Predicting mammographic density with linear ultrasound transducers

Mammographic density (MD) is defined as the proportion of the area of dense regions on a mammogram to the whole area of the breast. Percentage mammographic density (PMD) reflects breast tissue composition, with dense areas appearing lighter than non-dense areas [1, 2]. Women with a high PMD have a higher risk for developing breast cancer (BC) [3‐8]. A case–control study showed high correlation of the absolute dense area (DA) and PMD [7]. The DA was furthermore identified as a BC risk factor [9]. In the context of BC diagnostics, high MD was positively associated with potential masking of BC [10]. No difference in survival between interval cancers and screen-detected cancers for high breast density [11]. Moreover, no association of high MD with risk of death from breast cancer was observed [12]. A recent retrospective analysis added to these results, as no association between PMD and overall survival (OS) was observed despite MD being one of the strongest risk factors for BC [13].

Changes in PMD occur dynamically over the course of a lifetime, the causes of which appear to be multifactorial. MD and age are inversely related: while PMD is higher in premenopausal women, postmenopausal women have significantly lower breast density [14, 15]. In several previous studies, average PMD was not only inversely associated with age, but also with body mass index (BMI) [4, 16‐19].

Changes in breast tissue density are also observed in women exposed to exogenous hormonal influence. While hormone replacement therapy (HRT) is associated with higher density scores [20‐24], endocrine treatment with tamoxifen or an aromatase inhibitor can reduce breast density [25, 26].

MD is inversely associated with parity [27‐30]. Studies have shown that PMD declines with a higher number of pregnancies as well as with a younger age at the first birth [31, 32]. Within about 2 years after the first full-term pregnancy, an average loss of PMD by 12% can be observed [30]. A previous retrospective study assessed the association between the number of full-term pregnancies and PMD relative to age and BMI and observed an inverse correlation between PMD and the number of full-term pregnancies in patients older than 45 years, but not in patients younger than 45 years [31]. Breastfeeding is also associated with reduced breast density [32]. Data from a large analysis of epidemiological studies showed that the relative risk for BC declined by 4.5% per 12 months of breastfeeding [33]. These reproductive factors have been shown to reduce the risk for BC while low parity on the other hand is a considerable risk factor for BC [34]. Considering these factors, it might be possible that a reason for lower BC risk is, in fact, the decline in PMD.

Mammography is the standard-of-care method of imaging in BC screening, presenting a reproducible method which is applied at pre-specified intervals [35]. Mammography screening contributed to a decrease in mortality rates as well as in higher stage BC in Germany [36]. However, there are certain disadvantages to this method, creating the need for alternative assessment tools. While radiation doses in mammography have decreased over the last decades, exposure to such ionizing radiation, especially from repeated mammography may lead to radiation-induced BC [35, 37]. Mammography is, therefore, of limited eligibility outside of routine screening programs, especially with regard to younger women. One image-based approach to measure breast density without the use of ionizing radiation is the assessment via MRI. Studies showed reasonable correlation of MRI with PMD [38‐42]. While MRI-based methods may allow reasonable prediction of MD, they are expensive and of limited availability.

To learn more about PMD and possible informative value about BC risk, it is important to establish other examination methods. In a previous single-center study, we could demonstrate that B-mode ultrasound imaging was associated with PMD in women who underwent routine mammography [6]. Assessing breast density via ultrasound appears to be a time- and cost-efficient method that can be carried out repeatedly on young and/or pregnant women. However, up to date, there is no widely established and validated method for predicting PMD using ultrasound images. Ultrasound systems from different manufacturers could yield different results as image processing and the available formats can vary [43].

The aim of the present work was to assess the correlation of PMD and ultrasound imaging of the breast using a high-frequency (12 MHz) linear ultrasound probe and the VOLUSON^® 730 Expert system (GE Medical Systems Kretztechnik GmbH & Co OHG, Austria) for the prediction of PMD.

MD is a significant risk factor for BC and has, therefore, been a much-studied topic with regard to diagnostics and potential prevention. There is profound evidence that women with a high MD have a higher risk for the development of breast BC [3‐7]. Moreover, high MD is positively associated with potential masking of BC, potentially resulting in later diagnosis and thus unfavorable prognosis [10].

As MD is closely linked to the number of pregnancies, understanding changes in breast tissue composition during and after pregnancies is of importance [4, 53]. One study assessed changes in the breast volume at the beginning and after a pregnancy using a three-dimensional surface assessment technique [54]. Assessing the correlation of changes in breast volume and changes in MD could be of interest in future trials.

Mammography is used as the standard-of-care method of imaging in BC screening in early stages [35]. While radiation doses in mammography have decreased over the last decades, exposure to such ionizing radiation, especially from repeated mammography may lead to radiation-induced breast cancer [35, 37]. Availability of infrastructure and personnel is limited for mammography. The assessment itself is uncomfortable or even painful for the patients, as the breast is compressed to create reproducibility and enhance imaging quality. To sum up, the usage of mammography is of limited eligibility outside of routine screening programs, especially with regard to younger women.

Several other potential methods of measuring MD have been assessed with somewhat promising results. There are also non-imaging methods such as photo acoustics described in literature [55]. However, these approaches are not based on broadly available imaging techniques and are, therefore, of limited eligibility for clinical routine.

One image-based approach to measure breast density without using ionizing radiation is the assessment via MRI. Studies showed reasonable correlation of MRI with PMD [38‐42, 56]. While MRI-based methods to assess breast density may allow reasonable prediction of MD, they have the disadvantage of being somewhat expensive and elaborate and are of limited availability.

Ultrasound imaging of the breast, on the other hand, is a time- and cost-efficient assessment tool. Ultrasound systems are widely available and yield no radiation exposure for the patient. Sonography can be carried out repeatedly and in shorter intervals, depicting a feasible assessment technique for young and/or pregnant women. However, up to date, there is no widely established, validated method for predicting PMD using ultrasound images.

The aim of our previous work was to predict PMD based on sonography and elastography. There, we showed that B-mode images of the normal breast tissue allowed prediction of PMD as assessed via mammography [6]. These results were consistent with other studies [57, 58].

The present work was designed to assess the correlation of PMD and ultrasound imaging of the breast using a high-frequency (12 MHz) linear ultrasound probe in a rather large cohort and examine the predictive value of this imaging assessment method.

Data of 168 patients was evaluated in the course of the present work. Sonography assessment was not fit to support findings of Jud et al. [6]. Ultrasound images showed limited predictive power for PMD when correlated with the corresponding mammograms. R² values were consistently low, with the random forest model M7 turning out to be the most accurate prediction model. This effect could be due to heterogeneity in patient characteristics with regard to several parameters. Patients with a history of BC were included as well as patients without current or past malignancies of the breast. As BC could also have an influence on breast tissue density, this indicates heterogeneity for PMD as well as sonography findings, leading to limited eligibility for validation processes. Moreover, patients were eligible for assessment in the current trial regardless of number of pregnancies and time of breastfeeding. These parameters were collected but had no influence on assessment or sub-analyses.

For future trials, a special focus on a less heterogeneous patient collective, especially with regard to breast cancer history could lead to higher R² results. In a homogenous collective, different prediction models than the random forest model could turn out to be more accurate.

Ultrasound assessments were carried out by different investigators. In our previous work, automated ultrasound image analysis was performed. Inter-observer differences could potentially inflict the validity of assessment and result in deterioration of predictive power.

Differences in image processing between the used ultrasound machines could have potentially influenced outcome results. For example, the device used in the current work has integrated optimizing presets, which does not apply for the Siemens machine (Acuson Antares premium edition, Siemens, Germany) used in our previous work, which assessed raw data [6]. For future trials, this should be taken into consideration. The use of standardized presets could potentially minimize inter-observer bias caused by individually altered imaging settings. The use of raw data could provide unaltered images, if individual use of settings is avoided.

We assessed gray level bins and correlated these to PMD. However, additional texture features could be analyzed to predict PMD and provide different results in future studies.

Overall, the results allowed no reliable prediction of PMD using ultrasound imaging. These results differ from our previous work, where prediction was considerably stronger (R² = 0.67 for B-Mode ultrasound imaging) [6].

Our results did not align with previous studies. No reliable prediction of PMD using ultrasound imaging was observed. Larger studies are needed to assess ultrasound imaging with regard to prediction of mammographic density and eventually breast cancer risk. Identifying feasible non-radiation imaging methods of the breast and their predictive power in young or pregnant women remains an important topic and warrants further evaluation.