To evaluate the diagnostic accuracy of Magnetic Resonance Imaging (MRI) and Contrast-Enhanced Mammography (CEM) and to determine which better estimates lesion size.
Material and methodsBetween September 2017 and November 2023, 139 histopathologically confirmed lesions from 95 patients who underwent MRI and CEM were retrospectively analyzed. BI-RADS categories and lesion diameters were reassessed. Sensitivity, specificity, and area under the curve (AUC) were calculated. Breast density, background parenchymal enhancement, and discrepancies in lesion size compared to histopathology were evaluated.
ResultsAmong 139 lesions (117 malignant, 22 benign), MRI detected 112 true positives (TPs) and 5 false negatives (FNs); CEM reported 98 TPs and 19 FNs. For benign lesions, MRI showed 8 false positives (FPs) and 14 true negatives (TNs); CEM, 2 FPs and 20 TNs. Sensitivity was 95.7% for MRI vs. 83.8% for CEM (p = 0.002); specificity was 63.6% vs. 90.9%, respectively (p = 0.058). AUC was 0.90 for MRI and 0.86 for CEM (p = 0.45). Both techniques tended to overestimate tumor size (+3.35 mm MRI, +4.04 mm CEM), with no significant difference in accuracy (p = 0.389).
ConclusionMRI showed higher sensitivity than CEM for detecting malignant lesions, although their accuracy was similar. Both techniques tend to overestimate tumor size.
Morphological imaging techniques (mammography, tomosynthesis, and ultrasound) are routinely used for the detection of breast cancer. However, some lesions can be missed. This occurs more frequently in patients with dense breasts [1,2]. In this context, contrast-enhanced mammography (CEM) is a morphofunctional imaging technique with greater sensitivity than digital mammography [3], employing dual-energy acquisition and iodinated contrast to visualize tumor neovascularization in a similar manner to contrast-enhanced MRI [4]. Recombined images, obtained by subtracting LE (low energy images) from HE (high energy images), display contrast uptake areas and can be considered analogous to subtraction MRI images [5,6]. Because of its ability to show both anatomical and perfusion-related changes more rapidly [7], at lower cost, and with greater accessibility than MRI [8], CEM is emerging as an alternative to address the mentioned drawbacks of MRI (higher cost and lower accessibility), and sometimes limited by patient intolerance due to claustrophobia [9]. Moreover, international authors and societies differ in their opinion regarding preoperative MRI in newly diagnosed breast cancer patients [10,11], and several studies have reported that the higher lesion detection rate by MRI has not translated into lower recurrence rates [12,13].
In the preoperative setting of breast cancer, the diagnostic efficacy of CEM comparable to MRI has been described [14–16]. Given its potential advantages, there is growing interest in implementing CEM as a preoperative assessment, particularly as an alternative to MRI for patients who face its already-mentioned limitations. The primary objective of our study was to compare the diagnostic accuracy of CEM and MRI. A secondary objective was to determine which technique was more accurate in estimating tumor size.
Materials and methodsPatient selectionBetween September 2017 and November 2023, we conducted a retrospective, observational, single-center, multimodality study with patients who underwent both MRI and CEM. Because of the retrospective design, written informed consent was not required. Ethical approval was obtained from the Institutional Review Board of our center.
Inclusion criteria were patients who underwent CEM, MRI, and biopsy or surgery (histopathological evaluation) at our institution. Patients who had received neoadjuvant chemotherapy were also included for evaluation of CEM and MRI performance, although they were excluded from the measurement of the tumor size.
In the protocol of our center, CEM was mainly performed as an adjunct to digital mammography, digital breast tomosynthesis, and ultrasound, before biopsy and definitive diagnosis. MRI was performed after histological confirmation of malignancy, to evaluate for additional ipsilateral or contralateral cancers, and in patients with equivocal or non-concordant histology results, as a problem-solving technique (Fig. 1). Both techniques were also used to measure tumor size.
Study design diagram. DM: Digital Mammography, DBT: Digital Breast Tomosynthesis, US: ultrasound, CEM: contrast-enhanced mammography, B3: histological lesions with uncertain malignant potential, MRI: Magnetic Resonance Imaging, VAB: vacuum-assisted biopsy, w/o: without. *Microcalcifications without US correlation were excluded.
The data were collected and stored in an anonymous database that included age, breast density pattern, results from CEM and MRI, histological characterization, and the size of each lesion.
Image acquisitionCEM protocolCEM examinations were performed using a full-field digital mammography unit (MAMMOMAT Revelation, Siemens Healthcare GmbH, Forchheim, Germany). Intravenous iodinated contrast (iohexol, Omnipaque 350 mg/mL or 300 mg/mL; GE Healthcare, Cork, Ireland) was injected at 1.5 mL/kg with an automatic injector (Medrad Stellant Injector; Bayer Healthcare, Berlin, Germany) at 3 mL/s, followed by a 25 mL saline flush.
At the time of injection, the breast was not compressed to allow for its perfusion. Normally, the suspicious breast was positioned first during the waiting time of 2 min. Then, the breast was compressed, and a LE image was acquired between 28 and 32 kV. Afterward, a HE image was acquired at 49 kV with a titanium filter for each view. The sequence of image acquisition projections used was first, the affected breast in CC view; next, the contralateral breast in CC and MLO views; and finally, the MLO view of the initially imaged breast. For each projection, LE and HE images were acquired, which were then used to generate recombined images (subtracting LE from HE images).
MRI protocolMRI studies were performed on a 1.5T system (AERA, Siemens Medical Solutions, Erlangen, Germany) with a 4-channel breast coil, in prone position without compression.
The protocol included the following sequences and parameters: STIR-T2 axial (slice thickness: 4 mm; TR/TE: 4600/74; field of view [FOV]: 380 × 380 mm; matrix: 384 × 376); axial diffusion-weighted imaging (DWI) (slice thickness: 4 mm; TR/TE: 6300/82; FOV: 340 × 340 mm; matrix: 192 × 90); and axial 3D dynamic T1 sequences (T1-fl3d tra dynaVIEWS 1 + 5, fat suppression; slice thickness: 1.5 mm; TR/TE: 4.8/1.8; FOV: 300 × 300 mm; matrix: 384 × 342) after intravenous administration of 0.1 mmol/kg gadolinium (Gadovist, Bayer Healthcare, Leverkusen, Germany). Dynamic acquisition lasted 6 min to generate time–intensity curves.
Image evaluation and data analysisA breast radiologist with more than 20 years of experience, blinded to previous reports, retrospectively reviewed all images from both techniques (starting with CEM studies).
Breast density and background parenchymal enhancement (BPE) were assessed according to ACR criteria. Lesions detected by CEM and MRI were classified using BI-RADS categories; BI-RADS 1–3 were considered benign, and BI-RADS 4–5 malignant.
The maximum diameter of each visible malignant lesion was recorded using both imaging techniques. Tumor size was measured on recombined CEM images (CC and MLO views) and on the second subtraction sequence in MRI. In neoadjuvant cases, this measurement was excluded.
Histopathological evaluationHistopathological findings (biopsy or surgical specimens) were considered the gold standard. In neoadjuvant cases or patients with benign lesion, the gold standard was core biopsy; in the remaining patients, the surgical specimen was used.
Tumor size was measured on surgical specimens in patients without neoadjuvant chemotherapy.
Statistical analysisStatistical comparisons were performed using Wilcoxon, McNemar, paired Student's t-test, and Mann–Whitney tests.
Diagnostic accuracy of each technique was assessed with sensitivity, specificity, and ROC curve analysis. These curves compare test performance, with a maximum area under the curve (AUC) of 1, where higher values indicate better diagnostic accuracy [17].
Boxplots illustrated discrepancies between imaging-based and pathological tumor size.
All statistical tests were two-tailed, and p-values <0.05 were considered statistically significant. Analyses were performed using Stata (Release 14, StataCorp, College Station, TX).
ResultsSample characteristicsWe included 95 women with 139 histopathologically confirmed lesions. Mean age was 58 years (range 37–84). According to the ACR breast density classification, 19 patients had non-dense breasts (a and b) and 76 had dense breasts (c and d).
CEM showed significantly lower BPE compared to MRI (p = 0.001). Specifically, 89.5% of CEM studies showed minimal/mild BPE versus 76.8% of MRI studies. Among CEM examinations, 9.5% showed moderate BPE and only 1% marked, whereas 20% of MRI studies were classified as moderate BPE and 3.2% as marked. Table 1 describes patients' characteristics.
Patients' characteristics.
| Characteristic | Value |
|---|---|
| Number of patients | 95 women |
| Number of lesions | 139 histopathologically confirmed |
| Mean age (range) | 58 years (range 37–84) |
| Breast density (ACR classification) | 19 non-dense (a & b)76 dense (c & d) |
| BPE on CEM | 55 (57.9%) minimal30 (31.6%) mild9 (9.5%) moderate1 (1%) marked |
| BPE on MRI | 38 (40%) minimal35 (36.8%) mild19 (20%) moderate3 (3.2%) marked |
| Statistical significance between CEM and MRI BPE | CEM showed significantly lower BPE than MRI (p = 0.001) |
We identified 95 index lesions (87 malignant and 8 benign) and 44 additional lesions (30 malignant and 14 benign). Table 2 summarizes the histological diagnoses of malignant tumors, distinguishing between index and additional lesions. Notably, 5 out of 7 cases of DCIS did not show microcalcifications.
Histological subtypes of malignant index and additional lesions. In brackets, the number of lesions detected by CEM and MRI.
| Number of malignant lesions (%) | Histological subtype | Index lesions | Additional lesions |
|---|---|---|---|
| 48 (41%) | Luminal A invasive ductal carcinoma | 35 (CEM 31, MRI 32) | 13 (CEM 11, MRI 13) |
| 30 (25.6%) | Luminal B invasive ductal carcinoma | 24 (CEM 21, MRI 24) | 6 (CEM 3, MRI 6) |
| 9 (7.7%) | Triple-negative invasive ductal carcinoma | 8 (CEM 8, MRI 8) | 1 (CEM 1, MRI 1) |
| 8 (6.8%) | Her2 invasive ductal carcinoma | 5 (CEM 5, MRI 5) | 3 (CEM 3, MRI 3) |
| 7 (6%) | Ductal carcinoma in situ (DCIS) | 6 (CEM 3, MRI 5) | 1 (CEM 1, MRI 1) |
| 7 (6%) | Luminal B invasive lobular carcinoma | 3 (CEM 3, MRI 3) | 4 (CEM 1, MRI 4) |
| 5 (4.3%) | Luminal A invasive lobular carcinoma | 3 (CEM 3, MRI 3) | 2 (CEM 2, MRI 2) |
| 2 (1.7%) | Luminal A tubular carcinoma | 2 (CEM 1, MRI 1) | 0 |
| 1 (0.9%) | Luminal A papillary carcinoma | 1 (CEM 1, MRI 1) | 0 |
Enhancement patterns were classified as mass, non-mass, or absent. Among index lesions detected by CEM, 58 (61.1%) demonstrated mass enhancement, 16 (16.8%) non-mass enhancement, and 21 (22.1%) showed no enhancement (13 malignant and 8 benign). Conversely, on MRI, mass enhancement was observed in 73 (76.8%) index lesions (69 malignant and 4 benign) while 17 (17.9%) (16 malignant and 1 benign) exhibited non-mass enhancement. MRI did not show enhancement in 5 index lesions (5.3%) (two malignant and three benign).
For additional lesions, CEM revealed mass enhancement in 18 cases (40.9%) (16 malignant and 2 benign), non-mass enhancement in 6 cases (13.6%) (all malignant) and 20 lesions (45.5%) showed no enhancement (8 malignant and 12 benign). On the other hand, MRI demonstrated enhancement in most additional lesions: 35 (79.5%) as mass (25 malignant) and 8 (18.2%) as non-mass (5 malignant). Only one additional lesion (2.3%) showed no enhancement on MRI and was benign.
Diagnostic performance of CEM and MRIAmong 117 malignant lesions (87 index and 30 additional lesions), CEM identified 98 true positives (TPs) (76 index and 22 additional lesions) and 19 false negatives (FNs) (11 index and 8 additional lesions), while MRI identified 112 TPs (82 index and 30 additional lesions) and 5 FNs (all index lesions). Only 1 malignant additional lesion was contralateral. Among 22 benign biopsies (8 index and 16 additional), MRI had 8 false positives (FPs) (1 index and 7 additional lesions) and 14 true negatives (TN) (7 index and 7 additional lesions), whereas CEM reported 2 FPs (1 index and 1 additional lesion) and 20 TN (7 index and 13 additional lesions). Table 3 shows the diagnostic performance for each technique and type of lesion (index or additional). The overall sensitivity for detecting breast cancer was 95.7% for MRI compared to 83.8% for CEM (p = 0.002). Regarding specificity: 63.6% for MRI, while CEM reached 90.9%, showing a trend toward statistical significance (p = 0.058).
Diagnostic performance for each technique and type of lesion (index or additional).
| Sensitivity | Specificity | PPV | NPV | |
|---|---|---|---|---|
| Index lesions | ||||
| CEM | 87.4% | 87.5% | 98.8% | 38.9% |
| MRI | 94.3% | 87.5% | 98.8% | 58.3% |
| p value | 0.04 | 1 | 0.957 | 0.296 |
| Additional lesions | ||||
| CEM | 73.3% | 92.9% | 95.7% | 61.9% |
| MRI | 100% | 50% | 81.2% | 100% |
| p value | 0.013 | 0.041 | 0.106 | 0.053 |
ROC analysis showed an AUC of 0.90 for MRI and 0.86 for CEM (Fig. 2), without statistically significant differences (p = 0.45).
Focusing on the clinical significance of these outcomes, the 30 additional malignant lesions were identified in 21 patients. Their detection led to changes in surgical strategy in 13 cases. In 6 of them, a wider lumpectomy was performed based on findings from both CEM and MRI. In another 6 patients, treatment was modified to mastectomy; however, if only CEM findings had been considered, this change would have occurred in 3 of these cases, while in the remaining 3, CEM would not have influenced the surgical plan, and the change was due to MRI results. Finally, in only one patient, the detection of an additional contralateral lesion resulted in bilateral mastectomy.
Tumor size measurementIn 78 patients without neoadjuvant chemotherapy, average differences were +3.35 mm for MRI and +4.04 mm for CEM (Fig. 3). No significant differences were observed in tumor size measurement accuracy (p = 0.389) between CEM and MRI. Both modalities tended to overestimate lesion size.
Box plot. Discrepancies were observed between lesion size measurements obtained by MRI and CEM, compared to those performed by pathologists on surgical specimens. No significant differences were found in the accuracy of tumor size measurements between the two imaging techniques (p = 0.389). Both tended to overestimate lesion size.
MRI failed to detect 5 tumors (1 DCIS, 4 luminal A), all of them index lesions, whereas CEM missed 19 malignant tumors (11 index and 8 additional lesions), including 3 DCIS (index lesion, one 45 mm), 7 luminal A (5 index and 2 additional lesions), and 9 luminal B (3 index and 6 additional lesions). Both techniques detected all HER2 and triple-negative carcinomas, index or additional (Figs. 4 and 5).
Asymptomatic 45-year-old patient seen at our institution for routine mammography. A. The low-energy image showed a suspicious mass classified as BI-RADS 4. B. The recombined CEM image revealed an additional suspicious mass. C. Contrast-enhanced MRI (T1 post-contrast sequence) showed two suspicious nodular enhancements, with good correlation with the recombined CEM image. Pathological study diagnosed two luminal B invasive ductal carcinomas, true positive for both imaging techniques.
Asymptomatic 64-year-old patient. A and B. Low-energy and recombined images showing a metallic clip. No suspicious enhancement was detected (false negative for CEM). C. Preoperative MRI (T1 post-contrast sequence) showed a nodular enhancement corresponding to a true positive on MRI. Pathological evaluation confirmed the lesion as luminal A invasive ductal carcinoma.
In newly diagnosed breast cancer patients, it is essential to determine both the extent of the primary tumor and the presence of additional lesions (ipsilateral or contralateral) to plan the most appropriate treatment [18]. Traditionally, contrast-enhanced breast MRI has been considered the reference imaging technique for preoperative staging. However, studies questioning its benefits [12,13] and the lack of consensus on its indications have fueled debate among breast radiologists [19,20]. Along with MRI limitations (cost, availability, claustrophobia), this has encouraged exploration of alternatives such as CEM.
The appearance of CEM as a faster, more cost-effective and accessible technique has sparked interest in its potential use as an alternative to MRI for preoperative staging. This has led to studies evaluating whether CEM is a reliable method for determining the extent of breast cancer compared to MRI. Regarding CEM indications, it should be prioritized over MRI in patients with incompatible pacemakers, claustrophobia, or obesity. Conversely, it is contraindicated in patients with iodine contrast allergies, renal insufficiency, or when the breast lesion lies outside the mammographic field of view [2,4].
Based on the available information, our study represents the largest intraindividual comparison of diagnostic accuracy between CEM and MRI, using histopathological correlation as the reference standard, with each patient serving as their own control [21,22].
Our results indicate that breast MRI demonstrates significantly higher sensitivity than CEM when detecting index lesions (94.3% vs. 87.4%, p ≈ 0.04), while specificity is identical for both modalities (87.5%). When analyzing additional lesions, MRI achieved 100% sensitivity compared to 73.3% for CEM (p ≈ 0.013), but at the expense of a lower specificity for the first one (50.0% for MRI vs. 92.9% for CEM, p ≈ 0.041). MRI detected all malignant additional lesions but also identified numerous benign findings, whereas CEM was less sensitive but substantially more specific.
When all lesions were considered, MRI maintained its significantly higher sensitivity than CEM (95.7% vs. 83.8%), but its specificity decreased due to the inclusion of additional lesions, as specificity for index lesions was equivalent to that of CEM. Despite these differences, when regarding all lesions, it is important to note that the AUC for CEM does not differ significantly from that of MRI (0.86 vs. 0.90, respectively), mainly due to the lower specificity of MRI (63.6%) compared to CEMs (90.9%). These findings are consistent with observations reported by other authors [23–25].
These results confirm the clinical relevance of using advanced imaging techniques in surgical planning for breast cancer. The identification of 30 additional malignant lesions in 21 patients resulted in substantial modifications to the surgical strategy in more than half of the cases, including wider excisions and conversion to mastectomy. This impact is consistent with previous evidence indicating that preoperative MRI can alter the surgical approach by detecting multifocal, multicentric, or contralateral lesions not identified by other modalities [26]. Although CEM also contributed to these changes, MRI continues to play a decisive role in decision-making, particularly in scenarios where tumor extent is underestimated by conventional methods.
Regarding false negatives and their histology, MRI failed to detect 5 low-grade tumors (1 DCIS and 4 luminal A), all of them index lesions, while CEM missed 19 (11 index lesions and 8 additional ones), including 3 cases of DCIS (index lesions), some of which were extensive, with the largest measuring 45 mm. The remaining undetected tumors by CEM showed low to intermediate aggressiveness: 7 luminal A (5 index and 2 additional lesions) and 9 luminal B (3 index and 6 additional). Importantly, both modalities detected all aggressive subtypes (HER2 and triple-negative), either index or additional. This is likely attributable to tumor neoangiogenesis, which underlies the functional imaging techniques. Neoangiogenesis is a hallmark of aggressive tumors, characterized by disorganized and highly permeable neovascularization that facilitates contrast extravasation into the extracellular space, resulting in intense enhancement patterns [27]. Further evidence is needed to assess the clinical relevance of CEM false negatives in terms of overall survival, morbidity, and cost-effectiveness.
Taking these results into account, it can be considered that, in preoperative breast cancer assessment, a high rate of FP (as seen with MRI) is more acceptable than a high rate of FN (as with CEM), especially given the risk of missing intermediate-aggressiveness tumors (luminal B) with the latter technique. While FPs lead to an increasing number of biopsies, FNs mean leaving a cancer undiagnosed.
It is worth noting that MRI is more susceptible to BPE than CEM [28]. Although both CEM and MRI are morphofunctional techniques, their technical characteristics differ significantly. MRI, due to its tomographic nature, allows acquisition of high-resolution three-dimensional images. This volumetric acquisition and its multiparametric capabilities enable better evaluation of lesion morphology and distribution, contributing to its superior sensitivity, especially in dense breasts. However, this tomographic trait could increase the intensity of BPE, which may result in FP [29], potentially reducing specificity, particularly in dense breasts. This aligns with previous studies indicating that MRI shows a higher degree of background enhancement in cases of high breast density [30]. For this reason, CEM may be a useful tool in patients with marked BPE.
Regarding tumor size measurement, CEM and MRI showed comparable performance, with no significant differences between them. This study used measurements taken by pathologists on surgical specimens as the reference. Interestingly, both CEM and MRI tended to overestimate lesion size, in agreement with previous findings [23].
Our study has some limitations. Its retrospective nature, the review of images by a single radiologist, and the use of CEM as a problem-solving technique prior to biopsy, rather than in the preoperative context like MRI, may have influenced the analysis. Additionally, 5 of the 7 DCIS cases included did not present microcalcifications, which could introduce a selection bias. Another limitation is the relatively small number of benign lesions (22 cases), which may have contributed to the trend toward statistical significance when comparing the specificities of CEM and MRI. With a larger sample of benign lesions, significant differences might have been detected.
In conclusion, MRI was more sensitive than CEM in detecting malignant lesions, although they showed similar overall accuracy. Both techniques tended to overestimate tumor size. CEM should be considered a second-line diagnostic tool, despite its similar accuracy to MRI, given its higher false negative rate, though it remains reliable in specific contexts or where MRI is less accessible. Further studies are needed to evaluate the long-term clinical impact of CEM's lower sensitivity compared to MRI.
Informed consentThe authors have obtained informed consent from the patients.
Ethical considerationsThis study was reviewed and approved by the Ethics Committee of the University of Navarra, in accordance with the ethical principles set out in the Declaration of Helsinki and the relevant institutional guidelines.
FundingNone.
Authors' contributionsLP contributed to the conception, development, data collection, interpretation, and final approval of the study. PM and MBP participated in data collection, interpretation, and manuscript writing. AE collaborated in data collection, interpretation, and manuscript revision. CUI, CMH, and MTC contributed to the critical revision of the manuscript. All authors read and approved the final version of the manuscript.
The authors have no conflict of interests in this study.
