCA models (pre- vs post-menopausal) using a larger sample of women is warranted. Validation of the predictive ability of the model presented here on a new group of women will also be carried out to assess the ability of TIBS to discriminate high tissue density. Work is currently under way to study directly the relationship between TIBS and breast cancer incidence.


   Acknowledgments
 
The authors wish to express their gratitude to all participants in this study. The authors would also like to thank Gina Lockwood for her statistical advice during the preparation of this paper. This work was funded by CDMRP (Congress Directed Medical Research Plan), under DAMD 17-00-1-0393, and by CIPI (Canadian Institute of Photonics Innovation).

Received for publication May 23, 2006. Revision received September 7, 2006. Accepted for publication October 17, 2006.


   References
Top
 Abstract
 Introduction
 Methods and materials
 Results
 Discussion
 Summary and conclusions
 References
 

  1. National Cancer Institute of Canada. Canadian cancer statistics 2001. Toronto, Canada: National Cancer Institute of Canada
  2. National Cancer Institute of Canada. Canadian cancer statistics 2000. Toronto, Canada: National Cancer Institute of Canada
  3. Paci E, Duffy S, Rosseli di Truco M. Mammographic screening: from the scientific evidence to practice. In: The Oxford textbook of oncology. Oxford: Oxford University Press, 2000
  4. Salih AK, Fentiman IS. Breast cancer prevention: present and future. Cancer Treat Rev 2001;27:261–73.[CrossRef][Medline]
  5. Fabian CJ, Kilmer BF. Breast cancer chemoprevention: current challenges and a look toward the future. Clin Breast Cancer 2002;3:113–24.[Medline]
  6. Sakorafas GH, Krespis E, Pavlakis G. Risk estimation for breast cancer development; a clinical perspective. Surg Oncol 2002;10:183–92.[CrossRef][Medline]
  7. Sakorafas GH. The management of women at high risk for the development of breast cancer: risk estimation and preventative strategies. Cancer Treat Rev 2003;29:79–89.[CrossRef][Medline]
  8. Hartmann LC, Sellers TA, Schaid DJ, Nayfield S, Grant CS, Bjoraker JA, et al. Clinical options for women at high risk for breast cancer. Surg Clin North Am 1999;79:1189–206.[CrossRef][Medline]
  9. Dite GS, Jenkins MA, Southey MC, Hocking JS, Giles GG, McCredie MR, et al. Familial risks, early-onset breast cancer, and BRCA1 and BRCA2 germline mutations. J Natl Cancer Inst 2003;95:448–57.[Abstract/Free Full Text]
  10. Pike MC, Krailo MD, Henderson BE, Casagrande JT, Hoel DG. "Hormonal risk factors", "breast tissue age" and the age-incidence of breast cancer. Nature 1983;303:767–70.[CrossRef][Medline]
  11. Russo J, Hu YF, Silva IDCG, Russo IH. Cancer risk related to mammary gland structure and development. Microsc Res Tech 2001;52:204–23.[CrossRef][Medline]
  12. Drake DA. A longitudinal study of physical activity and breast cancer prediction. Cancer Nurs 2001;24:371–7.[CrossRef][Medline]
  13. Boyd NF, Greenberg C, Lockwood G, Little L, Martin L, Byng J, et al. Effect at two years of a low fat, high carbohydrate diet on radiologic features of the breast: results from a randomized trial. J Natl Cancer Inst 1997;89:488–96.[Abstract/Free Full Text]
  14. Boyd NF, Lockwood GA, Martin LJ, Byng JW, Yaffe MJ, Tritchler DL. Mammographic density as a marker of susceptibility to breast cancer: a hypothesis. In: Miller AB, Bartsch H, Boffetta P, Dragsted L, Vainio H, editors. Biomarkers in cancer prevention, IARC Scientific Publications No. 154. Lyon: International Agency for Research on Cancer, 2001
  15. Harvey J, Bovbjerg VE. Quantitative assessment of mammographic density: relationship with breast cancer risk. Radiology 2004;230:29–41.[Abstract/Free Full Text]
  16. Wolfe JN. Risk of breast cancer development determined by mammographic parenchymal pattern. Cancer 1976;37:2486–92.[CrossRef][Medline]
  17. Wolfe JN. Breast patterns as an index of risk for developing breast cancer. AJR Am J Roentgenol 1976;126:1130–7.[Abstract]
  18. Boyd NF, O'Sullivan B, Campbell JE, et al. Mammographic signs as risk factors for breast cancer. Br J Cancer 1982;45:185–93.[Medline]
  19. Brisson J, Merletti F, Sadowsky NL. Mammographic features of the breast and breast cancer risk. Am J Epidemiol 1982;115:428–37.[Abstract/Free Full Text]
  20. Whitehead J, Carlile T, Kopecky KJ, Thompson DJ, Gilbert FI, Jr, Present AJ, et al. The relationship between Wolfe's classification of mammograms, accepted breast cancer risk factors and the incidence of breast cancer. Am J Epidemiol 1985;122:94–1006.
  21. Wolfe JN, Saftlas AF, Salane M. Mammographic parenchymal patterns and quantitative evaluation of mammographic densities: a case–control study. AJR Am J Roentgenol 1987;148:1087–92.[Abstract/Free Full Text]
  22. Saftlas AF, Hoover RN, Brinton LA, Szklo M, Olson DR, Salane M, et al. Mammographic densities and risk of breast cancer. Cancer 1991;67:2833–38.[CrossRef][Medline]
  23. Oza AM, Boyd NF. Mammographic parenchymal patterns: a marker of breast cancer risk. Epidemiol Rev 1993;15:196–208.[Free Full Text]
  24. Boyd NF, Byng JW, Jong RA, Fishell EK, Little LE, Miller AB, et al. Quantitative classification of mammographic densities and breast cancer risk: results from the Canadian National Breast Screening Study. J Natl Cancer Inst 1995;87:670–5.[Abstract/Free Full Text]
  25. Boyd NF, Lockwood GA, Byng J, Tritchler DL, Yaffe M. Mammographic densities and breast cancer risk. Cancer Epidemiol Biomarkers Prev 1998;7:1133–44.[Abstract/Free Full Text]
  26. Yaffe MJ, Boyd NF, Byng JW, Jong RA, Fishell E, Lockwood GA, et al. Breast cancer risk and measured mammographic density. Eur J Cancer Prev 1998;7:S47–55.[CrossRef][Medline]
  27. Ringash J. Preventive health care, 2001 update: screening mammography among women aged 40–49 years at average risk of breast cancer. Can Med Assoc J 2001;164:469–76.[Abstract/Free Full Text]
  28. Erbas B, Amos A, Fletcher A, Kavanagh AM, Gertig DM. Incidence of invasive breast cancer and ductal carcinoma in situ in a screening program by age: should older women continue screening? Cancer Epidemiol Biomarkers Prev 2004;13:1569–73.[Abstract/Free Full Text]
  29. Graham SJ, Bronskill MJ, Byng JW, Yaffe MJ, Boyd NF. Quantitative correlation of breast tissue parameters using magnetic resonance and X-ray mammography. Br J Cancer 1996;73:162–8.[Medline]
  30. Lee NA, Rusinek H, Weinreb J, Chandra R, Toth H, Singer C, et al. Fatty and fibroglandular tissue volumes in the breasts of women 20–83 years old: comparison of X-ray mammography and computer-assisted MR imaging. Am J Radiol 1997;168:501–6.[Abstract/Free Full Text]
  31. Blend R, Rideout DF, Kaizer L, Shannon P, Tudor-Roberts B, Boyd NF. Parenchymal patterns of the breast defined by real time ultrasound. Eur J Cancer Prev 1995;4:293–8.[Medline]
  32. Kaizer L, Fishell EK, Hunt JW, Foster FS, Boyd NF. Ultrasonographically defined parenchymal patterns of the breast: relationship to mammographic patterns and other risk factors for breast cancer. Br J Radiol 1988;61:118–24.[Abstract/Free Full Text]
  33. Wrensch MR, Petrakis NL, Miike R, King EB, Chew K, Neuhaus J, et al. Breast cancer risk in women with abnormal cytology in nipple aspirates of breast fluid. J Natl Cancer Inst 2001;93:1791–8.[Abstract/Free Full Text]
  34. Simick M. Near infrared transillumination spectroscopy of breast tissue for correlation with mammographic density. Master's thesis, Department of Medical Biophysics, University of Toronto, 2002
  35. Simick M, Jong R, Wilson B, Lilge L. Non-ionizing near-infrared radiation transillumination spectroscopy for breast tissue density and assessment of breast cancer risk. J Biomed Optics 2004;9:794–803.[CrossRef]
  36. Blyschak K, Simick M, Jong R, Lilge L. Classification of breast tissue density by optical transillumination spectroscopy: optical and physiological effects governing predictive value. Med Phys 2004;31:1398–414.[CrossRef][Medline]
  37. International Electrotechnical Commission. Group safety of laser products, Part 1: Equipment classification, requirements and user's guide. Geneva: IEC, 1993
  38. Egan RL, Dolan PD. Optical spectroscopy. Pre-mammography marker. Acta Radiol 1988;29:497–503.[Medline]
  39. Peters VG, Wyman DR, Patterson MS, Frank GL. Optical properties of normal and diseased human breast tissues in the visible and near infrared. Phys Med Biol 1990;35:1317–34.[CrossRef][Medline]
  40. Key H, Davies ER, Jackson PC, Wells PN. Optical attenuation characteristics of breast tissues at visible and near-infrared wavelengths. Phys Med Biol 1991;36:579–90.[CrossRef][Medline]
  41. Tromberg BJ, Coquoz O, Fishkin JB, Pham T, Anderson ER, Butler J, et al. Non-invasive measurements of breast tissue optical properties using frequency-domain photon migration. Phil Trans R Soc London B Biol Sci 1997;352:661–8.[Abstract/Free Full Text]
  42. Quaresima V, Matcher SJ, Ferrari M. Identification and quantification of intrinsic optical contrast for near-infrared mammography. Photochem Photobiol 1998;67:4–14.[CrossRef][Medline]
  43. Cerussi AE, Berger AJ, Bevilacqua F, Shah N, Jakubowski D, Butler J, et al. Sources of absorption and scattering contrast for near-infrared optical mammography. Acad Radiol 2001;8:211–18.[CrossRef][Medline]
  44. Pogue BW, Poplack SP, McBride TO. Quantitative haemoglobin tomography with diffuse near-infrared spectroscopy: pilot results in the breast. Radiology 2001;218:2616
  45. Shah N, Cerussi A, Eker C, Espinoza J, Butler J, Fishkin J, et al. Noninvasive functional optical spectroscopy of human breast tissue. Proc Natl Acad Sci USA 2001;98:4420–5.[Abstract/Free Full Text]
  46. Durduran T, Choe R, Culver JP, Zubkov L, Holboke MJ, Giammarco J, et al. Bulk optical properties of healthy female breast tissue. Phys Med Biol 2002;47:2847–61.[CrossRef][Medline]
  47. Srinivasan S, Pogue B, Jiang S, Dehghani H, Kogel C, Soho S, et al. Interpreting haemoglobin and water concentration, oxygen saturation and scattering measured in vivo by near infrared breast tomography. Proc Natl Acad Sci USA 2003;100:12349–54.[Abstract/Free Full Text]
  48. Jong R, Fishell E, Little L, Lockwood G, Boyd NF. Mammographic signs of potential relevance to breast cancer risk: the agreement of radiologist's classification. Eur J Cancer Prev 1996;5:281–6.[CrossRef][Medline]
  49. Weersink RA, Marret LD, Lilge L. Validation of self-reported skin colour via principal component analysis of diffuse reflectance spectra of the skin. Proc Soc Photo-Opt Instrum Eng 2000;3917:232–7.
  50. Wise BM. PLS tool box tutorial: MatLab, version 6. Seattle, WA: Eigenvector Research Inc., 2000
  51. Haaland DM, Thomas EV. Partial least squares methods for spectral analysis. 1. Relation to other quantitative calibration methods and the extraction of qualitative information. Anal Chem 1988;60:1193–202.
  52. Haaland DM, Thomas EV. Partial least squares methods for spectral analysis. 2. Application to simulated and glass spectral data. Anal Chem 1988;60:1202–8.
  53. Guo XP, Martin LJ, Hanna W, Benarjee D, Miller N, Fishell E, et al. Growth factors and stromal matrix protein associated with mammographic densities. Cancer Epidemiol Biomarkers Prev 2001;10:243–8.[Abstract/Free Full Text]
  54. Alowami S, Troup S, Al-Haddad S, Kitlepatrick I, Watson PM. Mammographic density is related to stroma and stromal proteoglycan expression. Breast Cancer Res 2003;5:R129–35.[CrossRef][Medline]
  55. Sala E, Warren R, McCann J, Duffy S, Leben R, Day N. High risk mammographic parenchymal patterns and anthropometric measures: a case control study. Br J Cancer 1999;81:1257–61.[CrossRef][Medline]
  56. Gram IT, Bremnes Y, Ursin G, Maskarinec G, Bjurstam N, Lund E. Percentage density, Wolfe's and Tabar's mammographic patterns: agreement and association with risk factors for breast cancer. Breast Cancer Res 2005;7:R854–61.[CrossRef][Medline]



This article has been cited by other articles:


Home page
 Cancer Epidemiol. Biomarkers Prev.Home page
K. M. Blackmore, J. A. Knight, and L. Lilge
Association between Transillumination Breast Spectroscopy and Quantitative Mammographic Features of the Breast
Cancer Epidemiol. Biomarkers Prev., May 1, 2008; 17(5): 1043 - 1050.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Blackmore, K M
Right arrow Articles by Lilge, L
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Blackmore, K M
Right arrow Articles by Lilge, L

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
BJR DMFR IMAGING  ALL BIR JOURNALS