Diffuse optical tomography of the breast: a potential modifiable biomarker of breast cancer risk with neoadjuvant chemotherapy.

The purpose of this study is to evaluate whether a diffuse optical tomography breast imaging system (DOTBIS) can provide a comparable optical-based image index of mammographic breast density, an established biomarker of breast cancer risk. Oxyhemoglobin concentration (ctO2Hb) measured by DOTBIS was collected from 40 patients with stage II-III breast cancer. The tumor-free contralateral breast was used for this evaluation. We observed a moderate positive correlation between the patient's mammogram density classification and ctO2Hb, rs = 0.486 (p = 0.001). In addition, significant reduction in ctO2Hb levels were noted during neoadjuvant chemotherapy treatment (p = 0.017). This observation indicates that ctO2Hb levels measured by DOTBIS could be a novel modifiable imaging biomarker of breast cancer risk and warrants further investigation.

In recent years, several groups [11][12][13][14] have shown evidence that optical based imaging modalities may play an important role in assessing breast tissue composition by measuring optical property contrast from endogenous chromophores. The advantages of optical techniques are the use of non-ionizing radiation, ease of use, and relatively low cost. We developed a diffuse optical tomography breast imaging system (DOTBIS) which does not require the use of contrast agents or compression, and enables imaging of the whole volume for both breasts simultaneously using low intensity near infrared light capable to measure tissue concentration of total hemoglobin (ctTHb), oxy-hemoglobin (ctO 2 Hb), deoxyhemoglobin (ctHHb), oxygen saturation (StO 2 ) and water percentage.
We hypothesize that diffuse optical tomography imaging provides quantitative measurements of tissue functional components, such as ctO 2 Hb, which is directly related to tissue metabolism and vascular characteristics, and could be correlated to mammographic breast density, a known imaging biomarker of breast cancer risk [15]. In addition, we evaluate whether DOTBIS-measured oxyhemoglobin concentration is modifiable after neoadjuvant chemotherapy (NAC).

Materials and methods
This retrospective cohort study analyzed 40 patients with stage II-III breast cancer in a Columbia University Institutional Review Board (IRB) clinical protocol. The patient inclusion criteria for this subset were defined by availability of DOTBIS measurements at baseline, after two cycles of taxol drug infusion or at the end of NAC, and available mammographic breast density assessment before starting NAC. Each patient received a taxane-based regimen. Out of 40 patients, 32 subjects (80%) received 12 cycles of weekly paclitaxel followed by 4 cycles of doxorubicin and cyclophosphamide given every 2 weeks with growth-factor support (T x 12/AC x 4). Two patients received the same treatment but with an addition of carboplatin to paclitaxel (T/C x 12/AC x 4). Three patients received six cycles of docetaxel, carboplatin, and trastuzumab plus pertuzumab (TCHP). Lastly, another three patients received six cycles cytoxan plus taxotere (T/C x 6). Subject demographics according to DOTBIS imaging availability at baseline, after two cycles of chemotherapy, and at the end of therapy prior to surgery, are shown in Table 1. 3 a Average and standard deviation (range) of continuous variable. Abbreviations and Acronyms: NAC, neoadjuvant chemotherapy; T x 12/AC x 4, 12 cycles of weekly paclitaxel followed by 4 cycles of doxorubicin and cyclophosphamide given every 2 weeks with growth-factor support; T/C x 12/AC x 4, 12 cycles of weekly paclitaxel followed by 4 cycles of doxorubicin and cyclophosphamide given every 2 weeks with growth-factor support with an addition of carboplatin to paclitaxel T/C x 6, six cycles cytoxan plus taxotere; TCHP, six cycles of docetaxel, carboplatin, and trastuzumab plus pertuzumab; BMI, Body Mass Index.

Mammographic breast density measurement
Mammograms at our institution were performed on dedicated mammography units (Senographe Essential, GE Healthcare). The views obtained consisted of the standard mediolateral oblique and craniocaudal views. Additional views were obtained if clinically indicated or requested by the reading radiologist. A breast fellowship trained radiologist with 7 years of experience, blinded to the DOT measures, classified mammographic breast density (BD) in accordance with BI-RADS categories: BD I = almost all fatty tissue, BD II = scattered area IV = extremel

DOTBIS
By sampling ctHHb, StO 2 % a continuoussets of four ri cup, Fig. 1 ship between D mammograph lation between , p = 0.006, a nt correlated w gory was nd BD IV aseline for l n n DOTBIShic breast n baseline and water with breast density. After running multivariate correlative analysis, ctO 2 Hb combined to water percentage level was also associated with breast density categories, r s = 0.497, p = 0.001. A Kruskal-Wallis H test was conducted to determine if there were differences in baseline of ctTHb, ctO 2 Hb, ctHHb, StO 2 and water percentage levels between the breast density groups, Table 3. Median ctO 2 Hb levels were the only parameter statistically significant different between BD groups, χ2(2) = 9.374, p = 0.025, and increased from BD I (12.00 μM), to BD II (14.65 μM), to BD III (24.95 μM) to BD IV (35.00 μM), Fig. 3 (mean ± standard deviation are shown in Table 3). Subsequently, pairwise comparisons were performed using Dunn's procedure. A Bonferroni correction for multiple comparisons was made with statistical significance accepted at the p < 0.008 level. This post hoc analysis revealed statistically significant differences in ctO 2 Hb levels between BD II and BD III, p = 0.007. No significant results were available for the comparison with BD I or BD IV which is likely due to the small sample size for both groups (n = 1 for BD IV and BD I).  Thirty-five patients were also imaged after their second cycle of NAC. The first two NAC cycles elicited an overall decrease in ctO 2 Hb levels in 60% (21/35) of the patients, whereas 40% (14/35) participants had their ctO 2 Hb levels slightly increased. Table 4 summarizes the mean values for of ctTHb, ctO 2 Hb, ctHHb, StO 2 and water percentage at three different time points: baseline, after two cycles of taxol drug infusion and at the end of NAC. In Fig. 4, using a grouped scatter plot and drawing a line of equality, we can interpret the overall ctO 2 Hb reduction after two NAC cycles (left), and after NAC completion (right). The patients lying on the line correspond to contralateral breast tissue which experience no change to ctO 2 Hb levels. Those above the line were higher after two NAC cycles than before at baseline, i.e. display an increase in ctO 2 Hb, and those below the line have experienced a reduction in ctO 2 Hb. One can notice that majority of the patients are below the line. A pairedsamples t-test was used to determined that there was a statistically significant mean decrease in ctO 2 Hb levels of 2.39 (95% CI, 0.67 to 4.11) μM after two NAC cycles completion (17.04 ± 7.22 μM) in comparison with ctO 2 Hb levels at baseline (19.43 ± 8.17 μM), p = 0.008. The mean reduction of 3.22 (95% CI, 0.55 to 6.85) μM after NAC completion (n = 24) was also statistically significant, p = 0.024. Statically significant mean reduction was also observed for ctTHb after 2 weeks of therapy, p = 0.011, as shown in Table 4. Data are mean ± standard deviation. For better visualization of the longitudinal hemoglobin reduction trend across NAC, Fig. 5 display the mean values of ctO 2 Hb levels for eighteen patients imaged at all the 3 time points.

DOTBIS features
Baseline (n = 40)  Out of these 35 subjects who were imaged at time point baseline and two weeks after NAC, 26 had their residual cancer burden (RCB) scores available after NAC. To explore any correlation between changes in DOTBIS-measured parameters after 2 cycles of chemotherapy and tumor response, we run an independent-samples t-test to evaluate whether there was a difference between responders (RCB0) and non-responders (RCBI, II and III). The reduction levels of ctO 2 Hb, ctTHb, ctHHb, StO 2 and water percentage were not statistically different between responders and non-responder tumors. For instance, the results show a mean difference of 1.80 ± 2.44 μM in ctO 2 Hb levels after 2 weeks of NAC between responders and non-responders, but it is not statistically significant (p = 0.468).
The association with age was analyzed by considering ctO 2 Hb levels at baseline for all the 40 patients. A linear regression fitting indicates a moderate negative correlation between baseline ctO 2 Hb and age (r = −0.439, p = 0.005). From the coefficient of determination, r 2 = 0.20, we can determine that 20% of the variability in baseline ctO 2 Hb is explained by patient age at enrollment, Fig. 6.

Discussion
In our study, there was a moderate positive correlation between ctO 2 Hb and patient's mammographic density classification. Baseline ctTHb and water percentage level were also statistically significant correlated with breast density. High levels of ctO 2 Hb can be associated with the increased rate of metabolism in dense breasts due to the greater volume fraction of fibroglandular tissue compared to almost entirely fat, and the increased vascular demand required by dense tissues. Additionally, the moderate correlation with water can be related to higher water fraction in fibroglandular tissue compared to adipose tissue [18]. This observation indicates that the ctO 2 Hb levels measured by DOTBIS may be a novel biomarker of breast cancer risk.
Previous publications, summarized by Grosenick et al in their review of optical breast imaging and spectroscopy [19], have shown the feasibility of optically derived data to quantify breast density. Simick et al. categorized density measurements into two groups and presented a PCA model that predicts lower from high-density tissues with 80% of accuracy [20]. In addition, Blackmore et al. correlated breast tissue optical content (water, lipid and StO 2 ) with percent density in pre-and post-menopausal women [21]. In a simulation study, Ruiz et al. found that estimated percent breast density predicted from lipid and water maps was highly correlated to the true values from MRI [22]. Moreover, Fang et al. found linear correlation between ctTHb and the fibroglandular volume fraction derived from the 3D digital breast tomosynthesis imaging scans [23]. However, none of these references specifically correlates the optical data with the four breast density groups from BI-RADS classification, or investigates the changes across NAC treatment.
Similar to our study, O'Sullivan et al. demonstrated a moderate correlation between ctTHb level and FGT assessed by MR imaging, r = 0.597, p = 0.040 [14]. However, they did not find significant correlation with BIRADS classification of mammographic breast density, possibly due to lower number of patients in their study (n = 12). They observed significant difference between mean hemoglobin levels in BD II and BD IV, and water percentage levels in BD III and BD IV. With the larger sample size in our study (n = 40), we were able to detect a significant difference in mean ctO 2 Hb levels between BD II and BD III, a result not observed previously. In addition, unlike our study, the diffuse optical spectroscopy imaging system used in their study did not sample the entire breast volume, and it is unclear if the nipple region was excluded, which can impact accurate assessment due to varied levels of ctO 2 Hb involving the nipple. Taroni et al. also published a work showing direct correlation between mammographic breast density BI-RADS classification and water, lipid and collagen content. However, for a sample size of 49 patients, the correlation between breast density and ctTHb or StO 2 were not statistically significant [24]. It is important to highlight that the lacking of strong correlation with BI-RADS classification could also be justified by the qualitative measurement approach associated to this type of breast density assessment. High variability between radiologists are expected [25] and quantitative measurements of breast density could reveal stronger association with the optical features.
Because optical imaging provides quantitative metrics capable to measure and track changes in breast tissue composition, we reconstructed the 3D full volume of ctTHb, ctO 2 Hb, ctHHb, StO 2 and water percentage maps of 40 patients undergoing NAC treatment to determine whether this metric is modifiable. Our results indicate significant reduction in ctO 2 Hb levels after NAC completion. Several studies have reported a consistent reduction of breast density after NAC [26][27][28][29]. By using MRI or mammographic imaging, all these groups were able to detect change in percent breast density across NAC. For example, Chen et al., using 3D MR imaging, demonstrated that patients who underwent AC and taxane had a mean percentage reduction of 12.7% at the end of NAC [27]. Besides, Sandberg et al. and Knight et al. took a step further and showed that women who experienced a decrease of at least 10% in breast density after chemotherapy or tamoxifen, had decreased their risk in half of contralateral breast cancer. According to the literature, chemotherapy, including taxane based regimens, has been linked with amenorrhea and ovarian function suppression. The rapid reduction in estrogen from chemotherapy induced menopause contributes to secondary change in breast tissue composition, and consequently, an impact in mammographic breast density [30,31]. In similar fashion, our results showed that ctO 2 Hb measured by DOTBIS is also modifiable with NAC and it had a significant mean reduction of 12% after the first 2 cycles of NAC and approximately 17% at the end of NAC. However, no statically significant reduction was noticed in ctHHb levels. Similar finding was published by O'Sullivan et al. in the same work discussed before. The reduction of ctO 2 Hb levels and no significant changes in ctHHb might suggest that NAC induces reduction of blood perfusion by vascular damage, and consequently, breast density reduction. It is possible that ctO 2 Hb, which is more representative of arterial blood supply carrying oxygen to the breast tissue, could be more variable to meet the demands of the amount of fibroglandular tissue, resulting in a positive moderate correlation. While ctHHb, which is predominantly venous blood that carries away metabolic byproducts including deoxygenated blood after tissue consumption, may be less variable and resulting in non-correlation with the amount of fibroglandular tissue in our small sample size. While O'Sullivan et al. found statically reduction in ctO 2 Hb levels about 90 days after start of NAC, we found a significant reduction already after 15 days. Since all patients in our study were administered taxane based regimen for the first weeks of treatment, we believe that same treatment type was responsible to minimize any variability from drug-induced changes in the breast tissue, and improved the intra and inter-patient analyses.
Regarding correlations between changes in the hemoglobin levels and tumor response to NAC, our results showed that the reduction levels of ctO 2 Hb, ctTHb, ctHHb, StO 2 and water percentage were not statistically different between responders and non-responder tumors. However, after running a statistical power analyses using G*Power 3.1, a sample size of at least 40 patients would be necessary to see a statistically significant difference in the changes of ctO 2 Hb levels between these two subgroups with an 80% power. Given that our study was limited by a small sample size with RCB scores available (n = 26), the statistical power to detect such a difference in ctO 2 Hb between the two groups was only 47%. Therefore, further studies are needed to demonstrate how DOTBIS-measured parameters change within the context of response to NAC.
We also examined the relationship between baseline ctO 2 Hb and age. Breast density is known to decrease with increasing patient age due to postmenopausal alteration of glandular breast tissue [32,33]. Our results presented similar association. There was a significant inverse relationship between age and ctO 2 Hb measured at baseline (p < 0.005). As already discussed, breast tissue becomes less dense with age, therefore the correlation between the lower levels of ctO 2 Hb might be related to the reduced vascular supply and perfusion occasioned by less fibroglandular tissue present in the breast of the older women.
Major limitation of our study was its dependence on the patient's mammogram availability for her inclusion in the study. The protocol at our institution does not include a post-NAC mammogram, and we were not able to have an end-point correlation between ctO 2 Hb and mammographic breast density. Patient's menstrual cycles at baseline was not uniformly reported, and therefore were not accounted for analyses, which could cause variation in breast tissue composition and density assessment. In addition, we were not able to expand our analyses to correlations between MR based quantitative breast density measurement with ctO 2 Hb since the majority of the women in our cohort did not had access to MR screening at pre-NAC time point. Furthermore, our study population was already diagnosed with breast cancer. Further studies in BRCA mutation carriers, a high-risk group, are needed to correlate DOTBIS-measured ctO 2 Hb with breast density and how is related to cancer risk. If such a relationship can be established, it may be possible to use DOTBIS measurements to predict breast cancer risk.

Conclusion
Optical-based image index by DOTBIS may be a novel modifiable marker of breast cancer risk that is 3D quantifiable and without exposure to ionizing radiation. Potential for its use as a predictor of breast cancer risk as well as an assessment tool to longitudinally evaluate efficacy of various chemoprevention strategies is warranted.

Funding
Breast Cancer Research Foundation; National Institutes of Health (grant 11223309).