1. Introduction
Dynamic contrast‐enhanced (DCE) MRI has become an important adjunct in breast cancer detection, particularly for high‐risk screening and preoperative staging [1, 2]. However, nonmass enhancement (NME) remains one of the most challenging MRI findings, with wide variability in morphology and enhancement kinetics that often lead to diagnostic overestimation [3]. BI‐RADS 4 NME, especially Category 4a, carries a low and inconsistent positive predictive value, resulting in a substantial number of benign biopsies and unnecessary procedures for patients [4, 5].
Targeted ultrasound (TUS) is frequently used to correlate MRI‐detected lesions, but NME typically lacks discrete sonographic morphology, limiting its ability to refine risk [6]. Contrast‐enhanced ultrasound (CEUS) offers additional microvascular information and may help differentiate benign from malignant NME, yet its clinical utility in this specific context remains inadequately studied [7]. Evidence regarding whether CEUS can safely reclassify MRI‐detected NME and reduce biopsy burden is limited, and no prospective data focused specifically on BI‐RADS 4 NME are available [8].
Given the high clinical cost of MRI‐driven false positives and the need for more precise risk stratification tools, establishing whether TUS combined with CEUS can reliably identify low‐risk NME and detect clinically significant lesions such as ductal carcinoma in situ (DCIS) is of substantial importance. This prospective study evaluates the performance of TUS and CEUS for reassessment of MRI‐detected BI‐RADS 4 NME, with the goals of reducing unnecessary biopsies while maintaining high sensitivity for malignancy.
2. Materials and Methods
2.1. Patients
In this prospective study, clinical and imaging data were consecutively collected from patients with breast lesions treated at a single tertiary‐care center between January 2017 and December 2023. The inclusion criteria were as follows: (1) patients with breast lesions detected on DCE‐MRI and classified as BI‐RADS Category 4 NME by consensus between two senior radiologists in accordance with the 5th edition of the ACR BI‐RADS guidelines [9]; (2) availability of complete and high‐quality TUS and CEUS imaging corresponding to the MRI‐detected lesions; (3) definitive pathological diagnoses obtained through surgical resection; (4) no prior biopsy or neoadjuvant therapy before undergoing DCE‐MRI or US examination; (5) no contraindications to CEUS, with all patients providing written informed consent prior to the CEUS procedure; (6) the interval between MRI and US examinations did not exceed one week; and (7) all imaging and surgical procedures completed within one month. This study was approved by the institutional ethics committee of a tertiary‐care hospital (approval no.
2023‐383). Patients who did not meet the inclusion criteria were excluded.
Of 9679 patients undergoing breast DCE‐MRI, 1916 were classified as BI‐RADS Category 4. Among 323 NME cases assessed with TUS and CEUS, 167 patients with 169 lesions were included after exclusions for lost MRI images (n = 95), incomplete pathology (n = 36), poor sonographic quality (n = 22), and metastatic lesions (n = 1) (Figure 1). The cohort comprised 83 benign and 86 malignant lesions, with mean ages of 45.3 ± 12.6 and 57.6 ± 11.3 years, respectively (p < 0.001).
2.2. MRI Examination
Breast MRI was performed using a 3.0T scanner (Philips Ingenia, Philips Healthcare, the Netherlands) equipped with a 16‐channel breast coil. The imaging protocol included axial T1‐weighted imaging with TR/TE = 609/8 ms, 36 slices, field of view (FOV) = 280 × 340 mm, matrix = 352 × 424, and slice thickness = 3.85 mm, followed by axial T2‐weighted imaging with TR/TE = 4000/70 ms, 36 slices, FOV = 280 × 340 mm, matrix = 280 × 318, slice thickness = 3.85 mm, and flip angle = 90°. Diffusion‐weighted imaging (DWI) was performed with TR/TE = 5098/70 ms, b‐values = 0 and 1000 s/mm2, 32 slices, FOV = 350 × 240 mm, matrix = 140 × 93, and slice thickness = 5 mm.
Diffusion kurtosis imaging (DKI) was conducted with TR/TE = 4034/102 ms, b‐values = 0–3000 s/mm2, 32 slices, FOV = 350 × 240 mm, matrix = 140 × 94, and slice thickness = 5 mm. DCE T1‐weighted imaging was performed after intravenous injection of 0.2 mmol/kg gadolinium‐based contrast agent (Gd‐DTPA, Magnevist, Bayer Healthcare) at a rate of 3 mL/s, followed by a 20 mL saline flush. This sequence captured 29 phases at 9.7 s per phase, with TR/TE = 3.2/1.55 ms, FOV = 360 × 360 mm, matrix = 276 × 276, flip angle = 10°, and slice thickness = 3.0 mm.
2.3. Diagnostic Workflow for BI‐RADS Category 4 NME Lesions Using TUS and CEUS
Following the identification of BI‐RADS Category 4 NME lesions on DCE‐MRI, a TUS examination is conducted within one week. Two experienced radiologists (F.L. and C.J.) perform the examination using Aplio 500 and Aplio i900 systems (Canon, Japan), equipped with 14‐5L and 14‐8L transducers, respectively.
For localization of MRI‐detected lesions via TUS, the distance from the lesion margin to the nipple is considered the primary reference. Telegrafo et al. [10] reported this distance as a stable and reliable metric, with a minimal discrepancy of 0.7 ± 0.2 cm between MRI and TUS measurements. While the original validation by Telegrafo et al. was conducted primarily in a cohort of mass‐type lesions, subsequent studies applying a comparable distance‐ and landmark‐based localization strategy specifically to nonmass enhancement lesions have supported the applicability of this general principle to NME, albeit with lower detection rates than for mass lesions (MLs), reflecting the greater positional displacement and less distinct margins characteristic of NME [11].
Additionally, the lesion’s position relative to surrounding anatomical structures observed on MRI—such as breast glandular patterns, Cooper’s ligaments, fat morphology, vascular distribution, and adjacent abnormalities—facilitates precise localization during TUS. When a suspicious lesion is detected via TUS, it is classified as a ML or nonmass lesion (NML), with detailed ultrasonographic features documented according to the 5th and 6th editions of the BI‐RADS guidelines [9, 12]. Color Doppler flow imaging (CDFI) is then used to evaluate blood flow signals, with the color scale reduced to below 8 cm/s to detect small and low‐speed flow, and wall filtering minimized to prevent false positives.
After lesion confirmation on TUS, radiologists conduct CEUS at a low mechanical index of 0.07, with uniform overall and background gain set at 73 dB. The contrast agent SonoVue (Bracco, Italy) is prepared according to the manufacturer’s instructions, with 5 mL of contrast suspension injected intravenously, followed by a 5 mL saline flush. Timing starts concurrently with injection, and dual‐imaging is employed to prevent lesion displacement during contrast imaging.
The probe remains stationary for the first minute, after which dynamic scanning of the suspicious area is performed to detect additional lesions.
2.4. CEUS Feature Assessment and BI‐RADS Classification Adjustment Criteria
The CEUS‐based downgrade and upgrade criteria described below were prespecified prior to the initiation of patient enrollment in 2017, based on the breast CEUS literature available at that time and the research team’s preliminary institutional experience, and remained unchanged throughout the study period [13, 14]. Image interpretation and BI‐RADS reassignment were performed at the time of each examination, with radiologists blinded to clinical and pathological information and prior to the availability of the pathological reference standard.
CEUS interpretation evaluates the following parameters: (1) enhancement time, (2) peak enhancement intensity, (3) enhancement morphology, (4) enhancement margins, (5) enhancement order, (6) contrast agent distribution, (7) presence of increased perilesional vascularity, (8) perfusion defects, (9) enhanced scope, and (10) contrast agent retention. In this study, three radiologists were involved in image interpretation. LF.J and YY.C independently interpreted the images.
Discrepancies in interpretation were resolved through consensus among all three radiologists (LF.J, YY.C, and LF.D). The radiologists were blinded to the clinical and pathological data during the image review process.
For ML, benign features suggesting low malignancy risk on CEUS include synchronous or iso‐/hypo‐enhancement compared to adjacent breast glandular tissues in the arterial phase on CEUS or rapid, intense enhancement with circumscribed margins, regular morphology, homogeneous enhancement, no increased perilesional vascularity, and absence of contrast retention. Malignant features, indicating higher malignancy risk, include arterial‐phase hyper‐enhancement with irregular morphology, noncircumscribed margins, heterogeneous enhancement, increased perilesional vascularity, enlarged enhancement scope, and contrast retention. These criteria are consistent with enhancement principles reported in earlier breast CEUS studies [13, 14].
For NME lesions, benign features include iso‐/hypo‐enhancement relative to surrounding breast glandular tissues in the arterial phase on CEUS, consistent with general principles of breast CEUS enhancement kinetics established prior to study initiation; these features are also consistent with subsequently published NME‐specific data [15]. Malignant features, suggesting elevated malignancy risk, include intense arterial‐phase hyper‐enhancement with increased perilesional vascularity, contrast retention, and perfusion defects. Lesions primarily showing benign features are downgraded to BI‐RADS Category 3 or below, indicating low malignancy likelihood and supporting follow‐up rather than immediate intervention.
Conversely, lesions predominantly exhibiting malignant features are upgraded to a higher CEUS BI‐RADS category (e.g., 4 or above), necessitating further diagnostic or therapeutic measures. The overall study design, including prestudy criteria development and the prospective per‐patient diagnostic workflow, is summarized in Figure S1.
2.5. Lesion Localization for Surgical Excision
On the day of surgery, the same radiologists who performed the original TUS/CEUS examination re‐localized each target lesion using conventional ultrasound, based on the previously recorded distance to the nipple and surrounding anatomical landmarks. The corresponding skin projection was marked directly on the patient’s body surface, and this marking, together with the documented imaging description of lesion depth and relationship to adjacent structures, guided the surgeon in defining the excision margins. This procedure, consistently performed by the same radiologist team throughout the study period, ensured correspondence between the lesion identified on MRI, evaluated by TUS/CEUS, and submitted for pathological examination.
2.6. Pathological Findings
All 169 lesions underwent surgical excision, in accordance with the study inclusion criteria requiring definitive pathological diagnosis via surgical resection, with the resulting histological diagnosis serving as the reference standard for all diagnostic performance analyses. We documented the pathological diameter of malignant lesions, grade, hormone receptor status, Ki‐67, and the status of human epidermal growth factor receptor 2 (HER2) [16, 17].
2.7. Statistical Analysis
Data processing was conducted using Python 3.8.19 (Python Software Foundation, USA). For quantitative data that conform to a normal distribution, results are expressed as the mean ± standard deviation; if not conforming to a normal distribution, they are expressed as the median (interquartile range). Qualitative data are represented as frequencies and percentages (n, %).
To compare differences between groups, quantitative data were analyzed using the independent samples t‐test (for normal distribution) or the Mann–Whitney U test (for non‐normal distribution), while qualitative data were analyzed using the chi‐squared test or McNemar’s test. Correlation analysis was performed using Spearman’s rank correlation coefficient. Nonparametric bootstrap resampling was employed to estimate the 95% confidence intervals (CIs).
A p value of less than 0.05 was considered the threshold for statistical significance. The consistency of image interpretation between TUS, CEUS, and DCE‐MRI was assessed using Cohen’s kappa coefficient.
3. Results
3.1. Histopathologic Characteristics
Malignant lesions included 50 DCIS, 27 invasive ductal carcinoma (IDC), 4 solid papillary carcinoma (SPC), and 5 others. Benign lesions included 28 adenosis, 27 inflammatory changes, 9 intraductal papillomas, 8 usual ductal hyperplasia, 7 atypical ductal hyperplasia, and 4 others. The proportion of BI‐RADS 4a was higher in benign (72.3%) than malignant cases (47.7%, p = 0.002) (Table 1).
TUS detected 97.0% (164/169) of lesions (96.4% benign, 97.7% malignant). The kappa value for inter‐reader agreement was 0.810 (Table S1).
3.2. Correlation of TUS/CEUS Features With Pathology in BI‐RADS 4 NME Lesions
TUS identified 36 MLs (15 benign and 21 malignant) and 128 NMLs (65 benign and 63 malignant). Lesions were classified as BI‐RADS 3 (3/164) or 4 (161/164), with significant correlation to pathology (p = 0.049, p < 0.001). For ML on CEUS, pathological correlation was observed with enhancement time, peak intensity, enhanced shape, margins, contrast distribution, perfusion defects, perilesional vascularity, enhancement scope, contrast retention, and CEUS BI‐RADS (r = 0.357–0.832).
For NML, correlated features included peak intensity, enhancement order, contrast distribution, perilesional vascularity, enhancement scope, contrast retention, and CEUS BI‐RADS (r = 0.271–0.652) (Figure 2). Inter‐reader agreement kappa values ranged from 0.822 to 0.957 (Table S1).
3.3. Diagnostic Performance of CEUS in Downgrading BI‐RADS 4 NME Lesions
TUS alone reduced the rate of biopsies for benign lesions by 7.2% and the overall biopsy rate by 4.7%, with a sensitivity of 97.7%. When TUS was combined with CEUS benign indicators, the reduction in biopsies for benign lesions was 43.4% (36 out of 83, 95% CI: 32.5%–54.2%), and the overall biopsy rate decreased by 23.1% (39 out of 169, 95% CI: 16.6%–29.6%), maintaining a sensitivity of 96.5% (83 out of 86, 95% CI: 91.9%–100.0%) (Table 2).
Of the 39 cases avoiding biopsy, 3 were false‐negative malignancies and 36 were benign (adenosis: 47.2%, p = 0.002 vs. non‐downgraded benign lesions). No significant differences in ER, PR, Ki‐67, grade, molecular subtype, T stage, or N stage were observed between false‐negative and true‐positive cases (Table 3).
Of three false‐negative cases, two were missed by CEUS, and one was downgraded. One missed case involved an occult breast lesion with enlarged axillary lymph nodes, diagnosed as low‐grade IDC (Luminal B) with 10% DCIS (MRI: 27 mm). The other was Paget’s disease with high‐grade HER2‐positive DCIS, showing 24 mm nonmass enhancement behind the left areola on MRI.
The downgraded case was intermediate‐to‐high‐grade Luminal B DCIS with invasion, showing mild enhancement (8 × 12 mm) under the right areola on MRI, while CEUS indicated a slightly hyper‐enhanced lesion.
3.4. Stratified Performance Analysis of CEUS Application
Stratified analysis revealed higher MRI overestimation in asymptomatic cases, lesions ≤2 cm, homogeneous enhancement patterns, Type I kinetic curves, and MRI BI‐RADS 4a. Benign biopsy reduction was 53.3% (16/30) in asymptomatic versus 37.7% (20/53) in symptomatic cases; 71.4% (20/28) for lesions ≤2 cm versus 29.1% (16/55) for > 2 cm (p < 0.001); and 60.0% (6/10) for homogeneous enhancement (p = 0.034). Reduction rates by kinetic curve type were 57.7% (Type I), 37.7% (Type II), and 25.0% (Type III).
For MRI BI‐RADS 4a, the reduction was 50.0% (30/60) versus 29.4% (4b) and 16.7% (4c) (Figures 3, 4, and 5). No intergroup differences in apparent diffusion coefficient (ADC) values were noted (Table 2, Table S2). Inter‐reader agreement for MRI features ranged from 0.814 to 0.942 (Table S1).
3.5. BI‐RADS Reclassification With TUS and CEUS
In benign NME reassessment, TUS combined CEUS downgraded 36 lesions to BI‐RADS 3 (p < 0.001). For malignant NME, CEUS upgraded 34 cases to BI‐RADS 4c/5 compared to MRI and 45 compared to TUS (p < 0.001) (Table 4). Of 24 DCIS or DCIS with invasion rated as MRI BI‐RADS 4a, 87.5% (21/24) were reclassified to 4b or higher with TUS + CEUS (Table S3).
Inter‐reader agreement for BI‐RADS across modalities was good (kappa: 0.810–0.881, Table S1).
4. Discussion
This study demonstrates the value of combining TUS and CEUS for refining the diagnostic stratification of NME lesions initially classified as BI‐RADS 4 by DCE‐MRI. The combination reduced benign biopsy rates to 43.4% and overall biopsies by 23.1%, while maintaining high sensitivity (96.5%). CEUS further improved malignant NME detection, correctly reclassifying 87.5% of DCIS or DCIS with invasion from BI‐RADS 4a to 4b or higher, enhancing diagnostic confidence and biopsy targeting.
DCE‐MRI, despite high sensitivity, has variable positive predictive value for malignancy in NME (20.8%–93.8%), often overestimating malignant potential [18, 19]. Nonspecific enhancement patterns contribute to unnecessary biopsies, patient anxiety, and healthcare burden [5, 20, 21]. Subgroup analysis identified MRI overestimation factors: asymptomatic presentation, lesion size ≤2 cm, homogeneous enhancement, Type I kinetic curves, and BI‐RADS 4a classification, with biopsy reduction rates of 50%–71.4%.
These results indicate that TUS combined with CEUS effectively refines risk stratification and reduces overtreatment, consistent with reported benign NME rates of 16.4%–75% [22].
TUS alone detects BI‐RADS 4 NME effectively but has limited ability to downgrade lesions, likely due to the lack of specific BI‐RADS criteria for NMLs, leading to conservative reporting [23]. CEUS overcomes this by providing microvascular information, allowing confident downgrading of MLs with well‐defined morphology and nonmass or adenosis lesions lacking enhancement or peripheral vessels. Malignant lesions, including IDC and DCIS, exhibit abundant vascularity, increased peripheral vessels, lesion enlargement, and prolonged contrast retention; venous‐phase hyperintensity further enhances detection of subtle microvascular structures, improving lesion characterization and reducing unnecessary biopsies [24, 25].
Three false‐negative cases illustrate CEUS limitations in anatomically complex regions. Two lesions were missed, and one was incorrectly downgraded, involving occult or subareolar lesions, low‐grade IDC, Paget disease with high‐grade DCIS, and intermediate‐to‐high‐grade DCIS with invasive foci. These cases highlight challenges in detecting subtle lesions and emphasize the need for multimodal imaging and careful clinical assessment.
Limitations include the single‐center design, limited sample size, and potential selection bias. Variations in equipment and operator experience may affect reproducibility. Elastography was not incorporated, which could have further improved diagnostic accuracy.
CEUS did not provide additional findings for TUS‐invisible lesions, although prior studies suggest independent detection potential. Emerging methods such as deep learning and radiomics were not applied and warrant future investigation. Lesion localization relied on same‐day ultrasound‐guided skin marking rather than wire localization or clip placement, which may introduce greater spatial imprecision, particularly for small or deep lesions.
The distance‐based localization strategy was originally validated in predominantly mass‐type lesions; although NME‐specific studies support its general applicability, our high TUS detection rate (97.0%) should be regarded as cohort‐ and operator‐specific, likely reflecting our exclusive focus on BI‐RADS 4 lesions, dedicated experienced radiologists, and a multilandmark localization protocol. These findings may not be fully generalizable to other settings. Finally, 22 lesions were excluded due to suboptimal image quality; their notably lower malignancy rate (18.2% vs.
50.9% in the final cohort) should be considered when interpreting the generalizability of the reported diagnostic performance.
In conclusion, TUS combined with CEUS effectively reassesses BI‐RADS 4 NME, reducing unnecessary biopsies while maintaining high sensitivity, particularly for small, homogeneously enhancing, or BI‐RADS 4a lesions. Careful application is required in complex regions, and multicenter studies with standardized protocols are needed to optimize diagnostic accuracy and safety.
Author Contributions
Qinghua Niu and Ting Sun contributed equally to this work and are co‐first authors; An Chen and Fan Li contributed equally as co‐corresponding authors.
Funding
This study was supported by the Shanghai Chest Hospital Technology Development Fund, Shanghai Jiao Tong University K. C. Wong Medical Fellowship Fund, Shanghai Municipal Education Commission, and Songjiang District Science and Technology Research Project (2023SJKWGG042).
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting Information
Additional supporting information can be found online in the Supporting Information section.
Supporting information