Diagnostic value of contrast-enhanced spectral mammography for breast lesions: qualitative and quantitative analyses.

Introduction
Breast cancer is one of the most common malignancies in the world, with the highest incidence and mortality rate ranked in many countries (1). Early breast cancer detection promotes earlier diagnosis, a greater chance of successful treatment, and better survivorship (2). Implementing screening programs is an effective initiative for early breast cancer detection (2,3). The traditional breast screening modality is a two-view mammography, but sensitivity and specificity are not optimal (4,5). Contrast-enhanced spectral mammography (CESM), an emerging imaging technology, is based on the phenomenon of X-ray attenuation. Based on traditional mammography, CESM combines double-energy subtraction technology and uses a nonionic iodine contrast agent injected with a high-pressure syringe (6). In contrast to traditional mammography, the recombined image provides clearer visualization of mass and non-mass lesions, as well as microcalcifications and the mammary gland structure. CESM improves image contrast and facilitates the identification of lesion edges and enhances lesion visualization (7-10).
When traditional mammography and ultrasound cannot identify suspicious findings, CESM can be incorporated into the daily workflow as the next step in diagnosis, with good potential to differentiate benign from malignant breast lesions. Magnetic resonance imaging (MRI) also plays an important role in breast cancer screening and diagnosis (11,12) and shows high sensitivity for deep masses and axillary structures (13), but it has low specificity, high cost, high examination noise, and long examination duration. In contrast, CESM has a lower cost and a wider range of applications than MRI (MRI is contraindicated in patients with pacemakers and in those with claustrophobia), improving diagnostic specificity while maintaining performance comparable to MRI (14-16). CESM plays a guiding role in the diagnosis and treatment of breast cancer, and its diagnosis mainly includes qualitative and quantitative analysis. The qualitative analysis relies on radiologists’ subjective judgment, but the morphological characteristics of benign and malignant breast lesions may overlap, complicating differential diagnosis (17,18). Indeed, the qualitative assessment will inevitably lead to differences in judgment of the same image due to individual differences and the lack of a unified, accepted set of criteria (no enhancement, mild enhancement, moderate enhancement, significant enhancement), and some benign lesions can also be misclassified as malignant. Therefore, a more objective, rapid, and convenient method is needed to improve the diagnostic efficacy of CESM for breast lesions. Quantitative analysis of the functional information derived from a recombined image involves measuring the degree of enhancement to obtain specific numerical values. This approach not only mitigates the limitations imposed by subjective interpretation but also enables precise correlation between lesion morphology and blood supply information. Complementing the morphological features of the breast, it enhances the diagnostic accuracy and discriminatory power for breast cancer, thereby facilitating the detection of breast lesions.
Previous studies evaluated CESM’s ability to identify benign and malignant breast lesions and explored its feasibility in the clinical setting (7,16). The results showed that the qualitative analysis of CESM has good application value for the identification of benign and malignant breast lesions (7,16). However, relatively few studies have quantitatively analyzed the enhancement degree of breast lesions. Therefore, this study aimed to explore the diagnostic value of CESM for breast lesions through qualitative and quantitative analyses. It was hypothesized that combining qualitative and quantitative data would outperform either method alone. We present this article in accordance with the STROBE reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1916/rc).

Methods

Study design and patients
This retrospective study included patients who underwent CESM for suspected (clinical or ultrasound) breast lesions between January 2023 and February 2024 at the Shanxi Bethune Hospital. The study was approved by the ethics committee of Shanxi Bethune Hospital (No. YXLL-2023-051). The requirement for individual informed consent was waived by the committee because of the retrospective nature of the study. All methods were performed in accordance with the relevant guidelines. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.
The inclusion criteria were: (I) no neoadjuvant therapy or surgery before examination; (II) final diagnosis confirmed by pathological examination; and (III) complete imaging data. The exclusion criteria were: (I) pregnant or lactating; (II) overlap of some breast lesions and the background enhancement area, making the region of interest (ROI) delineation questionable; (III) the craniocaudal and mediolateral oblique projections of the affected side failed to show the whole picture of the lesion; or (IV) other malignant tumors.

Data collection and definition
All imaging data were extracted from the patient charts and immediately anonymized upon completion of the extraction process. During the study period, the CESM examination was performed using a Selenia Dimensions digital mammography system (Hologic, Inc., Bedford, MA, USA). All patients were injected with intravenous ioversol (Optiray; Guerbet, Villepinte, France) using a high-pressure syringe before scanning (320 mg/mL, 1.2 mL/kg body weight, flow rate 3 mL/s). After 2 minutes, craniocaudal and mediolateral oblique projections were acquired. CESM was performed using automatic exposure control. The kV was automatically adjusted according to the breast compression thickness, typically ranging from low kV 20–39 to high kV 40–49. The tube current (mA) was also automatically regulated and terminated once the desired image quality was achieved, depending on the patient’s individual tissue attenuation characteristics. The order of acquisition was craniocaudal of the most suspicious breast, craniocaudal of the least suspicious breast, mediolateral oblique of the most suspicious breast, and mediolateral oblique of the least suspicious breast. The examination was aimed to be completed within 7 minutes. DCE-MRI was performed using a Skyra 3.0 T MR scanner and an 18-channel dedicated coil (Siemens, Erlangen, Germany). The routine parameters for the T1-FLASH-3D sequence were used with the following parameters: repetition time (TR)/echo time (TE) =4.45/1.68 ms, slice thickness 1.6 mm, field of view (FOV) =360 mm, flip angle 10°, matrix size 360×360, voxel size 1 mm × 1 mm, and interslice gap 0 mm. After mask scanning, gadopentetate dimeglumine (Gd-DTPA) was injected intravenously using a high-pressure injector at a dose of 0.1 mmol/kg body weight and a flow rate of 2.5 mL/s, followed by a 20 mL saline flush to ensure complete delivery of the contrast agent into the circulation. Immediately after injection, six consecutive dynamic contrast-enhanced (DCE) phases were acquired, each with a 61-second acquisition time. Gd-DTPA was injected using a high-pressure syringe (0.1 mmol/kg body weight, flow rate 2.5 mL/s). Scans were performed at seven time phases after injection, every 75 seconds. All images were downloaded from the picture archiving and communication system (PACS). Two associate chief or chief radiologists conducted a double-blind assessment. Subsequently, a third radiologist reviewed and verified their findings (all had >15 years of experience in breast imaging). Since the two contrast agents cannot be administered simultaneously, an interval of 24–48 hours is required between them. In this study, the two examinations were performed 3–5 days apart, with a maximum interval of 5 days.

CESM analysis
The qualitative analysis used the 5th edition of the American College of Radiology (ACR) Breast Imaging-Reporting and Data System (BI-RADS) (19). The qualitative metrics included breast density (dense or non-dense), shape (round/oval or irregular), margin (smooth or microlobulated/spiculated), calcification (with or without), enhancement pattern (homogeneous, heterogeneous, or rim), enhancement intensity (mild, moderate, or marked), and enlarged axillary lymph nodes (with or without). For the quantitative analysis, the lesion and background were measured using enhanced gray values: lesion grey value (LGV) in the CESM recombined image, according to the method by Rudnicki et al. (20). Using PACS images, the ROI was manually placed by two radiologists. An ROI with an area of 0.3 cm2 was placed in the most prominently and homogeneously enhanced portion of the lesion to represent the lesion, while a 1 cm2 ROI was placed in the relatively homogeneous glandular tissue to represent the background. In clinical imaging practice, CESM has a limited ability to distinguish lesions smaller than 0.5 cm in diameter from surrounding glandular tissue; therefore, ROIs smaller than this size are not generated. For lesions that are clearly visualized, the ROI is placed in the area showing the most pronounced and uniform enhancement. A circular ROI was used for measurement, as this approach is simple, efficient, and well-suited to clinical practice. Delineating the entire lesion would be more labor-intensive, and because enhancement within the lesion is often heterogeneous, the measured values would be less representative. Placing the ROI in the region with the most evident enhancement better reflects the tumor’s blood supply characteristics. The ROI was measured more than three times, and the final result used the dataset with the smallest standard deviation (SD) to record the lesion and background LGV and SD, which were represented by LGVlesion, LGVbackground, and SDbackground, respectively. The quantitative metrics contrast-to-noise ratio (CNR), contrast ratio, and relative signal difference (RSD) were calculated using the formula:
According to the RSD of two adjacent phases in the craniocaudal and mediolateral oblique projections, the CESM enhancement features were classified as persistent type (type I): RSD >10%; plateau type (type II): RSD of 10%; washout type (type III): RSD <10% (21). Lesion enhancement curves in DCE imaging are classified by temporal characteristics (ascending, steady, or descending/washout). The RSD quantifies these changes, supporting reproducible classification across readers and studies (21). Quantitative RSD calculation reduces interobserver variability and provides a standard threshold for kinetic curve types, facilitating improved lesion classification and aiding in BI-RADS assessment. Using RSD reduces subjectivity in visual assessment, making interpretation more reliable for both clinical use and standardized research (21).

DCE-MRI analysis
The scanned DCE-MRI images were imported into a workstation for post-processing to generate and evaluate enhancement curves, which were classified into three types: inflow, plateau, and outflow. Specifically, the enhancement curve was defined as: Type I (persistent; continuous gradual enhancement), typically benign; Type II (plateau), which shows an initial increase followed by a flattening and is concerning for malignancy; and Type III (washout), which shows a rapid initial increase followed by a decrease and is strongly suggestive of malignancy (22,23).

Pathological examination
The gold standard for the pathological diagnosis of breast cancer is histopathological examination. Tissue samples are obtained through core needle biopsy or surgical excision, processed through a series of standard histological procedures, and then examined microscopically for diagnosis. Using the conventional hematoxylin-eosin (H&E) staining method, pathologists can make the essential determinations, whether the lesion is malignant, and if so, whether it is carcinoma in situ or invasive carcinoma, as well as perform histological grading. All breast lesions in this study underwent pathological examination.

Statistical analysis
All analyses were performed using SPSS 27.0 (IBM, Armonk, NY, USA). Graphs were created using GraphPad 9.0 (GraphPad Software Inc., San Diego, CA, USA). The P<0.05 was considered statistically significant. The continuous data were tested for normal distribution using the Kolmogorov-Smirnov test. The continuous data conforming to the normal distribution were presented as means ± SD and analyzed using Student’s t-test, otherwise, they were presented as medians [interquartile ranges (Q1, Q3)] and analyzed using the Mann-Whitney U-test. The categorical data were presented as n (%) and analyzed using the Chi-squared test. The diagnostic value of CESM was evaluated using receiver operating characteristic (ROC) curves. The area under the curve (AUC), sensitivity, and specificity were calculated. Meanwhile, a κ consistency analysis was performed between the CESM and DCE-MRI results.

Results

Basic characteristics and qualitative analysis
This study included 202 female patients: 51 with benign lesions (46.55±11.84 years) and 151 with malignant lesions (51.32±11.21 years). Their age ranged from 25 to 74 years, with a mean age of 50.12±11.53 years. DCE-MRI was performed on 144 patients. There were no significant differences between the 144 patients with DCE-MRI and the 202-patient cohort.
Differences were observed in the qualitative analysis of the CESM images between benign and malignant breast lesions regarding irregular shape (33.33% vs. 69.54%, P<0.001), ill-defined margins (35.29% vs. 65.56%, P<0.001), moderate + marked enhancement intensity (35.29% vs. 73.51%, P<0.001), heterogeneous enhancement (41.18% vs. 62.25%, P<0.015), enlarged axillary lymph nodes (11.76% vs. 50.33%, P<0.001), and type III (13.73% vs. 47.68%, P<0.001) (Table 1).

Quantitative analysis
Compared with benign breast lesions, malignant lesions showed higher LGVlesion, CNR, and contrast ratio in the craniocaudal projection (2,129.26 vs. 2,109.85, P<0.001; 2.81 vs. 1.96, P<0.001; 856.48 vs. 465.52, P<0.001) and mediolateral oblique projection (2,122.74 vs. 2,112.84, P<0.001; 2.59 vs. 2.13, P<0.001; 806.26 vs. 532.30, P<0.001) (Table 2). Typical cases are shown in Figures 1-5.

Diagnostic value of CESM
Using the pathological findings as the gold standard (extracted from the patient charts), the AUCs of craniocaudal projection LGVlesion, CNR, and contrast ratio for benign and malignant breast lesions were 0.778 [95% confidence interval (CI): 0.697–0.859], 0.767 (95% CI: 0.683–0.850), and 0.861 (95% CI: 0.800–0.922), respectively. In the mediolateral oblique projection, the AUCs of LGVlesion, CNR, and contrast ratio of benign and malignant breast lesions were 0.665 (95% CI: 0.577–0.754), 0.662 (95% CI: 0.575–0.749), and 0.823 (95% CI: 0.760–0.887), respectively. Regarding distinguishing between benign and malignant breast lesions, the CESM qualitative analysis demonstrated an AUC of 0.857, a sensitivity of 77.5%, and a specificity of 78.4%. For the CESM quantitative analysis, the AUC, sensitivity, and specificity were 0.895, 84.8%, and 80.4%, respectively. The combination of CESM qualitative and quantitative analyses showed the highest value for breast lesions, with AUC, sensitivity, and specificity of 0.957, 90.1%, and 94.1%, respectively (Table 3 and Figure 6).

Differences in the enhancement curves of benign and malignant breast lesions
The 144 lesions with CESM and DCE-MRI data were used for the κ analysis. In the judgment of the enhancement curve, the concordance rate of the two tests was 85.42%, among which type I was 76.19%, type II was 81.67%, and type III was 92.06%. A κ coefficient of 0.764 for both tests indicated good agreement between the enhancement curves of the CESM and DCE-MRI examinations (Table 4).

Discussion
The results revealed differences between benign and malignant breast lesions in the CESM qualitative and quantitative metrics. Combined qualitative and quantitative CESM analysis achieved high diagnostic value, and the CESM enhancement curve demonstrated concordance with those of DCE-MRI, suggesting that CESM may have a promising application in the early diagnosis of breast cancer.
The results showed that the LGVlesion, CNR, and contrast ratio of malignant lesions were all higher than those of benign lesions, as supported by available studies (24,25). Moreover, the quantitative metrics CNR and contrast ratio could significantly distinguish breast cancer from benign breast lesions, consistent with Rudnicki et al. (20), reporting that a greater degree of enhancement is associated with a higher likelihood of malignancy. Furthermore, compared with the mediolateral oblique projection, LGVlesion, CNR, and contrast ratio in the craniocaudal projection provided a more accurate differential diagnosis for breast lesions, with higher AUC values and sensitivities of 86.1%, 80.8%, and 78.1%, respectively. In addition, some studies have applied CESM quantitative analysis to other clinical aspects, showing significant advantages in identifying potential multifocal and multicenter breast lesions (26,27), predicting the molecular classification of breast cancer (28,29), reducing unnecessary biopsies (30), and monitoring the efficacy of neoadjuvant chemotherapy (31), providing a reference for non-invasive, accurate, and comprehensive preoperative diagnosis.
Malignant breast lesions have typical manifestations such as irregular shape, marginal microlobulated or spiculated margins, heterogeneous or moderate-to-marked enhancement, and axillary lymph node enlargement, while benign lesions are characterized by regular shape, smooth margins, homogeneous enhancement, and mild-to-moderate enhancement (32). Compared with the BI-RADS classification based on traditional mammography, the diagnostic value of qualitative analysis in CESM was relatively high, as evidenced by an AUC of 0.857, sensitivity of 77.5%, and specificity of 78.4%. In contrast, integrating qualitative and quantitative analysis yielded even more promising results. Specifically, this combined approach achieved an AUC of 0.957, sensitivity of 90.1%, and specificity of 94.1%, thereby demonstrating enhanced diagnostic performance. A 2020 meta-analysis comparing CESM and MRI for breast cancer detection found the pooled AUCs for CESM and MRI to be 0.979 and 0.916, respectively, highlighting the high diagnostic performance of both and, at times, a small superiority for CESM (33). High-performance metrics were also reported by Ventura et al. (34). Rudnicki et al. (35) showed that radiologists with less experience in qualitative and quantitative assessment in CESM would be more accurate in breast lesions than radiologists with more experience in qualitative assessment alone, which is consistent with the results of the present study. Therefore, leveraging the objectivity, rapidity, and other merits of CESM quantitative analysis is essential. This approach can enhance the diagnostic accuracy, address the limitations caused by subjective factors, and ultimately elevate the differential diagnostic value of CESM for breast lesions. Furthermore, it serves as a reminder to radiologists that when interpreting CESM images, focusing solely on the presence and degree of enhancement while neglecting the morphological characteristics of lesions may lead to the omission of some valuable information. Instead, radiologists should comprehensively consider both the morphological and functional features of breast lesions, and utilize multiple methods in combination for interpretation, aiming to maximize the diagnostic efficacy in differentiating between benign and malignant breast lesions.
In the malignant group, the CNR in the craniocaudal projection was higher than in the mediolateral oblique projection. In contrast, in the benign group, the CNR in the craniocaudal projection was significantly lower than in the mediolateral oblique projection, indicating that the enhancement intensity of different projections varied significantly. Therefore, this study explored the enhancement curve of CESM as a function of the RSD between the two adjacent images and compared it with the breast DCE-MRI. In this study, malignant lesions were mainly type III (47.68%) and type II (34.44%), while benign lesions were mainly type I (49.02%), similar to previous studies (36,37). Liu et al. (21) showed that type I accounted for 25.5% and 18.6% of benign and malignant lesions, respectively, type II for 16.7% and 27.9%, respectively, and type III for 10.8% and 51.2%, respectively, similar to the present study, but the enhancement curve was obtained from two different photographic projections (craniocaudal and mediolateral oblique), which may change due to the overlap of lesions or the adjacent tissue, thus affecting the study results. The concordance of enhancement curves between CESM and DCE-MRI was 85.42%, indicating a good agreement between the two modalities (κ=0.764), similar to Rong et al. (38). Therefore, it is feasible to judge the enhancement curve of CESM by the RSD of two adjacent photographs, which can provide a reference value for identifying benign and malignant breast lesions. Compared with MRI, CESM is less expensive, more acceptable, and safer for patients with cardiac pacemakers, claustrophobia, or gadolinium contrast allergy (39).
Using the difference in CNR between mediolateral oblique and craniocaudal projections to infer kinetic enhancement patterns is an innovative yet physiologically indirect approach compared to true temporal dynamic studies such as DCE-MRI. However, recent studies and biological considerations provide a rationale for this method (36,38). Both CESM and DCE-MRI rely on tumor neoangiogenesis, which results in abnormal vascular permeability and rapid contrast uptake/washout in malignant tissues. Enhancement differences relate to tissue perfusion and vascular properties, central to both methods (36,38,40). In a typical CESM protocol, the craniocaudal and mediolateral oblique images are acquired sequentially, with a short, real-time interval (often 1–2 minutes). While these are static projections, the tissue is exposed to evolving intravascular contrast levels during the interval, especially in highly vascular lesions. Thus, comparing CNR values between the two views can approximate early enhancement kinetics (38). Studies have shown that relative enhancement differences between craniocaudal and mediolateral oblique images track with kinetic pattern classification on DCE-MRI, with higher enhancement in the later view correlating with persistent “ascending” curves, while a reduction points to a “washout” kinetic, mirroring key concepts from MRI pharmacokinetic modeling (36,38). DCE-MRI acquires multiple serial images at fixed intervals post-injection, directly mapping the inflow (wash-in) and outflow (washout) of contrast over time for dynamic time-intensity curve (TIC) analysis (38,41). In contrast, CESM’s view-based CNR difference offers only two sampled points on the underlying time-attenuation/intensity curve (TAC). Therefore, while not identical, a significant change in CNR between sequential projections can serve as a coarse surrogate for dynamic kinetic behavior, especially in settings where only limited temporal sampling is possible (36,38). Superimposed anatomy and compression: differences in tissue composition or compression between views may also contribute to CNR differences, potentially confounding pure kinetic interpretation (36,38). Published studies report moderate concordance rates (up to 64%) between CESM CNR patterning and DCE-MRI kinetics, with diagnostic value in differentiating benign from malignant lesions (38). Therefore, although projection-based CNR differences in CESM do not capture full kinetics as DCE-MRI does, the timed acquisition of sequential static images leverages evolving contrast pharmacokinetics to approximate early kinetic behavior, offering a practical surrogate in certain clinical contexts where dynamic imaging is impractical or unavailable.
Dense breast tissue, particularly common in younger women, presents a significant challenge because it both increases breast cancer risk and substantially decreases the sensitivity of conventional mammography, leading to more cancers being missed and detected at later stages. CESM markedly improves lesion visibility in these patients and is recommended as a supplemental imaging technique for women with dense breasts (40,42,43). CESM combines standard mammographic images with functional imaging, using an iodine-based contrast agent to highlight areas of increased vascularity, which is frequently an indicator of malignancy (40,42). This contrast highlights lesions that may otherwise be masked by dense glandular tissue, enabling clearer differentiation of benign from suspicious findings on recombined images, even when the low-energy (conventional) images are non-diagnostic (40,43,44). Studies demonstrate that CESM increases cancer detection rates by 6.5 to 23.9 more cancers per 1,000 dense breast screens compared to digital mammography alone, and achieves sensitivity rates near or exceeding 95% in this population (40,42,43).
There are some limitations in this study. Firstly, the small sample size was a notable constraint, with benign lesions predominantly fibroadenomas and malignant lesions primarily invasive carcinomas. Yet, the imbalance between benign (n=51) and malignant (n=151) lesions can bias machine learning models and performance metrics. The number of benign cases should be expanded to confirm the results in a wider selection of lesions. Secondly, only breast lesions with enhancement features were included in this study, and further research is needed to investigate the characteristics of unenhanced masses. Thirdly, the menstrual cycle of young patients may affect the breast background, potentially influencing the results of quantitative analysis of lesions. Finally, because the study was retrospective, the data were limited to those available in the patient charts.

Conclusions
Differences were observed in both the qualitative and quantitative analyses of CESM between benign and malignant lesions. The combined qualitative and quantitative analysis of CESM demonstrated a high diagnostic value. There was a good agreement between the enhancement curves of CESM and DCE-MRI. These findings suggest that CESM may have promising applications in the early diagnosis of breast cancer and provide more accurate reference evidence for non-invasive and rapid identification of the nature of breast lesions in clinical practice. Additional studies are necessary for validation.

Supplementary
The article’s supplementary files as