INTRODUCTION
Breast cancer is the most common malignancy among women worldwide, with postoperative recurrence and metastasis being the leading causes of treatment failure and mortality [1]. As a key pathological feature driving metastasis, lymphovascular invasion (LVI), defined as infiltration of cancer cells into stromal capillaries and lymphatic vessels, followed by vascular dissemination, strongly correlates with increased risks of local recurrence, axillary lymph node (ALN), distant metastasis, and poor prognosis in patients with breast cancer [2-5]. Accurate preoperative assessment of LVI status is therefore critical for guiding clinical decisions, such as optimizing surgical margin and selecting neoadjuvant therapy strategies [5].
However, current LVI diagnosis relies on invasive biopsy or postoperative histopathology, lacking timely postoperative data for personalized treatment. This clinical limitation demands urgent, noninvasive preoperative LVI-prediction tools.
Both histopathological examination and magnetic resonance imaging (MRI) provide essential information for evaluating the biological behavior and detection of breast cancer. Histopathology reveals tumor cellular and molecular characteristics, whereas MRI offers a noninvasive means to assess tumor morphology and microenvironmental changes, making the two methods complementary in clinical assessment. Previous studies have shown that MRI features, such as peritumoral edema, the apparent diffusion coefficient (ADC), and adjacent vessel signs, are associated with LVI [6, 7].
However, their predictive value has not been fully established, and most research has focused primarily on peritumoral edema, neglecting broader patterns of breast edema. Focal breast edema (FBE), which includes peritumoral, prepectoral, and subcutaneous edema types [8], has recently gained attention as a potential biomarker of tumor aggressiveness. The breast edema score (BES), introduced in 2021 to categorize FBE on T2-weighted imaging (T2WI) [9], was subsequently shown to predict sentinel lymph node metastasis [10-12] and to assess response to neoadjuvant therapy [13, 14], supporting its utility in assessing tumor biology.
Despite evidence that breast edema is associated with tumor aggressiveness and prognosis, whether BES on fat-suppressed (FS) T2WI can serve as an imaging marker for predicting LVI, beyond peritumoral edema alone, has yet to be systematically evaluated.
This study aimed to assess whether preoperative T2WI-based BES of FS could predict LVI in breast cancer and to develop a model combining clinicopathological features for individualized treatment planning.
MATERIALS AND METHODS
Patients
This retrospective study was approved by our Institutional Review Board and Ethical Committee, with a waiver of informed consent. Consecutive patients who underwent breast MRI at our hospital from January 2021 to December 2023 were included according to the following inclusion criteria: (1) breast cancer confirmed by histopathology; (2) pathologically confirmed LVI status; and (3) preoperative MRI performed within two weeks before surgery. The exclusion criteria were as follows: (1) biopsy before MRI; (2) prior treatment before MRI; (3) neoadjuvant chemotherapy; (4) inadequate MRI image quality; and (5) incomplete clinical or pathological data.
A total of 574 lesions from 574 patients were included. The patient selection process is illustrated in Fig. (1).
MRI Technique
Breast MRI examinations were performed on a 3.0T scanner (Skyra, Siemens Healthcare) with a 16-channel breast coil and patients in the prone position. The parameters for the MR scans were as follows: (1) axial T1-weighted imaging (T1WI): repetition time (TR), 735 ms; echo time (TE), 8.1 ms; field of view (FOV), 320 mm × 320 mm; matrix, 240 × 320; and slice thickness, 4.0 mm; (2) axial FS-T2WI via Dixon fat suppression technology: TR, 3700 ms; TE, 101 ms; FOV, 320 mm × 320 mm; matrix, 224 × 320; and slice thickness, 4.0 mm; (3) axial diffusion-weighted imaging (DWI) with multiple b values (0, 50, 1000, 2000 s/mm2): TR, 5000 ms; TE, 70 ms; FOV, 280 mm × 170 mm; matrix, 114 × 188; and slice thickness, 5.0 mm; and (4) axial dynamic contrast-enhanced MRI (DCE-MRI) using a T1-weighted time-resolved interleaved acquisition sequence: TR, 5.24 ms; TE, 2.46 ms; FOV, 320 mm × 260 mm; matrix, 182 × 320; slice thickness, 1.5 mm; flip angle, 10°. Gadopentetate dimeglumine (0.1 mmol/kg) was injected at 2 mL/s, followed by a 20 mL saline flush.
Seven consecutive dynamic phases were acquired, each lasting 50 seconds.
Clinical-Pathologic Evaluation
Clinical data (age, menstrual status, family history) were obtained from electronic medical records. Pathological variables included histological type, estrogen receptor (ER) status, progesterone receptor (PR) status, human epidermal growth factor receptor 2 (HER2) status, Ki-67 index, ALN status, and LVI status. LVI was defined as clustered cancer cells within small blood or lymphatic vessels in the peritumoral area.
ER/PR positivity was defined as ≥1% nuclear staining. HER2 positivity was defined as an immunohistochemistry score of 3+ or a score of 2+ with confirmed gene amplification by fluorescence in situ hybridization (FISH) [15]. The Ki-67 index was defined using a cutoff of 14%.
Breast cancer subtypes were classified according to the 2013 St. Gallen consensus: luminal A, luminal B, HER2-enriched, and triple-negative [16].
MRI Analysis
MRI features were independently reviewed by two radiologists with 7 and 10 years of experience, blinded to clinical data. Discrepancies were resolved by a senior radiologist (>15 years of experience). MRI features assessed per Breast Imaging Reporting and Data System (BI-RADS) included: (1) fibroglandular tissue (FGT) density (fatty, scattered fibroglandular, heterogeneous dense, or extremely dense); (2) background parenchymal enhancement (BPE) (minimal, mild, moderate, or marked); (3) tumor laterality (right or left breast); (4) tumor size, measured as the maximum diameter on the second phase of DCE imaging; (5) lesion type (mass or nonmass enhancement); (6) mass shape (round/oval or irregular); (7) mass margin (circumscribed or noncircumscribed); (8) internal enhancement pattern of the mass (homogeneous, heterogeneous, or rim enhancement); and (9) time‒intensity curve (TIC) type (persistent, plateau, or washout).
FBE on FS T2WI was scored independently by two radiologists, blinded to clinical data. Disagreements were resolved via consensus with a senior radiologist. FBE was categorized into four BES grades based on prior studies [8, 9, 17, 18].
BES 1 (no edema) was defined as the absence of a visible edema signal on FS T2WI (Fig. 2a). BES 2 (peritumoral edema) referred to the presence of linear or localized areas of waterlike high signal intensity adjacent to the tumor (Fig.
2b); BES 3 (prepectoral edema) was characterized by linear or bandlike high signal intensity extending into the space anterior to the pectoralis major muscle (Fig. 2c); and BES 4 (subcutaneous edema) was defined as prominent high signal intensity within the skin and subcutaneous fat, excluding localized signal changes caused by direct tumor invasion of the skin (Fig. 2d).
Statistical Analysis
Statistical analyses were performed using SPSS version 22.0 and MedCalc version 19.6. Interobserver agreement was assessed using Cohen’s kappa and weighted kappa. Normality was evaluated via the Kolmogorov–Smirnov test.
Continuous variables are expressed as mean ± standard deviation and compared using a t-test. Categorical variables were compared using chi-square or Fisher’s exact tests. Variables with P < 0.05 in univariate analysis were included in multivariate logistic regression.
Diagnostic performance was evaluated using receiver operating characteristic (ROC) curves, area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. Model calibration was assessed by the Hosmer–Lemeshow test, and AUCs were compared using the DeLong test. A P value < 0.05 was considered statistically significant.
RESULTS
Interobserver Agreement Analysis
The κ coefficients for tumor size, lesion location, lesion type, mass shape, mass margin, and peritumoral edema were 0.845, 0.991, 0.892, 0.823, 0.787, and 0.855, respectively. Weighted κ coefficients for breast parenchymal type, BPE, internal enhancement pattern, TIC curve type, and BES were 0.871, 0.822, 0.784, 0.863, and 0.819, respectively. These results indicate substantial to almost perfect interobserver agreement for all MRI features.
Comparison of Clinical, Pathological, and MRI Characteristics
A total of 574 female patients were included (mean age, 57.1 ± 11.2 years; range, 28–81 years), of whom 174 were LVI-positive. Clinical, pathological, and MRI characteristics are summarized in Table 1. Compared with LVI-negative patients, LVI-positive patients were younger, had higher rates of ALN positivity, higher Ki-67 index, larger tumor size, and greater prevalence of peritumoral, prepectoral, and subcutaneous edema (P < 0.05).
No significant differences were observed between groups regarding menstrual status, family history, histological type, HER2 status, BPE, lesion laterality, lesion type, mass shape, margin, internal enhancement pattern, or TIC curve type (P > 0.05).
Univariate and Multivariate Analysis
Six variables significant in the univariate analysis, including age, ALN status, molecular subtype, Ki-67 index, maximum tumor diameter, and BES, were entered into multivariate logistic regression. Younger age (OR = 0.973, 95% CI: 0.956–0.992, P = 0.004), ALN positivity (OR = 2.316, 95% CI: 1.455–3.685, P < 0.001), a tumor size > 2 cm (OR = 3.618, 95% CI: 2.337–5.601, P < 0.001), and higher BES were identified as independent risk factors for LVI. Compared with BES 1, the odds ratios for LVI were 1.825 (95% CI: 1.085–3.069, P = 0.023) for BES 2, 2.047 (95% CI: 1.051–3.987, P = 0.035) for BES 3, and 4.341 (95% CI: 2.128–8.856, P < 0.001) for BES 4.
Detailed results are presented in Table 2.
Predictive Model Assessment
ROC curves and corresponding analyses are presented in Fig. (3) and Table 3. The combined BES–clinicopathological model demonstrated the highest predictive performance for LVI, with an AUC of 0.778 (95% CI: 0.742–0.811), significantly higher than that the clinicopathological model alone (AUC = 0.762, 95% CI: 0.725–0.796; Z = 1.983, P < 0.05) and the BES-only model (AUC = 0.678, 95% CI: 0.638–0.716; Z=5.200, P<0.05).
Its AUC was also higher than that of the FBE–clinicopathological model (AUC = 0.765, 95% CI: 0.728–0.799), although the difference was not statistically significant (Z = 1.637, P = 0.102). Among all models, the BES–clinicopathological model had the highest specificity (73.50%) and sensitivity (71.84%). The Hosmer–Lemeshow goodness-of-fit test indicated good calibration (P = 0.345).
DISCUSSION
LVI refers to the presence of tumor emboli within lymphatic or vascular structures surrounding breast cancer. Its occurrence indicates potential systemic micrometastasis, suggesting that systemic therapy may be particularly important for these patients. Therefore, LVI status plays a key role in guiding breast cancer treatment strategies [19].
Most previous studies examining the relationship between breast edema and LVI have focused mainly on peritumoral edema. In contrast, this study further evaluated prepectoral and subcutaneous edema and developed a predictive model for LVI based on BES. To our knowledge, similar studies are rare.
Our study demonstrated that the mean age of LVI-positive patients was lower than that of LVI-negative patients, indicating that younger women are at higher risk of LVI. Age was identified as an independent risk factor, consistent with the findings reported by Zhang et al. [20] and Houvenaeghel et al. [21], who also reported younger age as a high-risk factor for LVI in invasive breast cancer. Early-onset breast cancer, influenced by elevated estrogen and progesterone levels, often presents at a more advanced stage and is more likely to be associated with LVI [22].
However, no significant difference in family history of breast cancer was found, which is in line with previous reports [19-21, 23].
Univariate logistic regression analysis revealed that a higher Ki-67 index was a significant predictor of LVI in breast cancer, indicating that LVI-positive tumors tend to grow more rapidly and exhibit greater aggressiveness. Ki-67 is a widely established marker of tumor proliferation, metastasis, and prognosis, and previous studies have reported a positive association between Ki-67 expression and LVI [19, 23], consistent with our findings. In contrast, no significant correlation was observed between HER2 status and LVI, although some studies have suggested HER2 overexpression as a potential risk factor [24, 25].
LVI is a key pathway for lymph node metastasis. Our results showed that ALN positivity was an independent risk factor for LVI, indicating that tumors with nodal metastasis are more likely to exhibit LVI, in agreement with prior reports [26, 27]. Therefore, accurate preoperative imaging evaluation of the axillary lymph node may aid in predicting LVI.
Our study confirmed that the maximum tumor diameter in LVI-positive patients was significantly larger than in LVI-negative patients. Using 20 mm as a threshold, tumor size emerged as a strong independent predictor of LVI, consistent with previous studies [19, 24, 28]. To evaluate the predictive value of preoperative MRI features, the maximum tumor diameter was measured from maximum-intensity projection images, as in prior research [29].
This may reflect the fact that larger tumors often grow more rapidly, infiltrating surrounding tissues and invading capillaries and lymphatic vessels. Irregular tumor shape and noncircumscribed margins were common in both groups, with no significant difference. This contrasts with some prior reports [20, 21] and may be due to the predominance of invasive breast cancers in our cohort, which typically disrupt adjacent tissues and present with stromal proliferation and indistinct borders.
No significant differences were observed in internal enhancement patterns or TIC curve types between groups. Previous studies on DCE-MRI features and LVI have yielded inconsistent results [27, 30]. Although some reported a positive association between BPE and LVI in invasive breast cancer [20, 31], our findings did not support this, aligning with the findings reported by Cheon et al. [6].
Multiple studies have identified breast edema as a poor prognostic factor in malignant breast disease [6, 10, 18, 32]. In our study, FBE independently predicted LVI positivity, with tumors exhibiting edema showing a higher likelihood of LVI. Focal edema on FS T2WI may thus serve as a biomarker of tumor aggressiveness.
To further clarify the relationship between edema patterns and LVI, FBE was subclassified into peritumoral edema (BES 2), prepectoral edema (BES 3), and subcutaneous edema (BES 4). Compared with tumors with BES 1 (no edema), tumors with BES 2, BES 3, and BES 4 had 1.825-, 2.047-, and 4.341-fold greater odds of LVI, respectively, indicating a progressive increase in risk with higher BES levels. Our findings align with those of Cheon et al. [6] and Uematsu et al. [18], who reported a significant association between prepectoral edema and LVI.
However, other studies found no statistically significant differences in prepectoral or subcutaneous edema between LVI-positive and LVI-negative groups, despite higher frequencies in the LVI-positive group [28]. This inconsistency may result from selection bias or the limited sample sizes for prepectoral and subcutaneous edema in those studies.
Our findings demonstrated that the logistic regression model combining BES and clinicopathological variables achieved the highest AUC for predicting LVI in breast cancer, significantly outperforming the clinicopathological model alone, the BES-only model, and the focal edema model (P < 0.05). This combined model also showed the highest specificity among all evaluated models. These results suggest that FS T2WI-based BES provides additional biological insight into LVI, with the potential to improve assessment of tumor aggressiveness and inform clinical management.
The ACR is preparing to release the 6th edition of BI-RADS, with updates presented at the Society of Breast Imaging (SBI) Symposium and the Radiological Society of North America (RSNA) Annual Meeting between 2022 and 2024. Based on accumulating evidence of its prognostic significance in breast cancer [33], peritumoral edema will be added as a new associated finding. Peritumoral edema reflects a localized tissue response and is considered the mildest form of edema, whereas prepectoral edema is often associated with overt LVI, and subcutaneous edema typically indicates more advanced lymphatic involvement [34, 35].
These anatomical patterns of edema may correspond to different stages or severities of disease progression. Despite their potential clinical relevance, prepectoral and subcutaneous edema have received limited attention, and their significance remains unclear. Our study evaluated the predictive value of various types of FBE, contributing to the interpretation of T2WI features and understanding of tumor biology, and providing supplementary imaging markers to support clinical decision-making.
STUDY LIMITATION
This study had several limitations. First, as a retrospective study, patients who underwent invasive procedures, such as core needle biopsy prior to MRI, were excluded, which may have introduced selection bias. Therefore, prospective studies are needed to validate our findings.
Second, given the relatively small number of cases with prepectoral edema (18.4% vs. 9.8%) and subcutaneous edema (17.8% vs. 6.0%) in the LVI-positive and LVI-negative groups, respectively, further studies with larger multicenter cohorts are warranted.
Third, assessment of FBE on T2WI is subject to observer variability. Future research will focus on applying artificial intelligence (AI) for BES assessment and developing radiomic- and deep learning–based models to improve diagnostic accuracy.
CONCLUSION
LVI is a key factor in breast cancer metastasis. This study proposes a noninvasive approach for preoperative prediction of LVI using the BES derived from FS T2WI. Our findings demonstrate a significant correlation between BES and LVI status.
Higher BES levels were associated with progressively increased LVI risk, with BES 2, 3, and 4 showing greater risk than BES 1 (no edema). An integrated model combining BES with clinicopathological features significantly outperformed the clinicopathological model alone in predicting LVI. These results suggest that BES may serve as a potential imaging biomarker for distinguishing LVI-positive from LVI-negative breast cancer.
However, studies on BES remain limited, and its clinical value requires further clarification. Future work will focus on prospective multicenter validation and AI-based analysis of breast edema subtypes to improve model accuracy and reproducibility.