Predicting Molecular Subtypes of Breast Cancer with Mammography and Ultrasound Findings: Introduction of Sono-Mammometry Score

We studied the correlation of sonographic and digital mammographic features with molecular classification of breast cancer. Imaging features from 313 patients with preliminary ultrasound and digital mammogram between November 2017 and May 2020 were compared with histopathology and immunohistochemical analysis for the prediction of molecular classification of breast cancer. We also devised a score called “sono-mammometry” score consisting of few simple imaging features which can easily be performed in outpatient settings. We studied that non-triple-negative breast cancers are predominantly hypoechoic and strongly correlate with the presence of irregular spiculated margins along with peripheral echogenic halo, posterior shadowing, and microcalcifications, while there is considerable variation in imaging features of TNBC as some of its imaging features overlap with those of typical benign tumors. Although imaging characteristics are helpful in the prediction of molecular classification, the prognostication value of these imaging features is still weak. There is considerable variation in imaging features which warrants vigilance towards improved diagnostic performance. To help better understand these features, our sono-mammometry score can serve as straightforward test which is assumed to be functional and productive in resource-limited settings.


Introduction
Breast cancer is the most common cancer in women and is the second most common cause of death from cancer in women worldwide [1]. It is a diverse group of diseases which consists of a wide range of molecular and genetic subtypes [2,3]. During the last two decades, regrouping of breast cancer classification has been undertaken, from histopathologic subtype to the molecular categorization established by gene expression [1][2][3][4].
Breast imaging has a dominant role in the detection, staging, and follow-up of patients with breast cancer. ere is a substantial effect on clinical outcome from early recognition of breast cancer [5]. Although histopathological characterization is the gold standard for classifying molecular subtypes of breast cancer after percutaneous breast mass biopsy, the trend is moving towards the development of noninvasive diagnostic procedures that can aid in easy and quick interpretation of the disease process [6]. is highlights the importance for a radiologist to have sound knowledge of these molecular subtypes of breast cancer to improve the analysis of imaging findings. e assessment of molecular subtypes of breast cancer on imaging perhaps displays intricacies as it requires the combination of imaging features on both ultrasound and mammography. e purpose of our study is to identify the relationship of immunohistochemical markers with multimodality imaging features using mammogram and ultrasound. MRI breast was not available in our department. We also determined the association of baseline histopathological type, grade of tumor, and T and N stages of the breast cancers with each luminal classification. We also proposed a scoring system that is assumed to provide better insight into understanding imaging features of non-triple-negative breast cancer (non-TNBC) in contrast to triple-negative breast cancers (TNBC). Our hypothesis is that different sonographic and mammographic features can predict molecular classification on histopathology.

Study Population.
Our study was conducted at the Radiology Department of Sindh Institute of Urology and Transplantation (SIUT). Our Institutional Ethics Committee waived the requirement of individual informed consent for this descriptive retrospective study.
e study group was selected from both the screening and symptomatic populations. Patients with preliminary ultrasound, mammogram, and histopathological analysis done at our institution were regarded as being suitable for our study. Relevant information was gathered from ultrasound and mammogram images. Patients with incomplete records and unsatisfactory images and ductal carcinoma in situ (DCIS) were excluded from the study. Bilateral breast cancers were also excluded from the study because in our record we found only 5 such patients out of which in 3 patients we did not find satisfactory imaging. Very large inflammatory breast cancers were also excluded because these usually cause obscuration of fine details of ultrasound images of malignant masses secondary to excessive breast edema; however, locally advanced diseases in which breast edema was not present and we can comprehend ultrasound characteristics of malignant masses were included in the study. In this way, data from 313 patients with breast cancer were retrospectively collected between November 2017 and May 2020.

2.2.
Ultrasound. All ultrasound examinations were done according to ACR practice parameters for the performance of a breast ultrasound examination (revised 2016) [7]. ese were acquired according to the standard of care protocol as a routine practice of our breast cancer unit and these were not carried out under research study settings. For sonography, 2 scanners were used, both were Canon Xario 200, each with frequency of 14 MHz. Every sonographic examination included both breasts and was extended to involve both axillary regions.

Digital Mammography.
All mammograms were performed according to ACR practice parameters for the performance of screening and diagnostic mammography (revised 2018) [8]. All mammograms were done with Selenia Dimensions 3D Digital Mammography Tomosynthesis System (Hologic, Bedford, MA, USA). Images were reviewed on high-resolution workstations at the Radiology Department of our institute. Standard 2 views (craniocaudal view and mediolateral view) were performed routinely for all patients. Supplementary views were done whenever required.

Ultrasound and Digital Mammogram Assessment.
All ultrasound and mammogram images were reviewed by 1 of 2 senior consultants in breast imaging, both of whom had at least 5 years of experience in breast imaging. Both radiologists were blinded to the patient's clinical findings and histopathological results. In the case of disagreement, a consensus was reached after discussion. e sonographic and mammographic features of the breast carcinomas were assessed based on the analytical criteria of Breast Imaging Reporting and Data System (BIRADS). For ultrasound, these features include tumor size, shape, orientation, boundary, margins, echo texture, and posterior features. For the mammogram, the features that are assessed are mass shape, margins, presence or absence of suspicious microcalcifications, and architectural distortion/trabecular thickening. Additional features like T and N stages and multifocality were also assessed.

Histopathological Examination.
e histopathological specimens from the core needle biopsy were formalin-fixed and paraffin-embedded tissue blocks afterwards stained with H&E (hematoxylin and eosin). ese samples were assessed by an experienced pathologist for tumor type and grading. Invasive cancer was graded as grade 1 (well differentiated), grade 2 (moderately differentiated), or grade 3 (poorly differentiated) according to the Scarff-Bloom-Richardson system [9]. Subsequently, immunohistochemical analysis was done to assess ER, PR, Ki-67 index, and Her-2 neu expression. ER and PR were regarded as positive if at least 1% of the tumor nuclei were positively stained [10]. An additional fluorescence in situ hybridization (FISH) was analyzed to detect Her-2 positivity with scores of 2 or higher. Scores of 1 or 0 were defined as Her-2 negative [10]. Ki-67 index >14% was considered as high and <14% was considered as low expression [5]. (Table 1). We combined ultrasound and mammogram imaging features to design a scoring system for prediction of non-TNBC and TNBC. We used bivariate Chi-square test and multivariate binary logistic analysis to select variables. Variables with p values < 0.05 were considered significant. Factors significantly associated with univariable and multivariable logistic regression analysis are listed in Tables 2-4. Results of the univariable and multivariable logistic regression analysis were shown as p values, odds ratio (OR), and corresponding 95% CI. To estimate the discriminating power of the scoring system, the receiver operator characteristic (ROC) curve ( Figure 1) was created and the areas under the curves (AUCs) were calculated.

Sono-Mammometry Score
By multivariate logistic regression analysis, seven variables on sonography and mammogram imaging were significantly related to predict non-TNBC from TNBC and were assigned scores for the final prediction. ese are summarized in Table 5.
ese variables were as follows:     is scoring system is not applicable in the absence of a sonographically and mammographically detected mass; therefore, score 0 was assigned to the presence of pleomorphic microcalcifications on mammogram without coexistent malignant mass on ultrasound and mammogram. Echogenicity was not included in the scoring system because in our study we did not find a significant difference among the echogenicity of non-TNBC and TNBC groups. e minimum and maximum scores were 6 and 14, respectively. e lower value of the score anticipates the presence of TNBC. In speculating TNBC, ROC curve analysis was performed for this sono-mammometry score, which showed significant statistical association (p � < 0.01). AUC was demonstrated to be 0.719 with 95% CI of 0.645-0.792. At score 10.5, sensitivity is 94%, specificity is 47%, positive predictive value is 75%, and negative predictive value is 84%.
In our study, the majority of tumors (n � 202, 64.55%) showed peripheral echogenic halo, among which the most common was luminal A (n � 56, 90.3%) (p � < 0.01). With regard to orientation, the preponderance of malignant masses showed antiparallel orientation to the chest wall (n � 297, 94.9%) (p � 0.03). (Table 4). Mammography findings were normal in 7 (2.2%) patients and (n � 7, 2.2%) patients with focal asymmetry. In all these patients, ultrasound showed noncalcified mass. e most common mammographic feature was the presence of a mass without microcalcification (n � 197, 62.9%) followed by a mass with microcalcification (n � 92, 29.4%). Most common noncalcified tumors showed correlation with HR   (Table 5). We also performed an analysis on combined imaging findings. Breast malignancies with round/oval shape and well-circumscribed/microlobulated margins were strongly associated with the absence of tumor

Discussion
We studied the correlation of sonographic and digital mammographic features with molecular classification of breast cancer and also devised a score called sono-mammometry score to anticipate the presence of non-TNBC tumors. We also determined the association of baseline histopathological type, grade of tumor, Ki-67, and T and N stages of the breast cancers with each luminal classification. We used the term "Her-2 enrich" for tumors that are hormone receptor negative and Her-2 neu +ve, while the term "Her-2 +ve" is used to indicate both luminal B-2 and Her-2 enrich tumors collectively. We studied that non-triple-negative breast cancers are predominantly hypoechoic, less commonly complex in the echo pattern.
ese tumors strongly correlate with the presence of irregular spiculated margins along with peripheral echogenic halo and presence of posterior acoustic shadowing. Architectural distortion and trabecular thickening are also found frequently in this same subset. Presence of microcalcifications also correlates significantly with non-TNBC.
Spiculations in the margin are established criteria in the diagnosis of malignancy. Spiculated margins are generally secondary to drawing in of Cooper's ligaments into a tumor mass or the invasion of tumor cells into the neighboring breast tissues [11]. ese irregular margins are believed to be associated with slow multiplication of tumor cells, which gives enough time for stromal interactions and induces fibrosis surrounding the invasive edge [11,12]. Liu et al. [11] reported that masses with spiculated margins were significantly more common in patients with luminal A breast cancer than in those with other subtypes; however, we studied that HR-positive tumors (all non-TNBC) including Her-2 +ve luminal B and Her-2 enrich cancers presented with predominantly spiculated margins and we also deduced that because of this same reason non-TNBC tumor more often causes architectural distortion/trabecular thickening as secondary sign of malignancy. In our study, multivariate logistic regression analysis also determined the convincing association of irregular spiculated margins with Her-2 positivity. is is opposed to Liu et al.'s study [11] which determined that Her-2 overexpressing cancer almost never presents as spiculated masses. On the contrary, this is similar to the studies done by Lee et al. [3], Cho [4], and Elias et al. [13] who stated that circumscribed margins decreased the chance of HER-2 overexpression.
Lacroix et al. [14] determined that grade 1 and grade 2 tumors result in spicules and perilesional hyperechogenic halo. Our study supports this finding because, as discussed above, irregular margins were frequently associated with non-TNBC and this is the same subgroup that is more often related to low-grade (grade I and II) tumors. We also assume that peripheral echogenic halo which dictates the presence of desmoplastic reaction [10] can also be found routinely in this same luminal A, B and Her-2 enriched subgroups.
In invasive carcinomas, the unregulated and disorganized growth of malignant cells generates different tissue layers with variable acoustic impedance, which in turn causes an increase in the attenuation of the sound beams.
is, along with fibrosis secondary to stromal interaction, results in posterior acoustic shadowing in sonographic images [10,12]. We found a significant correlation of both HR-positive and Her-2 enriched tumors with the presence of posterior acoustic shadowing (Figures 2(a)-2(c)).
It is imperative that the presence of microcalcifications at mammography is not definite for any breast cancer subtype [3]. ese are more likely encountered in Her-2 +ve and HRpositive tumors [3]. In our study, we have divided luminal B cancers into two subgroups to scrutinize the impact of Her-2 positive status on imaging features. Multivariate logistic regression analysis demonstrates a convincing interconnection of microcalcifications (with or without the presence of a malignant mass) with Her-2 +ve disease. Cen et al. [15] stated that Her-2 enriched tumors were more expected to have heterogeneous and pleomorphic microcalcifications on mammogram. Likewise, Elias et al. [13] and Lee et al. [16] found that fine, pleomorphic or fine linear branching calcifications are the most characteristics findings for Her-2 +ve cancers. It is noteworthy that Her-2 +ve tumors belong to the subgroup that can be missed with ultrasound because of the presence of microcalcifications only [5]. In our study, two cases, in which ultrasound did not show any mass, had microcalcifications on mammogram and belong to Her-2 +ve subset (Figures 3(a)-3(c) and Figures 4(a)-4(c)).
TNBC has been known to be associated with considerable differences in clinical, radiological, and pathological features compared with its counterpart [12,17]. is is the subtype most discussed in the literature and it is the most commonly identified subgroup in our study. TNBC are also hypoechoic or show a complex echo pattern. We studied that TNBC can less commonly share imaging features similar to non-TNBC, i.e., mass with irregular margins and posterior shadowing. Imaging features that are distinctive to TNBC are similar to those that are also peculiar to benign tumors like noncalcified masses with well-circumscribed or microlobulated margins with posterior acoustic enhancement and more often with an abrupt boundary. Our findings were supported by studies by Lee et al. [3] and Lin et al. [12] who determined that benign "pseudofibroadenoma" type features can often be seen in TNBC. On sonography, the smooth well-circumscribed and microlobulated margin in TNBC is termed as "pushing border" microscopically and is considered secondary to rapid growth and high proliferation rate of malignant cells which leads to the lack of both stromal interaction and fibrosis. Cooper ligaments are believed to be displaced and not disrupted significantly in TNBC [10,12]. Identical to the benign masses, orderly and nestled growth of tumor cells in TNBC create fewer layers that lead to improved enhanced through transmission [10]. In addition, Radiology Research and Practice 7 these tumors show a tendency towards being high grade and are more cellular and therefore display posterior acoustic enhancement more frequently than non-TNBC [12,17]. Peripheral echogenic halo is believed to be less often found in TNBC. Our findings were also supported by studies done by Lacroix et al. [14] and Wojcinski et al. [10] who also found abrupt tumor parenchyma interface more frequently in these tumors. TNBC are also known to lack the presence of suspicious microcalcifications on mammogram [3,18] ( Figures 5(a)-5(c)). However, Lin et al. [12] believed that there are wide variations in imaging features for TNBC. In our study, we postulate that close resemblance of its imaging features with benign tumors warrants improving diagnostic performance for early suspicion and recognition of malignancy, particularly in outpatient clinics where prompt and precise decisions are required.
Although there are some characteristics imaging features that can be attributed to each luminal category, overall there is considerable overlap in imaging features of these 5 groups.
erefore, based on our results, we divided these cancers into two groups, non-TNBC and TNBC, and devised a score called "sono-mammometry" score (Table 5)  8 Radiology Research and Practice performed in outpatient settings. Score of 10.5 shows sensitivity of 94%. Its low specificity of 47% is a limitation which warrants validation of this score on a larger number of patients before future implication. However, this score is still useful, particularly in a developing country where resources are limited. Although histopathological diagnosis providing us with indispensable information regarding tumor biology and histological interpretation cannot be ignored, all laboratory tests have the incidence of false results [11]. It highlights the importance of a radiologist to be well-aware of common and uncommon imaging features, so a question can be raised in the matter of conflicting lab results and perhaps it may be beneficial to repeat testing of the immunohistochemical receptors. It is worthwhile to discuss the advancement of artificial intelligence with regard to medical imaging. In our study, human readers are involved in anticipating results with the help of simple features on mammogram and ultrasound; however, there are studies which have incorporated newer imaging techniques, e.g., Dilorenzo et al. [19] has shown the potential role of background parenchymal enhancement in MRI breast in discriminating different breast cancer subtypes. Fusion of ultrasound and MRI images also proved to provide information about additional occult breast lesions [20]. State-of-the-art techniques assimilate radiomics in discerning benign and malignant breast lesions on mammography [21,22]. Different variety of machine learning techniques has also been developed and integrated for early detection of breast cancer through textural analysis of microcalcification on mammogram and for further classification of breast cancers [23][24][25]. Contrast-enhanced spectral mammography is a recently introduced mammographic method in which Forgia et al. have (c) Ultrasound of the same patient shows a complex mass with partly ill-defined and partly irregular anterior margins. Abrupt boundary is seen with parallel orientation of this mass with the chest wall (wider than taller). No posterior features. Sono-mammometry score: 11. discussed imaging features of two cancer molecular subtypes, i.e., HER-2 positive and triple-negative, with the help of radiomics [26]. Although radiomics is an evolving methodology which has shown to provide insightful results in quantitatively and objectively elucidating tumor biology, there are issues of standardization and lack of reporting guidelines. Furthermore, in economically developing and resource-constrained environment, these advanced engineering techniques are still seemed to be far-fetched and human readers with manual interpretation of images take precedence. Our results should be comprehended after taking into account the limitations. First, it is a retrospective study with a small number of patients. Prospective studies with larger numbers of patients are needed to validate these results. Sono-mammometry score should be applied in a prospective population reassessment to see how predictive the model obtained is. Another is that breast density was not taken into account. Moreover, our analysis was carried out on selected images rather than images acquired during real-time scanning and operator-dependent nature of ultrasound can give rise to biases.

Conclusion
Breast cancer exhibits different imaging features according to molecular type. Although imaging characteristics are helpful in the prediction of molecular classification, the prognostication of these features is still weak. ere is considerable variation in the imaging features of TNBC as some of its imaging features overlap with those of typical benign tumors, which warrants vigilance towards improved diagnostic performance for early suspicion and recognition of malignancy. To help better understand these features, our sono-mammometry score can serve as a relatively inexpensive and straightforward test which assumes to be functional and productive in resource-limited settings. Our next step would be to apply this score in prospective study to assess its predictive value.

Data Availability
e data used in the study are available from the corresponding author (Sana Shaikh) upon request.

Conflicts of Interest
e authors declare that they have no conflicts of interest regarding the publication of this paper.