Prospective Evaluation of Contrast-enhanced Mammography for Early Prediction of Pathologic Response after Neoadjuvant Therapy

Abstract

Purpose

To assess whether changes in contrast-enhanced mammography (CEM)–derived lesion measurements after the first cycle of neoadjuvant therapy (NAT) can predict pathologic complete response (pCR) in individuals with breast cancer.

Materials and Methods

This prospective single-center pilot study enrolled consecutive participants with breast cancer treated with NAT who underwent CEM at baseline (May 2018 to December 2018). CEM was performed before and after the first cycle of NAT. Two breast radiologists independently evaluated the percentage change in the longest dimension of the lesion (CLD) and Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 criteria at CEM. Multivariable logistic regression was used to identify independent predictors of pCR, and predictive performance was assessed using area under the receiver operating characteristic curve (AUC).

Results

Thirty-six participants (mean age ± SD, 48 years ± 10.3) were included; 11 (30.5%) participants achieved pCR. A CLD of at least 20.93% independently predicted pCR (odds ratio, 9.52; 95% CI: 1.34, 67.23; P = .02), achieving a sensitivity of 73% (eight of 11) and a specificity of 88% (22 of 25). Response according to RECIST 1.1 criteria was not associated with pCR (odds ratio, 3.22; 95% CI: 0.46, 22.53; P = .24). In participants with hormone-receptor negative breast cancer, a CLD of at least 20.93% was associated with a higher likelihood of pCR (odds ratio, 40.00; 95% CI: 2.01, 794.27; P = .005) and had an AUC of 0.86 (95% CI: 0.65, >0.99; P = .005).

Conclusion

CLD at CEM after the first cycle of NAT may be an early predictor of pCR in individuals with breast cancer.

Keywords: Breast, Tumor Response, Mammography, Oncology, Neoadjuvant Therapy, Radiographic Image Enhancement, Pathologic Complete Response, Breast Tumor

Supplemental material is available for this article.

© RSNA, 2025

Summary

Changes in lesion size measurements at contrast-enhanced mammography after the first cycle of neoadjuvant therapy may be an early predictor of pathologic complete response in individuals with breast cancer.

Key Points

• ■ In this prospective study of 36 individuals with breast cancer undergoing neoadjuvant therapy, percentage change in the longest dimension of the lesion (CLD) at contrast-enhanced mammography after the first cycle of neoadjuvant therapy was an independent predictor of pathologic complete response (pCR) (odds ratio, 9.52; P = .02).

• ■ A CLD cutoff of 20.93% had a sensitivity of 73% and a specificity of 88% for predicting pCR.

• ■ Among participants with hormone receptor–negative breast cancer, high CLD (≥20.93%) was associated with higher likelihood of pCR (odds ratio, 40.00; P = .005).

Introduction

Neoadjuvant therapy (NAT) as a therapeutic option for patients with breast cancer has gained increasing attention (1,2). NAT offers similar long-term distant and local-regional cancer control to adjuvant chemotherapy (3,4) with the advantage of increasing the probability of breast and axilla–conserving surgery (1,4,5). Additionally, NAT provides a unique opportunity to monitor in vivo tumor response. Pathologic complete response (pCR) is a well-established surrogate marker for extreme chemotherapy responsiveness and a short-term goal of treatment (3,4,6). Although pathology measurements of residual disease are established predictors of survival after NAT, they are not suitable as early indicators. In current clinical practice, the response to NAT serves as a prognostic marker and aids in decisions of adjuvant treatments (3,4,7).

With increasing use of precision medicine, the need for early predictors of response gains relevance. The early identification of poor responders or nonresponders could spare patients and society the physical, psychologic, and financial burden of potentially ineffective and toxic treatments. Stratification into subgroups of responders could allow for personalized treatment, optimizing the trade-off between therapeutic gains and iatrogenic hazards in both responders and nonresponders.

Imaging allows in vivo monitoring of NAT and is currently used to evaluate residual disease for surgical planning (8,9). Previous studies have explored imaging techniques as early indicators of response, namely, MRI and PET/CT (10–16).

Contrast-enhanced mammography (CEM) is an emerging morphofunctional imaging technique that is gaining increasing interest and acceptance in clinical practice. CEM generates high-resolution mammographic images that detail morphology as well as recombined images (iodine map) that depict perfusion, similarly to MRI. Previous studies have shown that contrast-enhanced MRI is superior to physical examination, mammography, and US to assess the extent of residual disease after completion of NAT (8,9). Recent studies have explored CEM performance for NAT response assessment with encouraging results (17–22). In comparison with MRI, CEM is faster, less expensive, more patient-friendly, potentially more available, and has fewer contraindications (23,24).

Presently, CEM is considered an alternative for NAT response evaluation, when MRI is contraindicated, unavailable, or has limited access. The role of CEM as an early biomarker of response to NAT has not previously been established in the literature, to our knowledge. We hypothesized that CEM can be an early predictor of response in patients with breast cancer undergoing NAT. Thus, the purpose of this study was to assess the performance of changes in lesion size measurements at CEM after the first cycle of NAT for predicting pCR in individuals with breast cancer.

Materials and Methods

Study Sample

This prospective single-center study was approved by our institutional ethics committee (CES: 43/2018). Written informed consent was obtained from all participants. The study was conducted in accordance with Declaration of Helsinki and complied with the Standards for Reporting of Diagnostic Accuracy Studies (25). This study was part of a research project running from May 2018 to October 2019. This research project also aimed to compare CEM and MRI for NAT response assessment, regarding diagnostic performance after NAT and participants’ preference and tolerance, which have been previously reported. The data and the analysis presented herein are independent of those previously reported, with no overlap.

From May 2018 to December 2018, all consecutive patients of our tertiary cancer center, with histologically proven breast carcinoma and indication for NAT who underwent CEM at baseline were considered for inclusion in this study. Diagnosis of breast carcinoma was confirmed with core needle biopsy. The therapeutic pathway of each participant was independently decided at the Breast Unit Multidisciplinary Group of our institution. Exclusion criteria were cancers not depicted by CEM or MRI or known contraindications for either of these imaging techniques. Figure 1 illustrates the flowchart of participant inclusion and exclusion. Clinical, pathology, and imaging data were collected.

Figure 1:

Flowchart of participant inclusion and exclusion. CEM = contrast-enhanced mammography, NAT = neoadjuvant therapy.

Imaging Procedures

CEM was performed at two time points for NAT response monitoring: before NAT and after the first cycle but before the second cycle of NAT. MRI was performed at baseline, at the middle of NAT, and after NAT, because it is the standard of imaging response assessment at our institution; unlike CEM imaging, participants did not undergo MRI after the first cycle and before the second cycle of NAT. CEM images were acquired with a Senographe Essential mammography system (GE HealthCare). The first CEM examination was bilateral. Because no participant with bilateral cancer was included and our aim was to monitor NAT response, the subsequent CEM examinations were unilateral. Omnipaque 350, a nonionic low-osmolar iodinated contrast material (iohexol; GE Healthcare) was administered intravenously 2 minutes before CEM acquisition at a dose of 1.5 mL/kg and a rate of 3 mL/sec. Mediolateral oblique and craniocaudal views were acquired at low (26–32 kVp) and high (45–49 kVp) energy, bilaterally at baseline and unilaterally thereafter. All CEM examinations were completed within the previously established 10-minute window of breast tissue enhancement (26).

Image Assessment

Radiologic assessment was performed by two independent readers with more than 2 years of experience in CEM reporting: reader 1 (L.P.G.), with 7 years of experience in breast imaging, and reader 2 (A.T.A.), with 24 years of experience in breast imaging. Both readers were aware of the location of the cancer and of the benign enhancing lesions. This information was retrieved from the results of core needle biopsies performed at staging and is relevant for image interpretation, because it has been reported that benign enhancing lesions can also reduce in size during NAT at MRI (27) and CEM (28). The readers were blinded to any other information, namely, the identity of the participants, the histology results of the surgical specimen, and the molecular subtype of the tumor. The readings of reader 2 were used for interobserver variability analysis.

Radiologic interpretation of CEM images included assessment of the longest dimension of the lesion (Fig 2). CEM measurements were made considering solely the information of recombined images. The low-energy images (similar to a normal digital mammogram) were not used for the measurements. The same measurement orientation and methodology was used for CEM comparative analysis in all time points for each participant. The view that better depicted the longest dimension was chosen for the measurements at the discretion of the reader. In cases of multifocal or multicentric cancers, the longest dimension was determined for each lesion individually and then summed to obtain the measurement registered for each time point (Fig 2). The change in measurements of the lesions from before NAT to after the first cycle of NAT were assessed by the percentage change in the longest dimension of the lesion (CLD) (Fig 2). The CLD was calculated using the following formula: (longest dimension of the lesion before NAT − longest dimension of the lesion after the first cycle of NAT) ÷ longest dimension of the lesion before NAT. Lesion type, tumor regression pattern, and background reduction induced by NAT were assessed using CEM recombined images.

Figure 2:

Method of radiologic interpretation of contrast-enhanced mammography (CEM) images used in the study. The change in lesion measurements (A, C) before neoadjuvant therapy (NAT) and (B, D) after the first cycle of NAT was assessed by the percentage change in the longest dimension of the lesion (CLD) on CEM recombined images. (A, B) The longest dimension measurement of a unifocal breast cancer marked with an intratumoral clip on CEM recombined images, craniocaudal view, in a 54-year-old female participant (A) before NAT and (B) after the first cycle of NAT. The CLD was 33%. (C, D) The longest dimension measurement of a multifocal breast cancer on CEM recombined images, craniocaudal view, in a 48-year-old female participant (C) before NAT and (D) after the first cycle of NAT. The CLD was 11%.

The images at these time points were also evaluated according to Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 (29,30). Imaging response was classified as follows: complete response (disappearance of all lesions), partial response (≥30% reduction in longest dimension), stable disease (<30% reduction in longest dimension and <20% increase in longest dimension), and progressive disease (≥20% increase in longest dimension or new lesions). In multifocal or multicentric disease, the readers were instructed to measure the maximum number of five target lesions at every time point for RECIST assessment.

Histopathologic Analysis and Study End Point

The histopathologic examination of core needle biopsies at baseline and of surgical specimens after the end of NAT were performed at our tertiary cancer center. All pathology reports were performed or reviewed by a dedicated breast pathologist (C.L.), with 25 years of experience in breast pathology. The primary end point was pCR, defined as the absence of invasive residual cancer cells from both the breast and axillary lymph nodes at final surgical histology (ypT0 or is N0) (31). Non-pCR was defined as the presence of invasive residual cancer in either the breast or the axillary lymph nodes.

Statistical Analysis

Continuous variables are expressed as means and SDs. Categorical variables are expressed as the numbers and percentages of cases in each category. Summary tables and simple frequencies were used to describe the data. CIs were calculated with the binomial-based method.

The categories considered for CLD were a CLD less than 20.93% and a CLD of at least 20.93%, and for RECIST were nonresponse (stable disease, progressive disease) and response (complete response, partial response).

The area under the receiver operating characteristic curve (AUC) was calculated (34). The Youden index was used to derive the optimal threshold to identify pCR. Multivariable logistic regression analysis considered all variables associated with pCR at the P value less than .05 level identified at univariable analysis. Odds ratios (ORs) with 95% CIs and P values were calculated for each variable. No multiple comparison adjustments were made due to the exploratory nature of the study.

Interreader agreement for CEM-derived measurements (CLD and RECIST) was assessed using the Cohen κ statistic (32). Interreader agreement for CLD was also assessed using intraclass correlation coefficients (ICCs) with 95% CIs based on the absolute agreement and two-way random effects model (33). Strength of agreement according to the κ statistic was classified as follows: almost perfect (κ = 0.81–1.00), substantial (κ = 0.61–0.80), moderate (κ = 0.41–0.60), fair (κ = 0.21–0.40), slight (κ = 0.00–0.20), and poor (κ < 0.00) (32). Strength of agreement according to ICC was classified as excellent (ICC > 0.90), good (ICC = 0.75–0.90), moderate (ICC = 0.5–0.74), and poor (ICC < 0.50).

All calculations were performed using IBM SPSS statistics for Windows (version 24.0; IBM). All statistical tests were two sided, and a P value less than .05 was considered indicative of a statistically significant difference.

Results

Participant Characteristics

Forty-seven participants with histologically proven breast carcinoma, indication for NAT, and initial CEM were consecutively and prospectively included (Fig 1). Of these, 36 participants (76%) completed the study protocol for early prediction and were included in the final analysis. Dropout was most commonly due to a mild acute contrast material allergy–like reaction (35) after intravenous exposure to iodinated contrast material (three participants) and to gadolinium-based contrast material (one participant). Dropout also occurred due to progression with distant metastatic disease onset during NAT (two participants) and inadequate CEM coverage of the tumor location due to weight loss and tumor reduction during NAT (one participant); four participants did not complete the early prediction protocol.

The demographic, clinical, pathologic, and imaging characteristics of the study sample are summarized in Table 1. All 36 participants were female and self-identified as White. The mean age (± SD) was 48 years (± 10.3).

Table 1:

Characteristics of the Study Sample

All participants received anthracyclines and taxanes. Additionally, participants with the triple-negative subtype cancer received dose-dense carboplatin, and those with human epidermal growth factor receptor 2 (HER2)–positive cancer received pertuzumab and trastuzumab. Most of the participants underwent breast-conserving surgery (26 of 36; 72%), with three of these participants (three of 26; 11%) requiring another surgery due to nonadequate resection margins. In 10 participants (28%), a mastectomy was performed. Most participants underwent sentinel lymph node biopsy (23 of 36; 64%), and 13 participants (13 of 36; 36%) underwent axillary dissection.

Fourteen participants (39%) had luminal-like (hormone receptor [HR]-positive and HER2-negative) breast cancer, 11 participants (30.5%) had HER2-positive breast cancer (seven HR-positive and four HR-negative), and 11 participants (30.5%) had triple-negative breast cancer. After the end of NAT, 11 participants (30.5%) achieved pCR, 24 (66.5%) had pathologic partial response, and one participant (3%) disclosed nonresponse at pathology examination, which corresponded to progression clinically and at imaging.

The participants who achieved pCR were more likely to have HR-negative versus HR-positive breast cancer (nine of 11 [82%] vs two of 11 [18%]; P = .001) (Table 1). No evidence of significant differences was observed between the pCR and non-pCR groups in terms of age, menopausal status, clinical T stage, clinical nodal stage, histologic grade, HER2 status, the presence of invasive lobular cancer, the presence of ductal carcinoma in situ, the presence of microcalcifications at CEM, the type of lesion at CEM, the cancer distribution at CEM, the pattern of regression of the tumor with NAT at CEM, and the reduction in background parenchymal enhancement at CEM with NAT (Table 1).

Association between CME-derived Measurements and pCR

Table 2 displays the CEM-derived measurements in participants with and without pCR. According to RECIST 1.1 assessment after the end of the first cycle of NAT, eight participants (22%) exhibited partial response and 28 participants (78%) exhibited stable disease, and no participants exhibited complete response or progression.

Table 2:

Differences in Contrast-enhanced Mammography-derived Measurements between Participants with and without pCR

Receiver operating characteristic curve by the Youden index method was used to establish the optimal threshold to identify pCR for the CLD after the first cycle of NAT (Fig 3). According to CEM-derived measurements assessment after the end of the first cycle of NAT, 11 participants (30.5%) exhibited a CLD of at least 20.93% and 25 participants (69.5%) exhibited a CLD less than 20.93% (Table 2).

Figure 3:

Receiver operating characteristic curves (ROC) for the fitted models considering hormone receptor (HR)–negative status, percentage change in the longest dimension of the lesion (CLD) greater or equal to 20.93%, and a two-component model including a CLD greater or equal to 20.93% and HR-negative status (CLD+HR) for prediction of pathologic complete response. Area under the ROC curve (AUC) values are reported with SDs and 95% CIs.

A CLD of at least 20.93% (P < .001) and response (complete response, partial response) at RECIST 1.1 (P = .03) were statistically significantly associated with pCR (Table 2).

Table 3 summarizes the diagnostic performance of CLD and RECIST after the first cycle of NAT. The sensitivity, specificity, positive predictive value, and negative predictive value for the CLD cutoff were 73% (eight of 11; 95% CI: 39, 93), 88% (22 of 25; 95% CI: 68, 97), 73% (eight of 11; 95% CI: 39, 93), and 88% (22 of 25; 95% CI: 68, 97), respectively. Respective values for response according to RECIST criteria were 45% (five of 11; 95% CI: 18, 75), 88% (22 of 25; 95% CI: 68, 97), 62% (five of eight; 95% CI: 26, 90), and 79% (22 of 28; 95% CI: 58, 91), respectively.

Table 3:

Diagnostic Performance of Contrast-enhanced Mammography-derived Parameters for Predicting Pathologic Complete Response

Independent Predictive Factors Associated with pCR

At univariable analysis (Table S1), HR status, the CLD, and response according to the RECIST were associated with pCR and were included in the multivariable analysis (Table 4). As CLD and RECIST are both measurements calculated using the longest dimension of the lesion, a two-component model was used for multivariable analysis for each measurement and HR status. A CLD of at least 20.93% (OR, 9.52; 95% CI: 1.34, 67.23; P = .02) remained an independent predictor for pCR, whereas RECIST was not an independent predictor (P = .24).

Table 4:

Multivariable Analysis for Prediction of Pathologic Complete Response

Associations between CLD and pCR according to HR Status

A significant interaction between the CLD and HR was observed (P = .02). In the HR-negative subgroup (n = 15, 42%), including four HER2-positive and 11 triple-negative breast cancers, a CLD of at least 20.93% was associated with pCR (OR, 40.00; 95% CI: 2.01, 794.27; P = .005) (Table S2). The pCR rates were 89% (eight of nine participants) for high CLD (≥20.93%) and 17% (one of six participants) for low CLD (<20.93%) (P = .005). Among the HR-negative participants, the AUC of high CLD (≥20.93%) for predicting pCR was 0.86 (95% CI: 0.65, >0.99; P = .005) (Table S2). Figure 3 visually compares the fitted receiver operating characteristic curves assumed for HR-negative status for high CLD (≥0.93%) and for a model with HR and high CLD (34).

In the HR-positive subgroup (n = 21, 58%), including seven HER2-positive breast cancers, no statistically significant association between CLD and pCR was observed (P = .63). Figures 4 and 5 provide detailed examples of the added value of CLD for participants with HR-negative and HR-positive breast cancer, respectively.

Figure 4:

Contrast-enhanced mammography recombined images, craniocaudal view, in a 43-year-old female participant with hormone receptor-negative breast cancer (A) before neoadjuvant therapy (NAT) and (B) after the first cycle of NAT. The percentage change in the longest dimension of the lesion was 25%. Pathologic complete response was observed after NAT.

Figure 5:

Contrast-enhanced mammography recombined images, craniocaudal view, in a 55-year-old female participant with hormone receptor–positive breast cancer (A) before neoadjuvant therapy (NAT) and (B) after the first cycle of NAT. The percentage change in the longest dimension of the lesion was 6%. Pathologic partial response was observed after NAT.

Interreader Agreement for CME-derived Measurements

There was substantial agreement between both readers for the CLD measurement, with a κ of 0.76 (95% CI: 0.55, 0.97) and moderate agreement for RECIST, with a κ of 0.53 (95% CI: 0.25, 0.81). According to the ICC, both readers had moderate agreement for the CLD (ICC, 0.73; 95% CI: 0.54, 0.85).

Discussion

This single-center prospective pilot study evaluated the performance of CEM-derived measurement changes after the first cycle of NAT for predicting pCR in individuals with breast cancer. Our results demonstrated that the CLD was an independent predictor of pCR after NAT (OR, 9.52; P = .02), with good negative predictive value (88%) and substantial interreader agreement (κ, 0.76). In contrast, response after the first cycle of NAT according to RECIST 1.1 criteria was not an independent predictor of pCR. The study also showed that, among participants with HR-negative breast cancer, who are known to be good responders to NAT, high CLD (≥20.93%) was associated with a greater likelihood of pCR (OR, 40.00; P = .005).

With modern chemotherapy regimens, pCR after NAT has greatly increased. Unfortunately, this achievement has come with greater costs and toxicity, which have led to growing interest in de-escalation of neoadjuvant therapies and personalized neoadjuvant therapy protocols (16,36,37). CLD might serve as an early and in vivo imaging marker to aid in the selection of candidates at high risk of poor response and to tailor treatments of individual patients. Thus, our results are encouraging and consistent with previous publications, in which changes in imaging parameters at PET/CT or MRI before NAT and after one or two cycles of NAT could early predict pCR (10–16).

In the HR-positive group, no association between CLD and pathologic response was identified. This can be related to the low pCR rate observed in the HR-positive subgroup and, possibly, to the small sample size. Compared with RECIST, CLD exhibited better diagnostic performance and interobserver reproducibility. This pilot study was important to understand which of these CEM-based measurements is more promising.

The role of CEM as a predictor of pCR after NAT has good initial results (19–22). This relates to its ability of helping to evaluate perfusion, as MRI does. Currently, in clinical practice, CEM is considered a particularly valuable alternative for NAT response assessment when MRI is contraindicated, unavailable, or has limited access. There is increasing interest in understanding the role of CEM in the imaging toolkit. This may be linked to its favorable logistical profile. Compared with MRI, CEM has fewer contraindications and is faster, less expensive, and more patient-friendly (22–24). The potential high availability of CEM is particularly relevant in the setting of repeated imaging that NAT response assessment entails. Compared with MRI, CEM requires a totally dedicated breast machine (with no competing agenda with other areas), has a lower implementation cost, and is a faster examination, allowing a greater patient flow rate.

In the NAT scenario, the added radiation dose of CEM has minimal biologic significance because breast radiation therapy or mastectomy will be performed. Compared with MRI, CEM has a more restricted anatomic coverage and a higher rate of adverse effects. CEM requires exposure to iodinated contrast material, which is known to have a higher rate of adverse effects (0.153%) compared with gadolinium (0.0404%) (38). Our most frequent cause of dropout was mild acute contrast material allergy–like reaction after intravenous contrast material administration. In keeping with previous reports, it was more frequently related to iodinated contrast material (three participants) than to gadolinium-based contrast material (one participant). It must be emphasized that both types of contrast agents have a very low rate of adverse effects (38) and that most of these are mild and can be managed at the radiology department.

Interestingly, the role of CEM as an early biomarker of response to NAT has not previously been established in the literature, although the low cost, short examination time, and potential high availability of CEM make it an ideal technique for performing repeated assessments in vivo. To our knowledge, our results are the first to indicate that CEM might help in the selection of good and suboptimal responders to NAT regimens, which could help in the construction of personalized NAT regimens.

Notwithstanding the encouraging results, our study had some limitations. First, the small sample size is a considerable limitation. Indeed, this pilot study was the first step in a research pathway we consider very relevant. Second, our study was performed in a single center, used a single mammography equipment, included only White participants, and had no validation set, which may limit the generalizability of our results. A study performed in multiple centers with diverse mammography equipment and external testing would better evaluate the performance expected in everyday clinical practice. Third, due to the small sample size, the molecular breast cancer subtypes were not analyzed. An interesting area of future research would be to evaluate the impact of CEM in making treatment adjustments during NAT.

In conclusion, our study demonstrates that the CLD after the first cycle of NAT can independently predict pCR in individuals with breast cancer. The clinical value of CLD for early in vivo monitoring of NAT regimens needs to be further assessed in larger, multicenter studies.