Introduction
Breast density, an assessment of the volume fraction of the human breast that containsepithelial and connective tissues, is a risk factor for breast cancer. Numerous studieshave shown that women with the highest density category evaluated from mammography havefourfold to sixfold increased cancer risk compared with women with lower density[
1]. In addition, the International BreastCancer Intervention Study (IBIS-I) primary chemoprevention study, evaluating theselective estrogen-receptor modulator (SERM) tamoxifen [
2,
3], revealed that only women who exhibited>10% reduction in percentage mammographic density experienced the protective effect oftamoxifen in decreased cancer incidence [
4].Women in the treated group who showed <10% decreased density had exactly the samecancer rates compared with the control group. Similar results were also recently foundregarding the use of tamoxifen and aromatase inhibitors in the adjuvant setting[
5]. These and related studies suggest thatbreast tissue density, in addition to being a risk factor, may also be a surrogatebiomarker for monitoring, predicting, and optimizing individual response to hormonaltherapies. However, methods to measure breast density by mammography or MRI have notbeen adopted for various reasons, preventing the utilization of breast-densitymeasurements in the clinic to assess risk or predict outcome.
The only established criterion for assessing breast density is given by the BreastImaging Reporting and Data System (BI-RADS) [
6].That system defines four categories for breast density, qualitatively based on therelative amounts of fat and dense fibroglandular tissue in a mammogram. Although thesystem is useful for evaluating the probability that a tumor is obscured by dense tissueon a mammogram, it is not suitable for quantifying or measuring small changes indensity. An alternative approach is to quantify PMD by using computer-basedimage-analysis techniques of mammograms [
7‐
11]. Whereas it is possible to quantify density from the image, a 2Dradiograph projection is inherently limited in its ability to quantify longitudinaldensity changes accurately [
12]. The use ofionizing radiation also limits the frequency of measurements, making it unsuitable formonitoring adjuvant or preventive chemotherapy. We and others have investigated the useof MRI [
12‐
19], which is asafe and quantitative technique for measuring breast density and volume, but its highcost precludes it from being applied for risk assessment and screening in mostwomen.
Optical-based imaging modalities are a promising alternative to characterize breastdensity. Because of the absorption and scattering properties of breast tissue,near-infrared light (650 to 1,000 nm) is able to penetrate several centimeters deep.Researchers have measured breast tissue in transmission and reflectance geometries byusing continuous-wave spectroscopy, frequency-domain, or time-domain techniques[
20,
21]. Tocorrelate tissue optical measurements with breast density, several groups have comparedspectroscopic features or measured tissue components (such as blood, water, and lipid)with mammographic density. These optically measured parameters have been comparedagainst breast densities that have been qualitatively analyzed from mammograms,classified into only three or four density categories [
22‐
28]. As mentioned earlier, this limits the ability of themeasurement to detect small density changes. Promising work attempted to correlatemammographic density with transillumination optical spectroscopy [
29,
30], but these studies did notquantify tissue scattering and biochemical composition. This makes it difficult tocompare patient spectroscopic features and to link functional changes with underlyingmechanisms of breast density.
We hypothesize that diffuse optical spectroscopic imaging (DOSI) provides quantitativemetrics to measure and track changes in breast tissue composition and density. DOSIprovides a quantitative measure of tissue functional components, allowing noninvasiveimaging of breast tissue composition and metabolism [
31]. DOSI is capable of measuring tissue concentrations (ct) ofoxygenated hemoglobin (ctO
2Hb), deoxygenated hemoglobin (ctHHb), water, andlipid. These measurements are directly related to tissue metabolism and vascularcharacteristics. For example, high levels of ctO
2Hb are considered to be asurrogate marker for elevated vascular supply and perfusion. High levels of ctHHbreflect high oxygen consumption and tissue metabolism due to cell proliferation and/orpoor vascular drainage. Total hemoglobin (ctTHb) corresponds to the total blood volumein tissue and has been validated as an index that corresponds to increased vasculardensity [
32].
DOSI scanning is performed without compression or injection of contrast agents by usinga bedside hand-held probe. Much of this work has been focused on determining functionalchanges in breast tumors during chemotherapy, a topic that is now under investigation inan American College of Radiology Imaging Networks (ACRIN) multicenter clinical trial[
33]. The DOSI technique combineslaser-based frequency-domain photon migration with broadband near-infrared spectroscopyto separate optical absorption and scattering over a broad spectral range [
20]. This results in a quantitative dataset of tissueconcentrations that can be compared longitudinally in the same patient, or acrossdifferent patients.
Our previous studies [
34‐
36] showed that premenopausal women tend to have greaterwater concentration than do postmenopausal women, reflecting the high water content ofepithelial connective-tissue compartments. Similarly, premenopausal women have a higherhemoglobin concentration (both ctHHb and ctO
2Hb), because of greater vasculardemands of the glandular tissue, and a lower lipid concentration. Based on these data,we expect that breast density, which quantifies the abundance of hormonally controlledglandular tissue, exhibits a positive correlation with water and ctTHb and a negativecorrelation with lipid.
Because it is known that neoadjuvant chemotherapy (NAC) affects the density andcomposition of normal breast [
37], in thisstudy, we measured the contralateral normal side of breast cancer patients before andduring NAC treatment with MRI and DOSI. We examined baseline composition, comparingdifferences between pre- and postmenopausal women. We analyzed DOSI parameters formarkers of breast density and metabolism. The correlation between these results andfibroglandular tissue density measured with MRI was examined to test the hypothesis thatDOSI can provide a quantitative measure of breast density. We further hypothesize that,in addition to providing an optical index of breast density, DOSI may help provideinsight into mechanisms of chemotherapy-induced changes in breast metabolism.
Methods
Subject measurements
This study is a retrospective analysis conducted on a subset of subjects with newlydiagnosed, operative, primary breast cancer measured with DOSI or DOSI+MRI duringtheir neoadjuvant chemotherapy treatment between 2007 and 2012. Subject demographicsare shown in Table
1. DOSI measurements were acquired at aminimum of 30 locations (taken in a rectangular grid pattern with 10-mm spacingbetween measurement points) on a contralateral breast not suggestive of malignanciesat two or more time points during the first 120 days of therapy (
n = 28).The DOSI+MRI cohort is a subset (
n = 12) of subjects who also received MRIimaging before NAC. Post-NAC MRI images were available for nine subjects.
Table 1
Subject demographics
DOSI cohort | 47.4 ± 10.2 | 28 |
Premenopausal | 40.8 ± 5.0 | 17 |
Postmenopausal | 57.6 ± 7.2 | 11 |
DOSI+MRI cohort | 43.6 ± 11.0 | 12 |
Premenopausal | 38.l ± 5.3 | 9 |
Postmenopausal | 60.0 ± 4.0 | 3 |
All subjects provided informed written consent and participated in this study underclinical protocols approved by the Institutional Review Board at the University ofCalifornia, Irvine (2002-2306, 2007-6084, and 2011-7812). Exclusion criteria includedpregnant women and women who were younger than 21 years or older than 75 years. Allsubjects were histologically diagnosed with invasive carcinoma before neoadjuvanttreatment.
DOSI measurement
A comprehensive description of the diffuse optical spectroscopic imaging (DOSI)system and underlying concepts may be found elsewhere [
38‐
40]. Inbrief, the instrument combines frequency domain photon migration (FDPM) andcontinuous-wave near-infrared spectroscopy (CW-NIRS) measurements to determine theoptical scattering and absorption spectra (650 to 1,000 nm) of the measured tissue.The FDPM component consists of six laser diode sources (660, 680, 780, 810, 830, and850 nm) that are sinusoidally intensity modulated between 50 and 500 MHz. Therelative amplitude and phase of the detected signals compared with the source areinput into an analytic model of diffuse light transport to determine tissuescattering and absorption coefficients at these wavelengths. White-light illuminationat each measurement point is used for CW-NIRS spectroscopy. The detected broadbandreflectance spectra are fit and scaled to the frequency-domain scattering andabsorption measurements to obtain full broadband absorption spectra over the entirespectral range. Absolute tissue concentrations are calculated by using theBeer-Lambert law and known extinction coefficient spectra of ctHHb,ctO
2Hb, water, and bulk lipid.
All subjects received NAC before surgical resection of tumors and were measured withthe DOSI system before treatment (to establish a baseline measurement), and atseveral time points throughout their treatment. Based on our previous findings,baseline measurements were obtained at least 10 days after diagnostic biopsies tominimize their impact on DOSI scans [
41].Subjects were measured in a supine position. The DOSI probe was placed against thebreast tissue, and sequential measurements were taken in a linear or rectangular gridpattern by using 10-mm spacing. Measurement regions on the normal breast were takento mirror the area of the underlying tumor determined by ultrasound and palpation onthe ipsilateral breast. Total measurement time varied between 20 minutes and 1 hourper subject. Repeated DOSI scans were shown previously to be relatively insensitiveto probe-contact pressure fluctuations, displaying less than 5% average variation intest/retest studies of human subjects [
42].
Mammographic density analysis
Each subject was characterized by a BI-RADS density category. In this system,category I is described as fatty breast tissue, II is scattered density, III isheterogeneously dense, and IV is extremely dense. The categories were compiled fromprechemotherapy mammographic reports documented by the subjects' radiologists andwere available for 20 of the 28 subjects. Of these, four subjects were BI-RADSdensity II, eleven subjects were BI-RADS III, and five subjects were BI-RADS IV.
MR imaging and breast-density analysis
MR imaging examinations were performed by using a dedicatedsensitivity-encoding-enabled bilateral four-channel breast coil with a 3.0-T system(Achieva; Philips Medical Systems, Best, The Netherlands) at time points before,during, and after completion of NAC. The axial-view T1-weighted imageswithout fat suppression were used for the analysis of breast density in this study.The images were acquired by using a 2D turbo spin-echo pulse sequence with TR, 800milliseconds; TE, 8.6 milliseconds; flip angle, 90 degrees; matrix size, 480 ×480; FOV, 31 to 38 cm; and slice thickness, 2 mm.
The breast and fibroglandular tissue segmentation was performed by using a modifiedpublished method [
12,
43,
44]. Before the segmentation, the operator viewed thewhole axial T
1W images dataset and determined the superior and inferiorboundaries of the breast (the beginning and ending slices) by comparing the thicknessof breast fat with that of the body fat. The breast-segmentation procedures consistedof (a) an initial horizontal line cut along the posterior margin of each individualsubject's sternum to exclude thoracic region; (b) Applied Fuzzy-C-Means (FCM)clustering and b-spline curve fitting to obtain the breast-chest boundary; (c) a biasfield-correction method based on nonparametric nonuniformity normalization (N3), andan adaptive FCM algorithm [
44] was used toremove the strong intensity non-uniformity for segmentation of fibroglandular tissueand fatty tissue; (d) applied dynamic searching to exclude the skin along the breastboundary; and (e) the standard FCM algorithm was applied to classify all pixels onthe image. The default setting was to use a total of six clusters, three forfibroglandular tissue and three for fatty tissues. After completing the segmentationprocesses in all image slices, the quantitative breast volume, fibroglandular tissuevolume, and the percentage density (calculated as the ratio of the fibroglandulartissue volume to the breast volume ×100%) were calculated.
Neoadjuvant chemotherapy regimen
Most patients (n = 19) received a 12-cycle, once-a-week course of apaclitaxel (either albumin-bound (nab-paclitaxel) or Cremophor-bound) andcarboplatin. Other patients received only doxorubicin+cyclophosphamiden (AC therapy)(n = 2), or received additional AC therapy either before (n = 4)or after (n = 2) paclitaxel+carboplatin. The remaining patient receiveddocetaxel+carboplatin for six cycles, once every 3 weeks. Many patients also receivedbevacizumab as part of their treatment (n = 15), and some HER2/neu-positivepatients also received trastuzumab (n = 8).
Statistical analysis
Data description
Summary statistics including mean and standard error were calculated for theDOSI-measured parameters of water, bulk lipid, ctO
2Hb, ctHHb, ctTHb,oxygen saturation (stO
2), and tumor optical index (TOI) measured atbaseline and during NAC. TOI is a tissue optical index of metabolism that providescontrast for metabolically active tissue, developed for identifying tumors[
45], and is given by TOI = ctHHb× water/(% lipid). Subject measurements during NAC were recorded as havingoccurred in one of four intervals with interval midpoints 30 days (mean, 33.1;range, 21 to 43/
n = 16 pre-,
n = 11 postmenopausal), 60 days(mean, 60.8; range, 55 to 69/
n = 10 pre-,
n = 7 postmenopausal),90 days (mean, 89.9; range, 78 to 104/
n = 16 pre-,
n = 9postmenopausal), and 120 days (mean, 116.6; range, 106 to 127/
n = 6 pre-,
n = 4 postmenopausal) from the beginning of chemotherapytreatment.
Comparison of DOSI parameters between groups
The Mann-Whitney U test was applied to test whether mean values for DOSIparameters differed significantly between pre- and postmenopausal subjects. TheShapiro-Wilkes test was applied to each DOSI parameter to test the fit to a normaldistribution within BI-RADS categories II, III, and IV. Analysis of variance(ANOVA) was applied to compare mean values among BI-RADS categories for water,lipid concentration, ctO2Hb, ctHHb, ctTHb, and scattering power. Thenonparametric Kruskal-Wallis test was applied to compare the distributions ofstO2 and TOI values among BI-RADS categories. For pairs of BI-RADScategories (II versus III, II versus IV, and III versus IV), the mean values forDOSI parameters were compared with application of the Bonferroni-Holm method ofadjustment for multiple comparisons to maintain an experiment-wise significancelevel of 0.05 for each DOSI parameter.
Regression and correlation analyses
The linear relation between age and water concentration measured at baseline wasassessed with linear regression analysis and the Pearson correlation coefficient.The influence of values obtained from individual patients was assessed byexamination of regression residuals and the DFFITS statistic. In addition, thecorrelation between age and change in water concentration at 90 days during NACwas estimated with the Pearson correlation coefficient.
The correlation between DOSI parameters and MRI fibroglandular density measured atbaseline, as well as at 24 days and 82 days during NAC, was assessed with thePearson correlation coefficient. Similarly, the correlation between percentagechange in DOSI parameters and percentage change in MRI fibroglandular density at82 days of NAC (that is, the end of NAC treatment) was assessed with the Pearsoncorrelation coefficient. A significance level of 0.05 was used for assessment ofestimated correlations.
Generalized estimating equations
We applied a statistical method known as generalized estimating equations (GEEs)to estimate and compare the expected (mean) change from baseline for eachspecified DOSI parameter between pre- and postmenopausal subjects. For example,the GEE method was used to model the linear relation between the mean change inbreast-tissue water as a function of predictors including menopausal status,measurement day after NAC, and the interaction between menopausal status andmeasurement day. Measurement day from the beginning of chemotherapy wasrepresented by a categorical variable with four categories. In contrast toordinary linear regression, for which values measured in individual subjects areassumed to be independent, the GEE method takes in account the correlation betweenDOSI values measured within individual subjects. For application of the GEE, it isnecessary to specify nature of the linear relation between the mean value of theDOSI parameters and the predictors and between the mean and variance of DOSIparameter values. The nature of the within-subject correlations between DOSImeasurements must also be specified. In technical language, we specified a normalmodel with an identity-link function and an exchangeable correlation structure. Inbrief, these specifications indicated that the mean and variance of the DOSIparameter are related through a normal distribution and that the within-subjectcorrelation between repeated measurements of DOSI values was assumed to be thesame for each subject.
From the final GEE model for a given outcome, the estimated percentage change frombaseline was calculated and compared for both the pre- and the postmenopausalgroups 30, 60, 90, and 120 days from the beginning of chemotherapy. For analysisof each outcome, the Bonferroni-Holm method of adjustment for multiple comparisonswas applied to maintain an experiment-wise significance level of 0.05.
Discussion
Breast density is a strong independent risk factor for breast cancer and may also have arole in the risk of recurrence. Optical imaging is a low-cost imaging modality thatshows promise for assessing breast density in the clinic without the use of ionizingradiation. Because optical imaging measures functional characteristics of tissue, italso yields additional information that complements breast-density-assessment techniquesusing MRI and mammography, which are based primarily on structure. This additionalinformation may help to elucidate the origin of breast density and to clarify causes ofboth natural and treatment-induced changes of breast density. By understanding thefactors that modulate breast density and the source of its contrast in imaging, we willbe able to apply the parameter better in risk assessment.
Conflicting information exists in the literature correlating optically measured tissuecomponents with breast density. Most agree that increasing density correlates withincreasing water concentration and ctTHb. However, conflicting data are found on othermeasures, including bulk lipid, which has found to be negatively correlated with density[
27], or not at all [
24]. Oxygen saturation (the fraction ofctO
2Hb to ctTHb, or stO
2) has been shown to be reduced in densetissue [
25], whereas others do not find asignificant correlation [
24,
27]. In addition, Taroni
et al. [
27] included collagen as a basis chromophore and found that it isstrongly associated with BI-RADS categories II-IV. Optical scattering is also expectedto increase with greater breast density because of the underlying fibroglandularstructures, and some studies have indeed shown this [
25,
27]. It is possible that thesedisagreements are due to the broad classification of density categories, to the limitedability of the applied optical modalities to quantify tissue components accurately, orto some combination of the two.
In this study, we found that water, ctO2Hb, ctHHb, ctTHb, and TOI areassociated with the BI-RADS density category, whereas lipid tended to decrease withincreasing category but was not statistically significant. The largest mean differencesin hemoglobin were found between BI-RADS density categories III and IV, but not categoryII and IV. This contradiction is likely due to the small sample size (n = 4with BI-RADS II) and the limitations of a broad density-classification scheme. We expectlipid to be significantly associated with the BI-RADS category in a larger sample size.No association was discovered between stO2 or scattering power with BI-RADSdensity.
Several DOSI parameters were strongly correlated with MRI fibroglandular breast densitymeasured at baseline, including tissue water, lipid, and ctHHb concentrations. Nosignificant correlation was noted between density and scattering parameters orstO
2. This significant correlation with water is likely becausefibroglandular tissue has about 30% higher water content than does adipose tissue[
24]. These observations are in generalagreement with the findings of other groups, with the addition that here we show, forthe first time, that a quantitative measure of water can be compared longitudinally andbetween patients. Because DOSI is a measurement of chromophore concentration throughoutthe probed tissue volume, the increased water may also be due to the increased vascularsupply required by the dense tissue. Furthermore, we demonstrated a strong baselinecorrelation between ctHHb and breast density, reflecting the increased rate ofmetabolism in fibroglandular breast tissue. Because metabolism is strongly captured inthe tissue optical index (TOI), this metric was also shown to be a good predictor ofbreast density. Finally, greater hemoglobin concentrations were observed in patientswith dense breasts, suggesting greater vascular density in these subjects, and lowerlipid concentrations, as would be expected by the greater volume fraction offibroglandular tissue compared to adipose.
At baseline, we observed statistically significant differences in the breast compositionand indicators of metabolism for premenopausal and postmenopausal groups, confirmingprevious studies [
47]. Increased mean ctHHb andmean water concentrations were found, as well as decreased mean bulk lipid concentrationin the breasts of premenopausal subjects, compared with those of postmenopausalsubjects. This is all consistent with increased cell proliferation and metabolicactivity in the denser breast tissue of younger premenopausal women.
Significant changes in optical markers for vascular density and supply(ctO
2Hb and water) were observed during NAC treatment. A significant decreasein ctO
2Hb was observed in both premenopausal and postmenopausal groups. Thesteady reduction of ctO
2Hb without a corresponding decrease in ctHHb suggeststhat NAC agents act directly on the breast tissue, perhaps by causing a reduction ofperfusion. This may be caused by chemotherapy-induced vascular damage and may contributeto the reduction of breast density. In contrast, only premenopausal subjects experienceda significant loss of water in their breast tissue during NAC. This trend better matchesthe observed effect of NAC on MRI-measured breast density wherein premenopausal subjectsexperience a greater loss [
37]. Even though theabsolute DOSI parameters were not significantly correlated with MRI measurements at theend of NAC, the percentage change in water and TOI at the end of treatment compared withbaseline did show strong correlation with the percentage change in fibroglandular breastdensity over the same time.
These NAC-induced changes could potentially have been caused by direct antiproliferativeeffects of chemotherapy, or by the indirect effect of ovarian suppression and subsequenthormone-level reductions [
37]. Cytoxicchemotherapeutic agents are known to cause suppression of ovarian function andamenorrhea, whereas ovarian-secreted hormones (estrogen and progesterone) are known toincrease breast density [
49‐
51]. The greater reduction in breast-tissue water inyounger, premenopausal subjects suggests that their chemo-reduced ovarian hormone levelsmay have a role in reducing breast-tissue density. Consequently, the change inctO
2Hb also suggests a hormone-independent mechanism. The enhancedbreast-density reduction in premenopausal subjects could also be because breast tissuenaturally becomes less dense with age [
52,
53]. Evidence suggests that this is also a hormonal effect[
49‐
51]; however, it is due to natural aging and not induced by thechemotherapy. This is reflected in the reduced baseline water concentration of oldersubjects. If less fibroglandular tissue is present in the breasts of the older women,then it is possible that NAC may not be able to cause a significant change in breastdensity (that is, it has already been reduced to minimum). This complicates the abilityto separate the relative importance of chemotherapy as a direct or indirect modulator ofbreast density.
We note that relatively little change over time was observed in mean DOSI-measured bulklipid in both groups of patients undergoing NAC. This suggests that the rapid changes inbreast density induced by NAC occur because of the reduction of the fibroglandulartissue rather than by increases or replacement by bulk lipid.
This study is limited because it is a retrospective study and includes a small number ofsubjects, especially those with matched DOSI and MRI measurements. Therefore, thereported correlations should be interpreted with caution, and they point to the need forfurther studies. Furthermore, DOSI measurements did not sample the entire breast volume.Recent data have shown, however, that density-related measurements from a spatiallylimited optical sampling on the breast are strongly associated with the BI-RADS category[
28]. Multiple therapy regimens wereincluded in the analysis, and it is likely that the underlying changes in breast densityvary, based on the specific chemotherapy drugs given. Future work stratifying DOSIchanges by treatment type and outcome in a larger population may provide insight intomechanisms of these changes, as well as patient response to therapy. Additionally,timing of the subjects' menstrual cycles was not accounted for, which can causevariations in breast density, nor their pregnancy history. Nonetheless, the ability toidentify significant correlations over several years of optical measurements speaks tothe strength of the DOSI method as a promising quantitative tool.
Competing interests
A Cerussi and BJ Tromberg report patents owned by the University of California that arerelated to the technology and analysis methods described in this study. TheInstitutional Review Board and Conflict of Interest Office of the University ofCalifornia, Irvine, have reviewed both patent and corporate disclosures and did not findany concerns. No potential conflicts of interest were disclosed by the otherauthors.
Authors' contributions
TO designed the study, carried out the data processing and analysis, and drafted themanuscript. AL, JC, SB, DR, and AC assisted with data collection, processing, andanalysis. AM assisted with data processing, whereas CM and WC performed statisticalanalyses and assisted with interpretation of results. MS and BJT conceived of thestudy, participated in its design and coordination, and helped to draft themanuscript. All authors read and approved the final manuscript.