Ultrasound Utility for Predicting Biological Behavior of Invasive Ductal Breast Cancers

Determination of the molecular status of invasive breast cancer is useful as a prognostic and predictive factor, and it has become standard practice in the management of breast cancer because estrogen receptor (ER) and human epidermal growth factor receptor2 (HER2) positivity predict response to endocrine therapy or targeted therapy with monoclonal antibodies directed against HER2 (Bauer et al., 2007; Doreen et al., 2011). If it is possible to predict molecular status on the basis of imaging characteristics, it could assist in both pretreatment planning and prognosis, as well as add to our understanding of the biologic behavior of this disease. Breast ultrasound has gained widespread acceptance as an adjunct to mammography in diagnosis of evaluating clinical or radiological suspected abnormalities (Gordon et al., 1995; Rizzatto et al., 2001). Stavros et al. reported that it has high sensitivity (98.4%) and negative predictive (99.5%) value for diagnosing breast cancers (Stavros et al., 1995). Ultrasound (US), with its merits of safety and low cost, is becoming a preferred method for both physicians and patients. Hence, more attention is needed toward US imaging to determine whether certain type of tumor biologic factors can be predicted from imaging appearances. A few studies have looked into correlation between


Introduction
Determination of the molecular status of invasive breast cancer is useful as a prognostic and predictive factor, and it has become standard practice in the management of breast cancer because estrogen receptor (ER) and human epidermal growth factor receptor2 (HER2) positivity predict response to endocrine therapy or targeted therapy with monoclonal antibodies directed against HER2 (Bauer et al., 2007;Doreen et al., 2011).If it is possible to predict molecular status on the basis of imaging characteristics, it could assist in both pretreatment planning and prognosis, as well as add to our understanding of the biologic behavior of this disease.
Breast ultrasound has gained widespread acceptance as an adjunct to mammography in diagnosis of evaluating clinical or radiological suspected abnormalities (Gordon et al., 1995;Rizzatto et al., 2001).Stavros et al. reported that it has high sensitivity (98.4%) and negative predictive (99.5%) value for diagnosing breast cancers (Stavros et al., 1995).Ultrasound (US), with its merits of safety and low cost, is becoming a preferred method for both physicians and patients.Hence, more attention is needed toward US imaging to determine whether certain type of tumor biologic factors can be predicted from imaging appearances.
A few studies have looked into correlation between

Ultrasound Utility for Predicting Biological Behavior of Invasive Ductal Breast Cancers
Lei Zhang, Yu-Jie Liu, Shuang-Quan Jiang, Hao Cui, Zi-Yao Li, Jia-Wei Tian* ultrasound features and certain types of biologic behavior (Kim et al., 2008;Wang et al., 2008;Au et al., 2009;Ko et al., 2010;Irshad et al., 2013).Some studies also have discussed the features of triple-negative breast cancers (negativity of ER, PR, and HER2) (Dogan et al., 2010;Ko et al., 2010;Dogan et al., 2012;Krizmanich et al., 2012;Wojcinski et al., 2012;Li et al., 2014).However, correlations between ER, PR, HER2 and ultrasound findings in previous studies were varied and none had calculated the logistic regression model for reliable identification of molecular status (Kim et al., 2008;Wang et al., 2008;Au et al., 2009;Dogan et al., 2010;Ko et al., 2010a;2010b;Dogan et al., 2012;Krizmanich et al., 2012;Wojcinski et al., 2012;Irshad et al., 2013).The aim of this retrospective study was thus to 1)evaluating the correlation of images findings with the corresponding molecular features, 2)characterizing the various ultrasound features in breast cancer by using logistic regression models, in order to identify the factors that might help in predicting the status of molecular.

Patients
Our study group consists of 357 consecutive patients with invasive breast cancer, performed surgery in the Second Affiliated Hospital of Harbin Medical University between Dec. 22, 2011 andJan. 20, 2013, were initially detected by breast ultrasound examinations.25 of the 347 patients were excluded because of treatment with neoadjuvant, 48 were excluded due to the failure of surgery, 11 (have the lesions too deep) were excluded owing to indistinct ultrasound imaging and 10 were excluded on account of multiple lesions.Finally, 263 women with definite histological results were evaluated in the study.

Ultrasound examination
To keep the consistency of the imaging, all the realtime scanning was performed by one radiologist with 4 years experience in breast ultrasound, using HITACHI Vision 500 system (Hitachi Medical System, Tokyo, Japan) equipped with a linear probe of 5-13 MHz.The static images and cine clips of B-mode and elastograms, which contain transverse and a sagittal plane, were saved in the database for double-blind analysis.The ultrasound criteria were according to the Breast Imaging Reporting and Data System (BI-RADS) and the elasticity score criteria proposed by Itoh et al (Itoh et al., 2006).The characteristics considered were shown in Table 1.In this study, ''Not circumscribed'' margins were defined when the margin was indistinct, spiculated, angular, or microlobulated.Two breast radiologists with respectively 9, 13 years of clinical experience retrospectively and independently reviewed the ultrasound images.A consensus interpretation was reached in the cases of disagreement.

Histological examination
All resected tumors were stained with hematoxylineosin (HE) and performed in formalin-fixed, paraffinembedded material for pathological diagnosis.Each sample was routinely tested for ER, PR, and HER2 with IHC.The cutoff point for ER-and PR-positive expression level was 1%.For HER2 expression, staining intensity was divided into four grades, with grades 0 and 1 considered as negative, grade 2 as determinate by fluorescence in situ hybridization, and grade 3 as positive.

Statistical analysis
To test the primary hypothesis that ultrasound features are associated with molecular status, univariate logistic regression models were performed and expressed as odds ratio (OR) with 95% confidence intervals (CI).In order to find the best combination of ultrasound-based indicators of molecular, multivariate model were built by means of logistic regression analysis.The stepwise regression method was used to select the parameters which were included in the final models.For a more rigorously screening, the selection of SLE and SLS were all 0.05.The parameters in the final model are independently associated with the molecular.The Hosmer-Lemeshow goodness-of-fit test was used to evaluate the overall fit of the final models.Discrimination and classification of each model or predictor were assessed using the concordance statistic (an approximation of the area under the receiver operating characteristic curve [AUC]).Inter-observer variability was assessed with the Cohen's kappa statistics (Svanholm et al., 1989).All p-values are two-tailed and p -Values lower than 0.05 were considered statistically significant.All statistical analysis was performed using the SAS System, version 9.2.

The model is as follows: p=[1÷ (1+e -z )]
Where e is a base of natural logarithm, =2.71828..., and z is a linear combination of x i variables and their estimators bi included in the model:

Features
The study group comprised 263 patients for whom ultrasound images and molecular results were available and whose mean age at presentation was 50.63±9.90years (range 22-76 years).The mean size of the lesions was 2.33mm (range 1-10.6 cm).The ultrasound findings are summarized in Table 1.

Observer agreement
Cohen's kappa statistics showed that the better interobserver agreement was obtained.Level of inter-observer agreement were between 0.61~0.80,which means substantial agreement.

Univariate and multivariate analysis
Assessing the correlation between ultrasound features and biological markers, we found that differences in ultrasound criteria were most pronounced among women with different molecular status.Statistically significant results of the univariate and multivariate regression models, which was established using stepwise regression (SLE=0.05,SLS=0.05),comparing ultrasound parameters of 263 invasive cancers are given in Tables 3-5.
The ultrasound features of PR-positive breast cancers are similar with ER-positive breast cancers, except posterior tumor shadowing and low elasticity score, which had no statistical significance of PR positive cancers.The size of PR positive cancers was 21.2mm, whereas the size of PR negative cancers was 26.5mm (p=0.006,AUC=0.66).A higher percentage of lobulate (60.2%, p=0.0001) and angular (32.3%, p=0.007) margin contour than smooth margin (7.5%) were noted in PR positive cancers (AUC=0.61).PR positive cancers were less likely to be associated with calcification (pos versus neg, 41.6% versus69.6%,p=0.0189,AUC=0.58).

AUC analysis
The final logistic models were established by the statistically significant results of the univariate and multivariate regression models.These models, which were established using stepwise regression (SLE=0.05,SLS=0.05), for predicting molecular status as follows:  The Hosmer-Lemeshow goodness-of-fit test showed value of 3.16 (p =0.92), 6.30 (p=0.61) and 0.72 (p=0.70),respectively, which mean these models fitted well.The results of receiver operating curves (AUC) are shown in Figure 1-3.The performances of the model that represent ER, PR, and HER2 status were good with AUC of 0.78, 0.74, and 0.74, respectively.

Discussion
Clinically, breast cancer is a molecularly heterogeneous disease that has been categorized into three basic therapeutic groups.The ER positive group is the most numerous and diverse, with several genomic tests to assist in predicting outcomes for ER patients receiving endocrine therapy (Paik et al., 2004).The HER2 amplified group is   (Slamon et al., 1987).Triple-negative breast cancers (TNBCs, lacking expression of ER, PR and HER2) are a group with only chemotherapy options (Perou, 2011).Hence, these biological markers (ER, PR, and HER2) can applied to prediction of clinical response to medical treatment and prognosis.This retrospective study revealed that the three biological markers correlating with the ultrasound findings and established three logistic regression models on prediction of molecular status.
From this article, it was realized that the efficiency of single feature to estimate the molecular status was low.However, the efficiency was substantially improved by combination of the ultrasound features.Three models to predict the ER, PR, HER2 status based on patients' pathologic data were developed from the multivariate logistic regression models (SLE=0.05,SLS=0.05), which showed AUC of 0.78, 0.74, and 0.74 respectively.In our study, we found that ER positive was associated with small size, lobulate or angular or spiculated margin contour, absence of calcification, posterior tumor shadowing and low elasticity score; PR positive was associated with small size, lobulate or angular or spiculated margin contour and absence of calcification; HER2 positive was associated with presentation of calcifications and absence of echogenic halo.
Posterior shadowing is an important breast ultrasound criterion.Zonderland. (2000) reported that posterior shadowing is known as a parameter of a moderate power to differentiate between benign and malignant lesions.It has been known that shadowing is provided by a highly cellular fibroblastic proliferation, which more likely existed in low grade tumors.ER positive, which more likely be low grade, was associated with shadowing.This feature is in agreement with previous reports, Irshad.(2013) compared the ultrasound features of ER-negative/PR-negative and ER-positive/PR-positive cancers and concluded that the presence of the posterior shadowing was found to be a very strong predictor of a receptor ER-positive tumor (97% were receptor positive).Study of Ko ES, et al (Ko et al., 2010) showed that triple-negative breast cancer were less likely to be associated with posterior shadowing.Similar to other study, in our study posterior acoustic was more commonly posterior shadowing (27.7%) and less commonly posterior enhancement (16.0%) in ER positive cancers, which also demonstrated that a significant correlation between posterior tumor shadowing and ER status.The performances of posterior tumor shadowing predicted ER status with AUC of 0.66.Margin contour is another important ultrasound criterion.Correlation was noted between the angular or spiculated and low tumor grade that may be explained by the greater desmoplastic reaction in these leading to noncircumscribe margins (Irshad et al., 2013).Wang. (2008) noted that while they found an association between spiculated margins and HER2 status among patients with ER-negative, ER-positive tumors can also manifest as spiculated masses.In addition, IIdefonso, et al (IIdefonso., 2008) showed that 63% of spiculated masses were ER positive.Similar as our results, circumscribe margin were more frequent in ER or PR negative cancers and lobulate or angular or spiculated margin contour were more frequent in ER or PR positive cancers (93.1%).The performances of margin contour predicted ER status with AUC of 0.66.
Our study results show that ER/PR positive cancers were much more likely to be associated with small size and low elasticity score.Marquet, et al (Marquet et al., 2002) found a statistically significant correlation between tumor size and posterior shadowing, and shadowing was more exited in ER positive breast cancer.Hence, similar to our result, ER/PR positive cancers were tended to be small.Some studies found that elastography was useful in diagnosing breast lesions in the clinical setting (Itoh et al., 2006;Parajuly et al., 2010;2012).The higher elasticity score, the greater stiffness.This was also in agreement with our report that ER/PR positive cancers were tended to be low elasticity score.The performances of size and elasticity score predicted ER/PR status with AUC of 0.68, 0.51, respectively.
Assessment of HER2 positivity is important for the establishment of a treatment plan and the prediction of prognosis in patients with primary breast cancer (Taucher et al., 2003).Our results are consistent with prior studies (Kim et al., 2008;Ko et al., 2010) , which reported that expression of the HER2 oncogene strongly correlated with presence of calcification on ultrasound or mammography that may be related to prognosis.In addition, we found that the ultrasound findings among HER2 positive cancers were most commonly a mass without echogenic halo, which is defined as one of back scattering in US.The performances of calcification and echogenic halo predicted HER2 status with AUC of 0.74.
The results of our study might have some applications.Some sophisticated laboratory tests are not readily available or not cost-effective in many part of the world and biopsy is invasive and may cause physical and psychological discomfort in patients.On the other hand, ultrasound is a relatively inexpensive, easily operable, widely accessible tool, and recent advances in ultrasound technology, transducer design permit greater spatial, contrast resolution.It is now a key mode of imaging for the clinical diagnosis of breast cancer.If certain ultrasound features of breast cancer could predictive the biological behavior, it would lead to the recommendation to perform biopsies in the lesions of unknown character more frequently.Furthermore, our results could be useful for the implementation of a diagnostic ultrasound criterion for the breast cancer to indicate the probability status of the molecular, which similar to BI-RADS category as proposed by the American College of Radiology.Such an instrument does not exist which need a long exploration process.It will be assisted in both pretreatment planning and prognosis, as well as add to our understanding of the biologic behavior of this disease.
Our study has the following limitations.First, our study is a retrospective study with small sample size, which needs a large population to confirm our results.The statistical significance of these findings may be insufficient.Second, we did not correlate with cancer stage at diagnosis and did not analyze the incidence of associated DCIS.
In conclusion, ultrasound pattern is correlated with biological markers in breast cancers.Ultrasound features of small size, lobulate or angular or spiculated margin contour, absence of calcification, posterior tumor shadowing and low elasticity score can be used to predict ER status; ultrasound features of small size, lobulate or angular or spiculated margin contour and absence of calcification can be used to predict PR status; ultrasound features of presentation of calcifications and absence of echogenic halo can be used to predict HER2 status.These results can assist in pretreatment planning and prognosis adding to a greater understanding of biological behavior.

Figure
Figure 1.Graph Shows the Average ROC Curves of Predicted ER Status