1.

Introduction

Despite proven benefit in breast cancer specific mortality,^1,2 the utility of screening mammography has been called into question on the basis of limited specificity^3,4 and overdiagnosis.^5,6 Digital breast tomosynthesis (DBT) has shown potential to mitigate these limitations based on reduction in screening recall rate and preferential detection of invasive carcinoma.^7,8 The U.S. Preventive Services Task Force in its most recent recommendation concluded that there was insufficient evidence to make a recommendation for DBT as a primary screening tool.⁴ There is a need for a physiology-based screening tool that is not constrained by dense breast composition, and one that may identify breast cancer that is more likely to cause morbidity and mortality over a subject’s lifetime.

Physiology-based imaging, such as near-infrared spectral tomography (NIRST), is an ideal adjunct. It does not require administered contrast agents, making it suitable for imaging almost all women. NIRST uses light wavelengths between 650 and 1000 nm to image several functional markers in tissue, including hemoglobin concentration, oxygen saturation, and water and lipid levels. When used by itself, NIRST suffers from poor spatial resolution due to pervasive light scattering. However, NIRST has demonstrated improved resolution to nearly 1 mm⁹ when paired with other modalities, including x-rays, ultrasound, and magnetic resonance imaging (MRI). Earlier work by Fang et al.¹⁰ demonstrated the potential for NIRST combined with DBT to distinguish benign from malignant lesions in a trial of over 100 women. The study found higher total hemoglobin and higher scattering coefficients in cancer compared to benign lesions and fibroglandular tissue within the same breast.¹⁰

We have developed a prototype NIRST/DBT system that expands on the approach described by Fang et al. by (i) exploiting DBT data to derive anatomical priors to improve NIRST image resolution and accuracy of the recovered optical properties¹¹ and (ii) adding more optical illumination wavelengths (8 versus 2) and light detectors (75 versus 40).¹² These advancements enable assessment of water and lipid content, in addition to hemoglobin and oxygen saturation. One challenge with NIRST/DBT is the use of breast compression, which is required only for the DBT part and can alter these physiological signatures, notably the hemoglobin and oxygenation status of breast tissue in both normal and malignant cases.¹³

In this study, the largest cohort of women with normal (BI-RADS 1) mammography was imaged with the coregistered NIRST/DBT under two compression levels to evaluate optically derived tissue property changes. Subjects selected for this study included a range of breast densities enriched with a larger proportion of women with dense breasts, a subgroup that is less well served by mammography and at greater risk of developing breast cancer.⁷ The purpose of this study was to benchmark optically derived tissue properties across a range of breast densities and assess the effects of compression on estimates of hemoglobin, oxygen metabolism, and water and lipid concentrations obtained from the coregistered NIRST/DBT imaging system.

2.

Materials and Methods

Subjects enrolled in the study followed experimental protocols that were approved by our Institutional Review Board. Written informed consent was obtained from all participants. The study was compliant with Health Insurance Portability and Accountability Act guidelines. Sequential eligible subjects were selected for enrollment to provide a range of breast size and mammographic breast density that was enriched with greater breast density, i.e., more subjects with heterogeneous or extremely dense breasts than scattered or predominantly fatty breasts. Selection was based on the clinical radiology report of breast density and subject reported bra cup size. Subjects were imaged with NIRST/DBT from June 2013 to March 2014.

3.

Results

Thirty-two women with normal mammography (BI-RADS 1) within the prior 2 years were enrolled and underwent the NIRST DBT imaging protocol. The first five subjects were excluded due to poor DBT quality and unrecorded laser power settings that could not be determined retrospectively. Of the remaining 27 subjects, the data of one woman were restricted to the craniocaudal (CC) projection due to a software malfunction prior to the MLO acquisition. This yielded four image sets per subject (CC fully compressed, CC mildly compressed, MLO fully compressed, MLO mildly compressed), except for the one subject who was imaged only in the CC view; in that case, there were only two sets of images. In total $26\times 4+1\times 2$ images were obtained, half mild and half fully compressed. The exact numbers include 53 NIRST mildly compressed, 53 NIRST fully compressed, and 53 DBT fully compressed image acquisitions.

Figure 1 summarizes the characteristics of the women imaged, including their BMI, breast thickness, percent fibroglandular tissue,¹⁸ breast density, bra cup size, and age. The average BMI was 28.5, nearly 29—the US average for older women.²¹ The average age was 51 years. The average compressed breast thickness was 4.9 cm, similar to other studies (4.4 to 5.4 cm).^22,23 The average fibroglandular content of 19% was marginally higher than the average population (10% to 17%).^24,25

Fig. 1

Histograms of key demographic and clinical information for the subjects enrolled in the study.

3.1.

Optically Derived Breast Tissue Properties

Mean HbT was $14.3\pm 6.4\mu \mathrm{M}$, oxygen saturation was $53\pm 17\%$, water fraction was $12\pm 9\%$, and lipid content was $34\pm 18\%$ for all exams (including both full and mild compression), as shown in Table 1 with confidence intervals. Additionally, examination of adipose and fibroglandular tissues separately showed mean HbT of 14.1 versus $15.1\mu \mathrm{m}$, oxygen saturation of 55% versus 48%, water content of 11% versus 14%, and lipid content of 35% versus 29%, respectively, but none of these differences was statistically significant. This is not surprising, given the interpatient variability of breast tissue. For the group imaged with both the DBT and DOSI systems, a comparison of the optically derived property estimates using the individually recorded scattering versus the group averaged scattering demonstrated minimal differences ($r$ values for oxy-, deoxy-hemoglobin, and water were all greater than 0.9, data not shown). Comparing CC and MLO breast positions demonstrated no significant differences between the views as a whole (data not shown). In comparing individual subjects, statistically significant correlations in hemoglobin, oxygen saturation, and lipids between the two views were confirmed ($r=0.6$, 0.6, 0.7, respectively). Mean water values were within 1% when comparing CC and MLO and were not significantly different between the two groups, but when examining the correlation of CC and MLO views for individual women it did not reach a level of statistical significance ($p=0.14$) as there was less variation in water values across women than there was in lipid content.

Table 1

Review of selected publications of optically derived tissue properties of normal breast relative to the current study. Average values are shown with standard deviations in parentheses.

Study source	Number of subjects	Total Hb (μM)	O2 Sat %	Water %	Lipids %
Dartmouth NIRST/DBT	27	14.3(6.4)	53(17)	12(9)	34(18)
Ref. 26 (pre- and postmenopausal subjects)	95, 107	15.7(4.9), 12.4(4.7)	70.5(7.6), 68.5(8)	21.3(7.2), 18.2(5.9)	61.5(10.2), 68.3(9)
Ref. 27	49	20.4(7.1)	71.1(7.7)	28.9(11.7)	62.4(12.6)
Ref. 28	60	16.6(4.6)	68.3(7.2)	51.5(14.1)
Ref. 29 (pre- and postmenopausal subjects)	93, 102	13.64(4.1), 10.4(3.2)	86.8(9.7), 84.9(9.1)
Ref. 10	138	19.2(6.5)	73(6)
Ref. 30	87	17.3(6.3)	74(7)
Ref. 31	68	16.2(6.1)	71
Ref. 32	58	17.5(7.5)	67.7(9.3)
Ref. 33	14	18(5)	69(6)
Ref. 34	27	12(6)	71(10)
Ref. 35	3	17(2)	68(4)
Ref. 36	12	25.3(8.3)	70(7)
Ref. 37	19	22.3(7.5)	64.8(10.3)
Ref. 38	55	21.2(11.7)
Ref. 39	15	21.8(9.2)

As shown in Fig. 2, significant correlations between optically derived tissue properties and breast density were observed. The difference in HbT and lipids was statistically significant among the three higher density groups. No statistically significant differences in oxygen saturation occurred across the groups. Average values are compared to data from previous reports in the literature in Table 1. A full list of correlations between optically derived tissue properties and clinical parameters is summarized in Table 2.

Fig. 2

Comparison of optical data for total hemoglobin, oxygen saturation, and water and lipid concentrations for different breast density categories (F, fatty; S, scattered; HD, heterogeneously dense; ED, extremely dense). Statistical significance between breast density categories occurs where angled portions of the box plot bands do not overlap between two bars.

Table 2

Correlations between breast density, bra cup size, BMI, fibroglandular volume fraction, breast thickness, and DOSI with DBT/NIRST optically derived tissue properties using Student t tests. Bolded values indicate statistical significance.

Breast density	Correlation	p value	Bra cup size	Correlation	p value	BMI	Correlation	p value
HbT	0.64	0.001	HbT	$-0.18$	0.42	HbT	$-\mathbf{0.45}$	0.034
${\mathrm{StO}}_2$	$-0.10$	0.66	${\mathrm{O}}_2$ Sat	0.54	0.01	${\mathrm{O}}_2$ Sat	0.25	0.27
${\mathrm{H}}_2\mathrm{O}$	0.62	0.003	${\mathrm{H}}_2\mathrm{O}$	$-0.28$	0.22	${\mathrm{H}}_2\mathrm{O}$	$-0.38$	0.09
Lipids	$-\mathbf{0.74}$	$<\mathbf{0.001}$	Lipids	0.27	0.24	Lipids	0.646	0.002
Fibroglandular content	Correlation	$p$ value	Thickness	Correlation	$p$ value	DOSI	Correlation	$p$ value
HbT	0.68	$<\mathbf{0.001}$	HbT	$-0.18$	0.39	HbT	0.47	0.30
${\mathrm{StO}}_2$	$-0.12$	0.58	${\mathrm{O}}_2$ Sat	0.71	$<\mathbf{0.001}$	${\mathrm{O}}_2$ Sat	0.15	0.75
${\mathrm{H}}_2\mathrm{O}$	0.63	0.001	${\mathrm{H}}_2\mathrm{O}$	$-\mathbf{0.63}$	0.001	${\mathrm{H}}_2\mathrm{O}$	0.73	0.07
Lipids	$-\mathbf{0.80}$	$<\mathbf{0.001}$	Lipids	0.56	0.005	Lipids	0.92	0.003

Fibroglandular content was statistically correlated with total hemoglobin, water, and lipids, and opposite to the correlation observed in BMI (data not shown). The only significant correlation between bra cup size and optically derived tissue properties was found with oxygen saturation as shown in Fig. 3 and listed in Table 2. Although trends are the same for both bra cup size and thickness, bra cup size did not demonstrate significant correlation with lipid and water content, likely reflecting the self-reported nature of this data point.

Fig. 3

Comparison of total hemoglobin, oxygen saturation, and water and lipid concentrations with subject reported bra cup size. Statistical significance between categories occurs where angled portions of the bands do not overlap.

3.2.

Correlation with Compression

As shown in Fig. 4, optically derived properties were similar in fully compressed and mildly compressed cases with no significant differences except for oxygen saturation ($p=0.04$)—an average of 58% for the mildly compressed relative to 50% for the fully compressed cases. Additionally, breast thickness was highly correlated with oxygen saturation as well as with water and lipid content as seen in Table 2.

Fig. 4

Comparison of optically derived tissue property estimates in mildly compressed and fully compressed views.

4.

Discussion

Mean HbT was $14.3\mu \mathrm{M}$, which is on the low end of the previously reported values shown in Table 1, but is similar to the report by Fang et al. ($16\mu \mathrm{M}$³¹ and $19\mu \mathrm{M}$¹⁰). Both water and lipid contents were also lower than previously published results as shown in Table 1. This finding is likely influenced by light channeling in the plastic compression paddles, as the effect was similar for every participant, all of whom were examined with the same compression paddles.¹⁵ The effect is more significant for water and lipids than for hemoglobin content because light channeling increases at longer wavelengths, coinciding with the higher absorption spectra for water and lipids. Additionally, lower lipid content would be expected since the fatty breast subject population was underrepresented.

A more specific and individualized assessment of the NIRST/DBT system was obtained by comparing data from 15 women who were imaged with both DOSI and NIRST/DBT. DOSI has demonstrated successful differentiation of benign and malignant tissues, the ultimate goal of NIRST/DBT imaging.³² This comparison was important for two reasons. First, optically derived tissue properties from the two systems should be similar in these patients, and the DOSI data were useful in verifying results from the new NIRST/DBT system, although identical values are not expected given differences in imaging geometry and the volume of tissue assessed by the two approaches. Second, scattering values cannot be estimated from our NIRST/DBT continuous wave acquisitions and availability of DOSI values was important for assessing the consequences of not obtaining subject-specific optical scattering estimates for NIRST data/image analysis of the remaining participants. The impact of using individually measured DOSI scattering properties versus the average scattering value from the whole group on other optically derived tissue properties was relatively modest ($r>0.9$ for oxy- and deoxy-hemoglobin and water, data not shown). Hence, the DOSI-derived averaged scattering may be used in lieu of breast-specific scattering data with minimal effects on the resulting breast-specific absorption values. Although the data from breasts imaged with both DOSI and NIRST/DBT systems demonstrated a small 3% difference in water content (15% for DOSI and 12% for NIRST/DBT), but a larger disparity in lipid fraction (34% versus 58% for NIRST/DBT versus DOSI), these differences are likely due to the preferential subcutaneous fat sampling of DOSI compared to NIRST/DBT, which probes the entire breast volume—a hypothesis further supported by the correlation between the fat values found in NIRST/DBT and DOSI ($r=0.92$, $p=0.003$).

In comparing results from the CC and MLO views, no statistically significant difference was found between the two groups as a whole, a reassuring result for reproducibility. Significant correlations in hemoglobin, oxygen saturation, and lipids existed between the two views for individual subjects, supporting the potential of using continuous wave near-infrared methods for clinical breast imaging—a less expensive, less complex and thereby more easily integrated approach into the clinic, which requires efficient workflow. Furthermore, the rapidity of the NIRST acquisition, which is a much faster alternative to frequency or time domain optical methods, allows seamless integration into the clinical examination.

Our results confirm expected relationships between optically derived tissue properties and subject characteristics including breast density, fibroglandular volume, and BMI. While these correlations have been noted for stand-alone optical imaging systems, heretofore, they have not been demonstrated in optical fusion technology based on spatial priors from DBT and continuous wave optical data acquisition in the compressed breast. Denser breasts and breasts with higher fibroglandular volume demonstrated significantly higher water and hemoglobin content and lower lipid levels without demonstrable correlation with oxygen saturation. Increasing BMI showed opposite trends, with higher recovered lipid fractions and lower hemoglobin concentrations.

The degree of breast compression did not significantly alter total hemoglobin content as in prior reports. Previous studies involved imaging systems in which light sources and detectors were on the same side of the breast, limiting the sampling volume and preferentially probing more superficial tissues. Blood may be redistributed toward the stiffer, deeper fibroglandular tissue that can be detected by NIRST/DBT. Only one prior study by Busch et al. used transmission measurements of the breast under compression, and demonstrated changes in hemoglobin and oxygenation, as well as an 88% decrease in blood flow at mammographic compression levels.⁴⁰ Although a trend toward lower hemoglobin was noted in our study, the result was not statistically significant, and may have been due to unaccounted changes in optical scattering.

In contrast, full breast compression had a substantial effect on tissue oxygenation with a significant decrease occurring under full compression ($p=0.04$). Reduction in oxygen saturation is expected during compression, assuming blood flow is restricted, since tissues consume the remaining oxygen in the static blood to meet ongoing metabolic demands. Prior studies with fewer subjects have demonstrated similar results.^41,42 Additionally, oxygen saturation was lower in this study when compared to literature values, and may be due to compressive effects that occur even under mild compression, which would not be present in previous data as most other systems do not place any pressure on the breast tissue during examination.

A simpler approach to some of the analyses presented here would be to ignore spatial information and analyze the whole breast as homogeneous. However, the next step for this imaging system is to examine women with breast lesions, particularly the small lesions found on breast screening exams. Homogeneous analysis would likely be incapable of detecting minute changes in optical properties expected in small tumors. Spatial priors have been shown to improve quantification of tissue properties and incorporating this information simulates future abnormal breast studies more closely than a homogeneous analysis.¹¹

Several limitations in the study are related to subject selection, system design, and compression measurements. The sample of subjects is not representative of the general mammographic screening population, as we intentionally chose women with denser breasts and recruited only a single subject with fatty composition. The continuous wave signals acquired with NIRST do not measure differences in scattering, an optical property that has been shown to differ in benign and malignant tissue.¹⁰ Unfortunately, our research NIRST/DBT system was not equipped with a compression force sensor, so breast thickness was used as surrogate. In terms of the lipid and water changes, the results reported here may be confounded by other variables such as BMI, which was highly correlated with breast thickness ($r=0.79$, $p<0.001$), and shows the same trends as seen for thickness in regards to water and lipids. However, BMI was not related to changes in oxygen saturation, and a significant correlation was found between breast thickness and oxygen saturation ($r=0.71$, $p<0.001$), which can be attributed to the compressive forces applied to the breast.

In summary, our results benchmark optically derived tissue properties estimated through the NIRST/DBT system and reveal some important differences between our completely coregistered data and previously published values. The results presented here represent the largest and most comprehensive analysis to date on the effects of compression on optically derived tissue properties. Mammographic related compression alters the oxygen saturation of hemoglobin in the breast. Additionally, differences in breast density can be ascertained using the NIRST/DBT platform by measuring the optically derived tissue properties of water, lipids, and hemoglobin. This outcome is critical for the development and success of the NIRST/DBT platform, which benefits from the synergy of its two component technologies. From the mammography perspective, NIRST contributes important physiological information. From the optical imaging perspective, the addition of DBT may improve the accuracy and resolution of the recovered optically derived property data as has been previously demonstrated for MRI.¹¹ From a practical standpoint, the lower cost of the NIRST system design and the rapidity of the NIRST acquisition further adds to its clinical application. NIRST/DBT is a robust imaging technique that uses the detailed anatomic information from DBT to provide additional physiologic information that is highly correlated with breast density and minimally altered by changes in breast position or compression.