This single-centre, retrospective, observational study was approved by the Institutional Review Board (IRB), and the need for written informed consent was waived. The images of consecutive patients who underwent CEM and CE-MRI between October 2018 and September 2022 were evaluated. Data were collected within a prospective study comparing the diagnostic value of CEM to CE-MRI in a problem-solving setting (Ethics Review Board number 2282/2019).

The study included women with indeterminate mammographic or ultrasound findings (BI-RADS 0, 3) and suspicious lesions (BI-RADS 4, 5). Our population was heterogeneous. From the total of 388 patients were included in the study:

91.3% (n = 294/322) had undergone a screening recall or follow-up examination (of which 51.7% (n = 152/294) for indeterminate lesions and 48.3% (n = 142/294) for suspicious lesions),

8.7% (n = 28/322) presented suspicious symptoms

Exclusion criteria were CEM and CE-MRI performed more than one month apart the absence of one of the two examinations, and the lack of a reference standard.

The standard of reference was histology obtained by imaging-guided needle biopsy (core-biopsy or vacuum-assisted biopsy) or after surgery for all suspicious lesions and one-year follow-up for non-suspicious lesions.

Imaging acquisitionContrast-enhanced mammography

The system used to perform the examinations was a Mammomat Revelation unit (Siemens, Erlangen, Germany). During a single breast compression, a dual-energy examination consisting of high-energy (HE; 49 kVp) and low-energy (LE; 26–32 kVp) images was performed sequentially. An iodinated non-ionic contrast agent (Iomeron® 400, Bracco) was administered at 1 mL/kg body weight at a rate of 3 mL/s using a power injector (Ulrich Medical). Following contrast injection, 20 mL of saline was flushed. Image acquisition started 90–120 s after contrast injection. The examination was performed as follows: craniocaudal (CC) of the affected side, CC of the contralateral side, mediolateral oblique (MLO) of the affected side, and MLO of the contralateral side. The generation of subtracted CEM images was performed by weighted subtraction using a fully automated, locally adjusted, tissue thickness-dependent subtraction factor.

Breast magnetic resonance

Breast CE-MRI was performed on either 1.5-T or 3-T scanners, with dedicated breast coils and patients in the prone position. All protocols included a T2-weighted sequence and a T1-weighted series acquired before and after injection of a gadolinium-based contrast agent, in accordance with international guidelines and recommendations. A single-shot diffusion-weighted echo planar imaging (EPI) sequence (DWI) at b 0 and 800 s/mm2 was also included.

The scanner software automatically generated the ADC maps used for evaluation using a mono-exponential fit of the high and low b data.

Imaging analysis

The images were independently assessed by three fellowship-trained breast radiologists with 3, 4, and 6 years of experience in breast imaging. The clinical data of the patients, the presence and location of the lesions and their histopathological findings were blinded to the readers. Evaluations were performed on dedicated workstations in separate sessions. Readers assessed the CEM images in the first reading session and the MR images in the second session, with a least 2 weeks of washout period to avoid bias due to readers recalling specific cases.

Readers evaluated on CEM the breast density on LE images, following the ACR BI-RADS 5th edition. On CEM, the lesions were described in terms of their characteristics, location, side, and size. This was done using a mammography lexicon associated with descriptors for internal enhancement characteristics (homogeneous, heterogeneous, rim) when enhancement was detected. For recombined (RC) image-only findings, the descriptors applied were mass, non-mass enhancement or enhancing asymmetry. The descriptors were assessed in both CC and MLO projections. On CE-MRI, readers assessed the amount of fibroglandular tissue on native T1-weighed sequences—with and without fat subtraction—the size and the enhancement characteristics using T1-weighted post-contrast sequences using a CE-MRI lexicon. Lesion were measured on LE and RC images on CEM, if a lesion was not visible on both the lesion size was measured as 0 by readers. Lesions were measured on T1w post-contrast sequences on CE-MRI, if the enhancing lesion was not visible, lesion size was measured as 0 by readers.

Readers assessed the lesion conspicuity in three separate categories (low, moderate, and high) as described in the CEM lexicon. Subsequently, an evaluation of lesion conspicuity was conducted utilising a five-point categorical scale based on image quality criteria, with the objective of discerning the subtlest differences in conspicuity. Each reader assessed a lesion conspicuity score for the identified lesions in each image.

This scoring system consisted of:

Grade 1—Not visible: The lesion does not show enhancement.

Grade 2—Poor conspicuity: The lesion is barely discernible, with significant difficulty in identification.

Grade 3—Fair conspicuity: The lesion is moderately visible but lacks clear distinction from surrounding tissues.

Grade 4—Good conspicuity: The lesion is clearly visible with good contrast and separation from surrounding structures.

Grade 5—Excellent conspicuity: The lesion is very well delineated, with clear and distinct visibility.

After readings, a fourth reader, who was not involved in the image analysis, matched the histopathological findings with the reading to proceed with the statistical analysis.

Statistical analysis

Statistical analysis was performed using SPSS 27.00 (SPSS, IBM) and Med-Calc 20.216 (MedCalc Software Ltd.) software.

Categorical variables were reported as absolute numbers and percentages, and continuous variables as mean ± standard deviation (SD).

Univariate non-parametric Spearman correlation analysis was performed to identify potential covariates influencing lesion enhancement on CEM.

Data were analysed using visual grading characteristics (VGC) analysis by calculating the area under the receiver operating characteristic (ROC) curve with 95% confidence intervals (95%CI).

In VGC analysis, readers use a multi-level rating scale to indicate their evaluation of the fulfilment of specific image quality criteria [12] This can be described as an iterative image criterion scoring process in which the reader adjusts the criterion threshold, in a similar way to a reader modifying the threshold using ROC scale steps to indicate confidence in positive or negative decisions.

The classification parameter was the ordinal visual grading scale (e.g. lesion conspicuity), and the reference criterion was the image acquisition method. The study was powered to detect an AUCVGC of 0.6 (zero hypothesis equal lesion conspicuity between CEM and CE-MRI indicated by an AUCVGC of 0.5) at alpha and beta errors of 5% and 20%. The analysis was performed for all lesions and for malignant and benign lesions separately. Mann–Whitney test was performed to evaluate the median lesion conspicuity scores of malignant and benign lesions per reader. A further analysis was conducted on a subset of the data, focusing on lesion types, mass and non-mass lesions. Inter-operator agreement was assessed using the Fleiss’ kappa. Fleiss’ kappa coefficients were interpreted as follows: 0.21–0.40, minimal agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; 0.81–0.90, strong agreement; > 0.90, almost perfect agreement.