The study received approval from the Institutional Review Board and was granted a waiver for patient informed consent due to its retrospective design. We retrospectively included patients with biopsy-proven BC who underwent either CEM or MRI at our institution for preoperative assessment between January and December 2022. Indications for preoperative imaging followed the European Society of Breast Cancer Specialists (EUSOMA) criteria [29] or were based on clinical request. The selection between CEM and MRI reflected our institutional policy, in routine clinical practice, of allocating to CEM women aged ≥ 60 years and/or those with relative or absolute contraindication to MRI, reserving MRI to all other patients. The age cutoff of 60 years was chosen as a practical, non-mandatory guideline for the following reasons: (1) Adherence to radiation protection principles by favoring techniques like MRI, that do not use ionizing radiation, in younger patients; (2) The greater ease of execution and patient tolerability of CEM in an older population; (3) The lack of an established, standardized age-based criterion in the published literature [30].

After applying the exclusion criteria reported in Fig. 1, the final population consisted of 195 women who underwent CEM or MRI in 91 and 104 cases, respectively.

Study flowchart. BC, breast cancer; CEM, contrast-enhanced mammography; MRI, magnetic resonance imaging

The reference standard consisted of histopathological evaluation of surgical specimens, performed by one of three board-certified pathologists with 5–25 years of subspecialty experience in breast pathology, in accordance with College of American Pathologists guidelines [31]. Breast cancer histotypes were documented, as well as the presence or absence of intraductal extension. NAC involvement was histologically confirmed when pathological examination of the surgically resected specimen demonstrated the presence within the nipple tissue of either: (1) invasive carcinoma components, (2) ductal carcinoma in situ (DCIS), or (3) Paget’s disease cells. In nipple-sparing mastectomy cases, histopathological evaluation included assessment of both the retroareolar disk and surgical margins [32]. Full histopathological assessment of the NAC was reserved for total or skin-sparing mastectomy specimens, as for central quadrantectomies. For patients treated with quadrantectomy or lumpectomy, negative surgical margins were deemed to infer negative nipple status.

CEM was performed with a dedicated system (Selenia Dimensions 3D, Hologic) following intravenous injection of 1.2 mL/Kg of iodinate contrast medium (Iobitridol 350 mg/mL, Guerbet) and a 20 mL bolus of saline solution. Injection was performed under remote control (Accutron CT-D, Medtron) at an injection rate of 3 mL/s. The protocol included the acquisition of right-sided cranio-caudal (CC) and medio-lateral oblique (MLO) views 2 min after contrast administration, immediately followed by left-sided CC and MLO views. Seven minutes post-contrast injection, CC and/or MLO views of the affected side were obtained, depending on the decision of the attending radiologist.

MRI examinations were performed on one of two 1.5 T magnets (Magnetom Aera or Magnetom Avanto, Siemens Medical Solutions), with a bilateral 16-channel coil and the patient in prone position. Standard breast MRI protocol included diffusion-weighted imaging (DWI), T2-weighted imaging and T1-weighted, non-fat-saturated dynamic contrast-enhanced imaging (DCE) with and without digital subtraction, as detailed in Table 1. A contrast dose of 0.1 mmol/kg gadoteridol (Prohance, Bracco Imaging) was administered at an injection rate of 2 mL/s, followed by a 20 mL bolus of saline solution. Injection was performed under remote control.

An independent study coordinator, not involved in the image interpretation process, arranged dedicated reading sessions with two radiologists who met EUSOMA and European Cancer Concord expert criteria [33]. Reader 1 (R1) and reader 2 (R2) had 15 and 4 years of continued clinical activity in breast imaging, respectively. Both readers were blinded to clinical, histopathological or follow-up information, except for the knowledge of BC presence and location of the index lesion (IL), defined as the previously confirmed tumor prompting locoregional staging.

Study readings were preceded by a training session during which R1 and R2 reviewed 10 CEM and 10 MRI exams, randomly selected by the study coordinator. The training phase aimed to standardize the evaluation based on normal findings (e.g., linear symmetrical or asymmetrical enhancement [34]) and the suspicion criteria listed below. During study reading sessions, R1 and R2 were asked to independently review randomly presented CEM or MRI exams, focusing on NAC. CEM included both low-energy (LE) and recombined (REC) images, while MRI included all the images and maximum intensity projections reconstructions obtained from the first post-contrast DCE sequence. NAC involvement was assessed based on one or more of the following criteria: (1) nipple retraction and invasion; (2) abnormal enhancement of NAC, particularly when compared with the contralateral NAC; (3) tumor nipple enhancement (TNE), i.e., presence of enhancement between tumor and nipple base [35]; (4) peri-areolar skin thickening, i.e., thickening of the peri-areolar zone; (5) abnormal morphology or asymmetry of the NAC; (6) a tumor-to-nipple (TTN) distance ≤ 10 mm, measured from the nearest tumor margin to the base of the nipple. In cases of multifocal or multicentric disease, the shortest distance between any tumor focus and the nipple base was considered [7]. Moreover, contrast-enhancement pattern (mass vs. non-mass), tumor size, and, for CEM examinations, associated microcalcifications were also assessed.

Peri-areolar thickening may, in some cases, reflect reactive or inflammatory edema rather than true neoplastic infiltration. In our study, the distinction between these conditions was based on a combined evaluation of morphology, enhancement pattern, and T2-weighted signal characteristics (on MRI). Edema was defined as diffuse and symmetric thickening with preservation of the smooth contour and layered structure of the NAC, showing homogeneous hyperintensity on T2-weighted sequences on MRI and minimal or no enhancement on CEM. True NAC infiltration, instead, was identified when architectural distortion, irregular or nodular enhancement, and loss of the normal dermal–epidermal interface were observed, often contiguous with the primary lesion or involved ducts. These radiologic criteria are consistent with previously reported features in breast MRI literature [12, 36].

The readers performed independent evaluations, exercising complete discretion in subjectively integrating the aforementioned imaging features to formulate a binary determination regarding NAC involvement (present/absent). As a general guideline, it was suggested to consider that the presence of ≥ 2 criteria increased the likelihood of NAC involvement.

Ancillary features such as location and IL enhancement, breast density category according to the Breast Imaging Reporting and Data System (BI-RADS) [37, 38] and Background Parenchymal Enhancement (BPE) were also recorded. For the purpose of analysis, BI-RADS density categories A and B were considered as “non-dense” while categories C and D were regarded as “dense” either directly on CEM or based on previous mammography examinations in the MRI group.

After independent readings, discordant cases were resolved through consensus to establish a single, unified set of CEM and MRI readings for analysis.

We used descriptive statistics to report the main clinical and imaging variables. Given that the Shapiro–Wilk test indicated a non-normal distribution of continuous variables, they were summarized with median values and interquartile range (IQR). Comparisons between the CEM and MRI groups were performed with the Mann–Whitney U test. Categorical variables were compared with the chi-square test. Relevant proportions were complemented with 95% confidence intervals (95% CI).

Given the inherently different populations, it was not feasible to apply a propensity score matching approach. Therefore, we assessed whether the main patient-related characteristics associated with referral to CEM or MRI—namely age and breast density—were significantly different between the two groups (Table 2).

By matching the radiologists’ readings with the reference standard, we calculated the sensitivity, specificity, positive (PPV), and negative predictive value (NPV) for NAC involvement. We used Fisher’s exact test to compare the diagnostic performance of CEM versus MRI. Inter-reader agreement for the binary CEM and MRI rating was quantified using Gwet’s AC1 as the primary statistic, given the known instability of Cohen’s κ under marked prevalence imbalance, percent agreement and Cohen’s κ were also reported for transparency [39,40,41].

Logistic regression analyses were applied to identify independent predictors of NAC involvement among clinical and imaging variables. A stepwise selection method was employed, including variables with a significance threshold of p < 0.05. Two separate models were developed: one for CEM and one for MRI. The clinical variables included in both models were as follows: age (divided into quartiles according to its distribution within the study cohort), histological type of breast cancer (divided into IBC-NST, ILC, DCIS and other), in situ component (present or absent) and immunohistochemical profile (divided into Luminal A, Luminal B/Her2-, Luminal B/Her2+, Triple-negative and Her2+). In addition, the following imaging-specific variables were included in each model: breast density (Dense vs. Non-dense breasts), BPE enhancement (divided into low, mild, moderate, marked), nipple retraction and invasion, abnormal enhancement of NAC (present or absent), TNE, peri-areolar skin thickening, abnormal morphology or asymmetry of the NAC, TTN distance (divided into quartiles according to its distribution in the cohort), contrast-enhancement pattern (mass vs. non-mass), tumor size, and, for CEM examinations, associated microcalcifications.

As a second step, we fitted an additional multivariable model to test whether the association of age and breast density with NAC involvement differed according to the imaging modality (CEM versus MRI). This model included the imaging modality as a categorical variable (CEM vs. MRI), as well as age, breast density, and their interaction terms with the imaging modality (calculated as the product of each variable and the modality indicator) [42].

Analyses were performed using commercially available software (MedCalc® Statistical Software version 23.1.3, MedCalc Software Ltd.) and Stata/BE 19.5. Alfa level was 0.05.