In this study, we evaluate a quantitative US method (g-SoS) for assessing breast density based on global SoS measurement, which is found to be highly repeatable and strongly correlated with m-ACR categories. Dense breasts and extremely dense breasts were successfully classified with high accuracy. This study shows that quantitative global SoS measurement with standard US hardware and conventional hand-held transducers can effectively differentiate between different breast density categories.

Our results show that, while visual B-mode assessments do not correlate with m-ACR categories (r = 0.133), there is a significant correlation and accordance between m-ACR categories and g-SoS measurements (r = 0.773) similar to that reported in [23] (r = 0.746). This corroborates the correlations observed between volume-averaged SoS values from 3D US computed tomography and mammography percent density [20, 22]. In [31] the patients classified as density type 4 (dense breast) were found to have a significantly higher breast SoS value compared to those categorized as density type 1 (fatty), 2, and 3; classified according to the BI-RADS 4th edition that used numbers instead of letters. In contrast to these US-based approaches, the method utilized here only requires conventional US hardware and does not necessitate any additional physical components. Note that during US examination we used the FDA- and CE-certified UF-760AG for collecting raw data, which we later processed using the described methods to measure global SoS in the breast, which is then an off-label use of this device.

In this study, for extremely dense breast (m-ACR D) classification using g-SoS, the AUC was 0.906, reflecting a high level of diagnostic accuracy. With the chosen g-SoS threshold, all extremely dense breasts were identified correctly with 100% accuracy, as seen in Table 3. The AUC for dense (C or D) breast classification was 0.931, further indicating a high accuracy. With the chosen g-SoS threshold, all non-dense breasts (m-ACR A or B) were correctly classified as non-dense. This is a major improvement over [23], where 17% of patients with non-dense breasts (6/36) were incorrectly classified as dense breasts (C or D). Furthermore, at 100% specificity, the method presented here demonstrates a sensitivity of 78%, which is substantially higher compared to the 15% reported in [23].

Breast density is known to decrease with aging. Our results show a moderate, negative linear relationship (r = –0.581) between quantitative g-SoS and age. This corroborates the findings of [30], where an inverse relationship between age and mammographic breast density was shown with 7007 participants. In that study, a sizeable proportion of young women exhibited predominantly fatty breasts, while a subgroup of older women showed extremely dense breasts [30]. This aligns with our observations from Fig. 6, where some young patients present g-SoS values lower than some elderly patients. Hence, age is not a surrogate for breast density, further emphasizing the need for its independent assessment and reporting.

We found that SoS measurements were highly repeatable at both assessed positions, with an ICC of 0.988 for position 1 and 0.991 for position 2; comparable to those intrareader agreements reported in [23] (ICC = 0.966–0.994). When comparing measurements taken at different orientations of the same breast anatomy, the g-SoS repeatability was also good, with an ICC of 0.908. Notably, these ICC values are considerably superior to intrareader agreements reported for ACR mammography breast density, e.g., ICC = 0.77 in [32] or inter-reader agreements (ICC = 0.57 [32] and ICC = 0.731–0.774 [23]). Being an automatic and quantitative method, our presented approach is inherently more robust to reader variability. Note that the goal of our repeatability study was not to investigate the repeatability of readings throughout a breast, but rather to test the robustness and repeatability of multiple measurements and different probe orientations around the same location, the UOQ in this case. Our results indicate that the g-SoS-based density measurement from the UOQ alone gives comparable results to X-ray mammography-based readings. This finding highlights the clinical applicability of g-SoS at a single indicative location (UOQ).

The clinical use of the terminology breast density refers mainly to the amount of glandular tissue observed. Breast epithelium and stroma attenuate X-rays more than fat and hence appear whiter on mammograms compared to fat [6]. From a medical physics and biomechanics perspective, SoS in a medium is a function of the medium’s (bulk) stiffness and density, where SoS is proportional to stiffness but inversely proportional to density. According to this, for higher breast (physical) density, a lower SoS should be observed. This then appears to contradict the findings herein as well as several earlier works. Nevertheless, this can be explained considering that the glandular tissues are expected to also have higher stiffness compared to fat. The empirical observation that SoS is higher in clinically denser breasts indicates that the breast (bulk) stiffness is increasing more than its (biomechanical) density. Differentiating these two biomechanical variables and potentially assessing their comparative value as a cancer risk factor independently can be an interesting future research direction. Such differentiation nevertheless requires other concurrent measurement modalities (e.g., Hounsfield units to factor out physical density) and/or ex-vivo biomechanical experiments on excised breast tissues, as in [33, 34], which are beyond the focus of our current work.

Our study has a few limitations. Our cohort includes a few women younger than 40 and a few on hormone replacement therapy, so efficacy in these groups may require further studies. Also, in our cohort, there were only two patients with m-ACR A, which inherently limits the statistical significance achievable in categorizing this class separately. Nevertheless, differentiating the category A alone potentially has limited clinical relevance. Hence, we do not make any ACR A classification alone, but instead group them together with ACR B in the analyses. This is also clinically meaningful since ACR A and B typically do not have different risk profiles, prognoses, or patient management options. In the case of dense breast classification, 30% of patients (14 of 47 in m-ACR C) were inaccurately classified by g-SoS as non-dense breasts (A or B). Diagnosing ACR category C is indeed challenging because these breasts exhibit both dense and non-dense regions. This was also indicated by our inter-reader observations as well as in the literature [1]. While mammography and US are intended to examine the same anatomical part of the breast, any disparity between m-ACR and g-SoS in breast density classification may be attributed to potential shifts in analyzed regions of interest between these modalities, e.g., their projective vs. cross-sectional imaging natures as well as m-ACR assessing all breast quadrants while g-SoS in this study having been measured only in the UOQ. Furthermore, the differences between m-ACR and g-SoS in breast density classification may also arise from the subjective nature of mammography assessment. Accordingly, one could speculate the quantitative g-SoS measurements are comparatively more reliable and observer-independent, but this requires further studies to confirm with additional gold-standard measurement approaches such as MRI.

Our US-SoS technique utilizes a conventional transducer, covering only a cross-sectional slice of the breast, lacking comprehensive coverage. In our study, data was collected from two different probe orientations around the same location in the UOQ since fibroglandular tissue and mammary gland volumes and hence the quadrant percent breast density are known to be higher in the UOQ than in the other quadrants [35, 36], and the latest BI-RADS lexicon [1] instructs that the densest breast part should be used for breast density classification. Therefore, mammography-based breast density classification is mostly likely to depend on the fibroglandular tissue content in the UOQ. This is also clinically relevant because denser breast portions obscure the mammography sensitivity to inclusions [1], while such parts also present the highest risk for breast cancer [37, 38]. These earlier works are corroborated by our results showing that g-SoS measured in the UOQ alone correlates strongly with m-ACR assessed throughout the breast. This finding highlights the clinical potential of g-SoS with measurements at a single indicative location (UOQ). These results may also suggest that, even when dense fibroglandular tissue is observed locally in X-ray mammography, the breast SoS might be changing globally across the entire breast, including the UOQ that we observed. Further studies are required to test this and other hypotheses arising from our work, e.g., by increasing the number of measurement locations, such as different quadrants, for broader breast coverage. Combining g-SoS measurements from different quadrants could also further improve classification performance, which is already quite promising with measurements from UOQ alone in this work. Additionally, despite the operator placing the transducer on breast cross-sections without tumors (guided using B-Mode images), there is a possibility that an unnoticed tumor could lead to an artificial increase in g-SoS value. One should also note that m-ACR and g-SoS assess inherently different tissue properties, where the former assesses the visual tissue inhomogeneity, and the latter quantifies the physical value of sound speed.

In this study, we showed that g-SoS is a repeatable, quantitative, and non-ionizing biomarker, that is promising for breast density classification. Evaluating the breast density with g-SoS prior to mammography provides several advantages. The sensitivity of mammography decreases as breast density increases [1]. Therefore, especially for young women with an increased risk, SoS-based density assessment could help determine whether a mammogram is appropriate. If the patient has high breast density, it might then be preferred to include tomosynthesis in the diagnostics or to use US or MRI instead of mammography. Through repeated measurements over time, SoS-based density assessment could help to determine the best moment for a mammogram or MRI, particularly for women using an IUD, undergoing hormonal substitution, and/or without regular periods [39]. Additionally, breast density is an independent risk factor for breast cancer [3, 4]. SoS-based density assessment could be integrated into future risk calculation tools alongside other risk factors, to provide a more precise and personalized breast cancer risk assessment. This could help in primary care; for example, regular SoS-based breast density assessments during gynecological check-ups, combined with other risk factors, may provide improved cancer risk assessment or stratification for X-ray mammographic examination.