The Fat-glandular Interface and Breast Tumor Locations: Appearances on Ultrasound Tomography Are Supported by Quantitative Peritumoral Analyses.

OBJECTIVE
To analyze the preferred tissue locations of common breast masses in relation to anatomic quadrants and the fat-glandular interface (FGI) using ultrasound tomography (UST).


METHODS
Ultrasound tomography scanning was performed in 206 consecutive women with 298 mammographically and/or sonographically visible, benign and malignant breast masses following written informed consent to participate in an 8-site multicenter, Institutional Review Board-approved cohort study. Mass locations were categorized by their anatomic breast quadrant and the FGI, which was defined by UST as the high-contrast circumferential junction of fat and fibroglandular tissue on coronal sound speed imaging. Quantitative UST mass comparisons were done for each tumor and peritumoral region using mean sound speed and percentage of fibroglandular tissue. Chi-squared and analysis of variance tests were used to assess differences.


RESULTS
Cancers were noted at the FGI in 95% (74/78) compared to 51% (98/194) of fibroadenomas and cysts combined (P < 0.001). No intra-quadrant differences between cancer and benign masses were noted for tumor location by anatomic quadrants (P = 0.66). Quantitative peritumoral sound speed properties showed that cancers were surrounded by lower mean sound speeds (1477 m/s) and percent fibroglandular tissue (47%), compared to fibroadenomas (1496 m/s; 65.3%) and cysts (1518 m/s; 84%) (P < 0.001; P < 0.001, respectively).


CONCLUSION
Breast cancers form adjacent to fat and UST localized the vast majority to the FGI, while cysts were most often completely surrounded by dense tissue. These observations were supported by quantitative peritumoral analyses of sound speed values for fat and fibroglandular tissue.


Introduction
Breast cancer locations can be described relative to their imaging appearance and histopathologic origins. Mammographically, greater cancer incidence within the upper outer quadrants (1,2) has been ascribed to greater parenchymal content or epithelial distribution (2,3), which can obscure cancer detection in women with dense breasts. Breast MRI has used percent fibroglandular volume as a surrogate for localized breast density, but only 20% of cancer locations corresponded to the quadrant with the highest density (ie, percent fibroglandular volume) (4). The fat-glandular interface (FGI) has also been noted on MRI as the dominant location for up to 94%-99% of clinical breast cancers, but they did not address cysts or evaluate the coronal imaging plane (5,6). Automated breast ultrasound (ABUS) is used for screening women with dense breasts (7,8), correlates well with MRI fibroglandular volume (9), and its reconstructed coronal view improves both reading efficiency (10) and mass discrimination (11).
Breast cancer initiation and growth have strong associations with peritumoral fat cells, or adipocytes, and their fat-secreted hormones, adipokines, that mediate blood pressure, reproductive function, appetite, glucose homeostasis, angiogenesis, immune function, and cancer growth (24)(25)(26)(27)(28). Adipokines, such as leptin, have been implicated in the initiation of breast cancer via aromatase expression when the balance tips toward an excessive proinflammatory state (26)(27)(28). Tumor growth of cancer cell lines become markedly accelerated in the presence of cancer-associated adipocytes and adipocyte-derived fibroblasts that contribute to breast cancer progression. These complex peritumoral stromal processes may correlate with hyaluronan deposition and peritumoral apparent diffusion coefficient values by breast MRI (29,30), as well as a stiff peritumoral rim by shear wave elastography (31,32). The FGI thus describes the fibroglandular boundaries with the subcutaneous adipose layer, which may represent the largest endocrine source for breast cancer origin and growth (26).
We hypothesize that cancers will also be preferentially found by UST at the FGI, which is well seen by sound speed imaging in the native coronal imaging plane, and that quantitative tissue properties can objectively differentiate cancers from benign masses, including cysts.

Subjects
For this study, data were used from the clinical diagnostic arm of an 8-site multicenter Institutional Review Board approved, Health Insurance Portability and Accountability Act compliant study of SoftVue UST for dense breast screening (clinicaltrials.gov-NCT #02977247). All participants gave written informed consent for study participation for this observational descriptive cohort study.

Key Messages
• The majority of breast cancers (95%) are located at the circumferential fat-glandular interface (FGI), a high-contrast structure by sound speed (SS) in the coronal plane. • Tissue locations for cysts also suggest they are more likely (64%) to be surrounded by fibroglandular tissue, while fibroadenomas are intermediate in location at the FGI (63%) and surrounded by fibroglandular tissue (35%). • Location of masses relative to the fat-glandular interface may be useful to incorporate in computer-aided US diagnostics for screening in women with dense breasts.
Women of all breast densities were eligible to receive additional UST imaging as part of their clinical visit for evaluation of a palpable or mammographic abnormality. The main inclusion criterion was their willingness to participate with a SoftVue scan during their clinical visit. Exclusion criteria included age <18 years, body weight >350 pounds (ie, SoftVue scanning table projected limit), inability to provide informed consent, inability to lie prone on the UST table, and any open sores or wounds on the breast precluding immersion into the UST water bath for their own safety (ie, sanitized water is exchanged by the system between each patient). No comparable data overlap is noted with any prior or current publications.

Image Acquisition and Analyses
All masses were included between UST scan dates 4/2017-10/2018 for this consecutive data set, using the same version of the SoftVue unit and associated reconstruction algorithms across all centers of the trial. The SoftVue unit displays image acquisition in the coronal plane ( Figure 1) and Table 1 gives clinically relevant performance parameters (13)(14)(15)(16)(17)(18)(19)21,22). To avoid associations with mammographic density, dense breast parenchyma was referred to as fibroglandular tissue and segmented from fat by SS (15,18,19,21,22). All identified masses on UST were biopsy-confirmed by subsequent or prior histology, unless considered as a characteristic cyst by ultrasound criteria. All complicated cysts underwent aspiration with cytologic confirmation. Some women had more than one mass in each or both breasts.

Qualitative Tumor Locations
Mass locations in relation to their anatomic quadrants and FGI were recorded by a central reviewing radiologist (PJL) not affiliated with any of the participating trial sites, with extensive UST experience and 23 years of experience in breast imaging. Masses were categorized according to standard quadrant positions (ie, upper outer, upper inner, lower outer, lower inner) based on distance from the nipple and clock position of the mass on coronal UST, comparing with available clinical imaging (ie, mammography, handheld US and/or breast MRI) for concordance. If a mass location was along a border of 2 adjacent quadrants, it was considered within the quadrant containing the greatest bulk of the tumor on UST.
The boundaries of the FGI were defined on coronal SS images as the high contrast interface between bright fibroglandular tissue and the circumferential darker subcutaneous fat (5,6) ( Figure 2). UST tissue locations were visually sorted into 3 groups: (1) completely surrounded by higher SS (m/sec) fibroglandular tissue, (2) completely surrounded by lower SS fat, or (3) partially surrounded by both (ie, at the FGI) ( Figure 2), rather than using previously reported subjective percentages of mass extension into fat and/or fibroglandular tissue (6), or simply being noted at the FGI (7). A mass could thereby be considered as being at the FGI if, at one extreme, it was surrounded by fibroglandular tissue but abutted a small margin of fat, or conversely if it was surrounded by fat yet abutted adjacent fibroglandular tissue.

Quantitative Tumor Locations
Mass boundaries were hand-traced by the reviewing radiologist to generate quantitative regions of interest (ROIs) using MIM viewing software (MIM Software Inc, Cleveland, OH). Mass margins were traced on their best visualized appearance on a single SS and/or reflection image to generate ROI surface areas ( Figure 3). Once tumor margins were traced, a peritumoral ROI was computer-generated by dilating the tumor margins by 20% of the average tumor diameter, comparable to a symmetric peritumoral "band" (30). The 20% diameter expansion was arbitrarily chosen as a representative compromise, rather than using the complexity of sequential concentric rings (29,30) at this time, which allowed the peritumoral band to remain proportionate for every tumor.
Mean SS values for each tumor and its peritumoral ROI allowed calculations of their differences and ratios between mass types. The tumoral and peritumoral ROIs were further segmented into two regions by SS, corresponding to fibroglandular and fatty tissues using k-means clustering (15,16,18,19,21,22). This allowed for similar comparisons of percent fibroglandular tissue between each ROI and mass type. The amount of fat and/or fibroglandular tissue surrounding masses could thus be quantified and compared between masses.

Statistical Analyses
The study used only descriptive statistics and was not powered to define a specific hypothesis. Comparisons of mean values between the mass types were performed using analysis of variance (ANOVA) analyses. Chi-squared tests were used to assess frequency differences with significance declared at P < 0.05.

Results
A total of 206 women (239 breasts) were included in this study. The average age for study participants was 48.9 years (standard deviation 11.6 years, range 18-82 years). A total of 298 benign and malignant breast masses were noted within 239 breasts (Table 2). Average tumor diameter was larger for cancers as compared to fibroadenomas and cysts (1.3 cm, 1.1 cm, and 1.0 cm, respectively, P = 0.007 Figure 2. Qualitative locations of a cancer, fibroadenoma, and cyst as seen on ultrasound tomography sound speed (SS-top row) and reflection (bottom row). A: 45-year-old woman with heterogeneously dense breast parenchyma and a mildly spiculated 0.7 cm mass that is an invasive cancer (arrow) in the right upper inner quadrant at the fat-glandular interface (FGI) (arrowheads). It is best seen on SS and is ill-defined on reflection, a common finding for small cancers. B: 52-year-old woman with extremely dense breast parenchyma and a 1.6 cm fibroadenoma (white arrow) in the left lower inner quadrant at the FGI (arrowheads). Note that the mass is abutting fat on a small margin, which is more conspicuous on reflection. A fat lobule surrounded by parenchyma creates a pseudomass (black arrow). C: 40-yearold woman with extremely dense breast parenchyma and a 1.5 cm simple cyst (arrow) located within the fibroglandular tissue of the left breast, best seen on reflection and obscured by the diffuse white parenchyma on SS. The SS image shows refraction artifacts blurring the upper and lateral skin margins, compatible with its posterior level as the breast extends toward the axilla, seen only as a thicker skin line on reflection.

Qualitative Tumor Locations
The four-quadrant anatomic distribution (Table 3) showed significantly greater cancer occurrence of 43.6% (34/78) within the upper outer quadrant compared to other quadrants (chi-squared, P = 0.001). Similarly, 37.1% (39/105) of fibroadenomas and 42.9% (39/91) of cysts were also more commonly seen in the upper outer quadrant (chi squared P = 0.003 and P < 0.001, respectively), such that no significant trend was noted separating individual tumor types in the upper outer quadrant (P = 0.648), as well as when comparing cancer with the group of all benign masses (P = 0.688). All tumors were least commonly located in the lower inner quadrant.

Quantitative Tumor Locations
Mean quantitative SS and percent fibroglandular tissue were grouped according to mass type for the tumoral and peritumoral regions in Table 4. The peritumoral region of cancers had the lowest mean SS and percent fibroglandular tissue (1477 m/s and 47.1%), whereas cysts had the highest values (1518 m/s and 84.0%) and fibroadenomas were intermediate (1496 m/s and 65.3%). These quantitative results support the qualitative location results and were indicative of cancers at the FGI being surrounded by both fatty and dense tissue, while cysts were more frequently surrounded by dense tissue. Considering all masses, those located at the FGI had lower mean peritumoral SS and percent fibroglandular tissue than masses located in dense tissue (1484 m/s versus 1524 m/s, P < 0.001; and 53.3% versus 90.7%, P < 0.001, respectively).
Boxplots of the peritumoral SS and percent fibroglandular tissue grouped by mass type are seen in Figure 5. Although there is overlap between cancers and fibroadenomas, in particular, the median peritumoral percent fibroglandular tissue for cysts and cancers are 98.5% and 44.7%, respectively. The majority of the cysts are thus almost entirely surrounded by dense tissue, while cancers are surrounded more by fat.

Discussion
The results of our study show that UST localizes 95% (74/78) of cancers to the FGI, which is seen as a high contrast interface between fat and fibroglandular tissue on coronal sound speed imaging. Conversely, only 55% (120/220) of all benign masses were noted at the FGI (P < 0.001), whereby only cysts were predominantly surrounded by fibroglandular tissue (64%) and a minority of fibroadenomas (35%). All masses were more common in the upper outer quadrant, such that anatomic location  Overall P value between cancer and all benign, P < 0.001. Similarly, overall P values for mass types and cancer versus all benign for anatomic quadrants were P = 0.648 and P = 0.658. showed no significant mass differentiation. These qualitative appearances were supported by peritumoral analyses that segmented quantitative sound speed values and confirmed greater fat surrounding cancers and fibroglandular tissue surrounding most cysts. Clinical mass location results were consistent with mammographic descriptions of greater occurrence of solid benign masses and cancers in the upper outer quadrants (1-3), making quadrant location alone insignificant for mass differentiation. The highly significant 95% cancer occurrence at the FGI by UST was similar to prior MRI studies (5,6) but now includes cysts and supportive quantitative tumoral:peritumoral data. Tumor location respective to the FGI is not a part of the US Breast Imaging Reporting and Database System (BI-RADS) (33), but as a potential future UST criterion, FGI location could be viewed as having a PPV of ~38%, or comparable to US BI-RAD Category 4B (ie, >10% to ≤50% likelihood of malignancy). These initial findings only used SS imaging for quantitation and further UST analytics suggest feasibility for future biological correlates, risk evaluation, and computer-aided detection and/or diagnostic efforts.
Cancer location at the FGI appears to be a reasonable visual search criterion for future UST screening of women with dense breasts. The circumferential periphery of the FGI is readily evaluated by the native coronal imaging plane of UST. Mass margins, as seen by UST, were visually compared to standard imaging and then hand-traced, which is clinically impractical. UST software using automated mass margin detection is being evaluated as part of computer-aided diagnostic efforts for further mass characterization. Moreover, future pixel-based enhancement of the fibroglandular margins that abut fat at the FGI, or associated fat-subtraction techniques, appear feasible and may improve cancer detection in women with dense breasts.
Our effort to quantify the FGI using tumoral:peritumoral data builds on UST work correlating SS imaging to mammographic density (15)(16)(17)(18)(20)(21)(22) and parenchymal distribution by MRI (19). The quantitative nature of SS imaging by UST implies that the peritumoral values are a better differentiator of mass type than values inside the tumor. Tumor ROI characterization by mean SS and percent fibroglandular tissue produced the weakest statistical   Table 3. Note: The "other benign" category was not shown for clarity due to their small numbers.
characterizations (P = 0.073 and P < 0.001, respectively). The strength of the characterizations dramatically increased when the peritumoral regions were measured or compared to the tumoral regions. SS imaging is one of several different images created during a UST scan, and these other quantitative image stacks may contribute to tissue/tumor characterizations but were beyond the scope of this article concentrating on the FGI. For mass evaluations, it may appear counterintuitive that the mean tumor SS of a cyst in this series (1536 m/s) displayed higher values than solid invasive cancers (1527 m/s). Some fibroadenomas may also have had fibrotic components to account for their high mean tumor SS (1535 m/s), which typically occurs as they involute after menopause. The cancers in this data set tended to be irregular in shape and/or have spiculated margins, making their true borders difficult to accurately trace. Cancer ROIs therefore may have inadvertently included some adjacent fat, such that ill-defined margins led to volume averaging and reduced mean tumor SS. Since cysts and fibroadenomas were more circumscribed, their traced boundaries would have more closely resembled their actual boundaries and limited this effect. Additionally, in breasts with multiple cysts, only 2 ROIs were drawn per breast and cysts, with a size of ~1 cm favored. Therefore, smaller cysts may have had cellular, proteinaceous, or inspissated debris that produced higher mean SS values (34,35), rather than common larger simple cysts with average mean SS approaching the value of water (eg, ~1520 m/sec).
Quantitative and volumetric SS parameters also have biological and clinical implications that warrant further work. From a whole breast perspective, defining the predominant origin of cancer at the FGI may better explain that only 20% of cancers occurred in the quadrant with the greatest percent density by MRI volumes (4). Cancer risk may relate more to the actual proportion of the FGI within each quadrant, such that the fat-related biological effects of cancer initiation (24)(25)(26)(27)(28) at the FGI may arise from random genetics occurring within susceptible adjacent fibroglandular tissue within any quadrant. While quantitative tumoral:peritumoral analyses confirmed that cancers were more likely to be surrounded by fat, further comparison of other UST parameters (ie, attenuation and stiffness) to pathology outcomes are needed to better understand biological changes within that fat. These may correspond to peritumoral MRI ADC values (29,30) and/or the stiff rim sign of elastography (31,32), suggesting the need for further developments of computer-aided diagnosis and detection, as well as the defining the optimum diameter and extent of the peritumoral band itself (29,30). Additional quantitative UST metrics may then be feasible for future computer-aided detection and mass characterization in support of dense breast screening and biological correlations. Finally, future quantitative comparisons to tissue specimens may offer insights to the complex biology of adipocytes, adipokines, and cancer cells near the FGI.
Several weaknesses are inherent when a single radiologist used a new breast imaging modality to analyze a subjective Figure 5. Boxplots of the quantitative mean peritumoral sound speed (left) and percent fibroglandular tissue (PFG) (right) grouped by mass type showing the highly significant peritumoral differences noted in Table 3. Note: The "other benign" category was not included due to their lower numbers. criterion, such as the location of a breast tumor residing at the FGI. We chose a simplified three-point system (5,6) incorporating the extremes of a mass being completely surrounded by fat or fibroglandular tissue, compared to any combination of fat and fibroglandular tissue abutting a mass at the FGI. Yet the quantitative peritumoral analyses supported the subjective appearances of the FGI. Additional correlates of peritumoral UST analyses with biological and/ or molecular parameters are also needed, including larger analyses of different cancer types, as well as the peritumoral stiffness more often seen around hormone receptor positive cancers. While the UST coronal plane likely highlights tumor growth and detection (10,11), we also acknowledge that further work is needed on the three-dimensional assessment of both anterior and posterior tumor margins near the FGI. While no apparent recruitment bias was intended (ie, all breast densities eligible), no women with fatty breasts and only a few with scattered fibroglandular densities were encountered in this smaller clinical cohort from a much larger dense breast screening study. As such, our results may not be generalizable to women with lower breast density. Finally, this clinical data set may have been biased by including cancers and masses that were simply more likely to be seen, as opposed to those that may be found during screening from a future data set.

Conclusions
The large majority of clinical breast cancers were visibly found by UST at the FGI, which is biologically relevant to cancer initiation and progression. The significantly greater occurrence of cancers at the FGI, compared to benign masses, was not true of anatomic quadrants where intra-quadrant frequency differences between the different types of masses were not significant. Moreover, quantitative UST results for individual masses and peritumoral regions corroborated the more subjective clinical appearance of greater fat surrounding cancers than benign masses. This study supports the use of the FGI to help guide future visual searches for clinical cancers, comparisons with their biological correlates, computer-aided detection and/or diagnostic efforts, and eventual incorporation into clinical practice for dense breast screening by UST.