Development and internal validation of a kinetic heterogeneity-based nomogram by dynamic contrast-enhanced magnetic resonance imaging to differentiate benign and malignant breast BI-RADS 4 lesions

Introduction

According to global epidemiological data from 2022, breast cancer is the second leading cause of global cancer incidence in 2022, with an estimated 2.3 million new cases, comprising 11.6% of all cancer cases (1). In recent years, both the incidence and mortality rates of breast cancer in China have shown a steady upward trend, with an increasingly younger age at onset. Notably, the mean age of diagnosis in China is approximately 49 years (2).

The Breast Imaging Reporting and Data System (BI-RADS), developed by the American College of Radiology (ACR), stratifies breast lesions based on imaging features such as lesion type, morphology, and enhancement characteristics, including enhancement patterns and signal intensity curves (3). Among these, BI-RADS 4 lesions carry a wide-ranging malignancy risk of 2% to 95%, indicating that a substantial proportion of benign lesions undergo unnecessary invasive procedures, including biopsies and surgeries, thereby imposing considerable psychological and economic burdens on patients (4). Thus, the development of a novel, non-invasive method to accurately distinguish benign from malignant BI-RADS 4 lesions remains a pressing challenge and a major focus in breast imaging research (5).

Although the multi-phase dynamic contrast-enhanced technique of breast DCE-MRI provides valuable morphological and microcirculatory perfusion data, its conventional quantitative parameters are limited in capturing global tumor heterogeneity (6,7). This limitation results in high diagnostic sensitivity but comparatively low specificity. With the development of artificial intelligence and radiomics, current efforts mainly focus on diffusion-weighted imaging (DWI), radiomics, or a combination of both to improve the diagnostic efficiency of distinguishing benign and malignant breast BI-RADS 4 lesions (8,9). However, radiomics requires special software, it is complex to operate and has poor reproducibility, as well as some features have unclear clinical significance and weak practicality.

Kinetic analysis with dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI), as an emerging computer-aided diagnostic (CAD) tool, provides a quantitative assessment of intratumoral hemodynamic properties (10). Unlike conventional DCE-MRI parameters, kinetic heterogeneity analysis leverages advanced computational methods to capture pixel-wise temporal changes in contrast enhancement, enabling precise volumetric quantification of early and delayed tumor enhancement and incorporating measures of MRI-based intratumoral heterogeneity (11).

As a novel non-invasive and quantitative technique, DCE-MRI kinetic heterogeneity analysis offers an objective and reproducible approach to evaluating tumor-wide hemodynamic characteristics, independent of radiologists’ subjective judgment (12). Prior studies have demonstrated that kinetic heterogeneity parameters can reflect histopathological and prognostic information in breast cancer. For instance, Nam et al. (13) have reported correlations between kinetic parameters and tumor malignancy as well as histological grade. Kim et al. (10) have revealed that certain kinetic features are associated with poor disease-free survival. Furthermore, Yao et al. (11) have found that kinetic heterogeneity parameters are statistically significant in differentiating benign from malignant breast lesions and show high interobserver agreement. Nevertheless, their studies are limited by a small sample size and the evaluation of only a limited set of kinetic parameters, highlighting the need for comprehensive analysis that incorporates BI-RADS classification and clinical imaging risk factors.

In this study, we aimed to investigate whether kinetic parameters extracted from DCE-MRI could serve as reliable preoperative biomarkers to differentiate benign from malignant BI-RADS 4 breast lesions. Moreover, by integrating clinical risk factors, we sought to establish a nomogram to improve predictive performance and provide a practical tool for individualized risk assessment. We present this article in accordance with the TRIPOD reporting checklist (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-314/rc).

Methods

Study population

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of the Affiliated Tumor Hospital of Nantong University (No. 2018-003-01). Informed consent was waived in this retrospective study. Female patients who underwent preoperative DCE-MRI and were diagnosed with BI-RADS 4 breast lesions between January 2018 and June 2023, with definitive histopathological results, were consecutively identified from the institutional database. BI-RADS classification was assigned according to the 5th edition of the ACR BI-RADS guidelines published in 2013 (3).

Female patients were included if (I) they underwent their first DCE-MRI scan of the breast and were classified as BI-RADS 4; and (II) if they had a definitive pathological diagnosis confirmed through surgical excision. Patients were excluded if (I) their breast lesions were too small to be reliably identified on MRI; (II) if the image quality was poor or artifacts interfered with diagnostic interpretation; (III) if they had a prior history of radiotherapy, endocrine therapy, or chemotherapy; or (IV) if their clinical or pathological data were incomplete. Ultimately, 271 patients were included in the study. A detailed flow chart of patient enrollment for the study is shown in Figure 1.

Figure 1 The flow chart of inclusion and exclusion criteria. BI-RADS, Breast Imaging Reporting and Data System; DCE-MRI, dynamic contrast-enhanced magnetic resonance imaging.

DCE-MRI examination

All MRI examinations were performed using a 3.0TMR scanner (Magnetom Verio DOT, Siemens Healthcare, Erlangen, Germany) equipped with an eight-channel dedicated breast coil. Patients were positioned prone, with both breasts naturally suspended in the openings of the coil. The imaging protocol included conventional pre-contrast sequences, DWI, followed by DCE-MRI.

The scanning parameters were set as follows to ensure optimal image quality: an axial T2-weighted sequence with a repetition time (TR)/echo time (TE) of 4,940/63 ms, matrix size of 320×256, field of view (FOV) of 350×350 mm, and a slice thickness of 3.0 mm; an axial T1-weighted fat-suppressed three-dimensional (3D) fast low-angle shot (FLASH) sequence with TR/TE of 5.5/2.5 ms, matrix size of 320×320, a flip angle of 12°, FOV of 350 mm × 350 mm, and a slice thickness of 1.2 mm; and DWI acquired using b values of 0 and 800 s/mm², with an FOV of 350 mm × 350 mm and a slice thickness of 4 mm.

For DCE-MRI, gadopentetate dimeglumine (Gd-DTPA; Bayer, Leverkusen, Germany) was intravenously administered at a dose of 0.1 mmol/kg using a power injector at a flow rate of 2 mL/s, followed by a 20 mL saline flush to ensure complete delivery. A baseline unenhanced scan was performed prior to contrast injection. Subsequently, five consecutive post-contrast phases were acquired, each with a temporal resolution of 60 s, yielding a total of five dynamic series. All imaging data were systematically archived in the picture archiving and communication system (PACS) for detailed post-processing and analysis.

Assessment of breast MRI features

All MRI features were independently evaluated by two experienced breast radiologists (Observer A and Observer B), with 7 and 10 years of expertise in breast imaging interpretation, respectively, and any discrepancies were resolved through consensus discussion. The evaluated MRI features are described as follows.

• Enhancement type: categorized as mass or non-mass enhancement. Mass enhancement is defined as a space-occupying lesion with solid components, with or without displacement or infiltration of surrounding normal tissue. Non-mass enhancement is defined as enhancement without a distinct mass effect, which is neither focal (<5 mm enhancement) nor a mass, and may present as linear, focal, segmental, regional, multiple regional, or diffuse enhancement.
• Background parenchymal enhancement (BPE): minimal if less than 25% of the glandular tissue enhances; mild if 25–50% enhances; moderate if 51–75% enhances; and marked if more than 75% enhances.
• Shape: classified as round, oval, or irregular.
• Maximum tumor diameter: the largest diameter of the largest tumor measured on the axial enhanced sequence at the slice showing the maximum tumor extent.
• Tumor margin: defined as circumscribed or non-circumscribed. A circumscribed margin appears smooth in contour, while a non-circumscribed margin presents as lobulated or spiculated.
• Time-intensity curve (TIC) type I (persistent): continuous increase in signal intensity with an increase of more than 10% beyond the initial enhancement point. Type II (plateau): initial uptake followed by a plateau phase where the signal intensity does not vary by more than 10% from the peak initial enhancement. Type III (washout): uptake followed by a decrease in signal intensity of more than 10% below the peak.
• Internal enhancement characteristics: classified as homogeneous or heterogeneous.
• Apparent diffusion coefficient (ADC): the ADC value was measured by first identifying the largest cross-sectional area of the lesion on the enhanced images. A region of interest (ROI) was manually drawn on this section within the lesion boundaries. Care was taken to avoid areas of necrosis, hemorrhage, or artifacts. The ROI was placed on areas showing high signal on DWI and corresponding low signal on the ADC map. The measurement was repeated three times for the same mass, and the final average value was recorded.
• Peritumoral edema: defined as the presence of high signal intensity surrounding the tumor on T2-weighted MRI images, indicating fluid accumulation.
• MRI axillary lymph node status: an axillary lymph node was classified as positive on MRI if it met any of the following criteria: short-axis diameter greater than 10 mm, a long-axis to short-axis ratio of less than 1.5, loss of the fatty hilum, or eccentric cortical thickening.

Extraction of DCE-MRI kinetic analysis parameters

ROI delineation

The DCE-MRI images were exported in DICOM format and uploaded to ITK-SNAP version 3.8.0 (http://www.itksnap.org/), where Observers A and B independently delineated ROIs on the first post-contrast DCE-MRI phase (Figure 2A). For patients with a single lesion, slice-by-slice segmentation was conducted to encompass the entire tumor volume, meticulously excluding necrotic, cystic, or hemorrhagic areas. In cases involving multifocal lesions, only the largest lesion was selected for analysis.

Figure 2 Kinetic heterogeneity analysis maps. (A) ROI delineation. (B) Kinetic heterogeneity analysis maps, where persistent, plateau, and washout voxels are represented in blue, green, and red, respectively. ROI, region of interest.

Extraction of kinetic analysis parameters

The delineated ROIs, along with the corresponding DCE-MRI images, were subsequently imported into MATLAB (version 2021a) and SPM12 software for further analysis. A dedicated script was employed to extract seven kinetic heterogeneity parameters, including persistent component (%), plateau component (%), washout component (%), volume, predominant, peak, and heterogeneity, with detailed descriptions provided in Table 1. The extraction process involved four steps: (I) identifying enhancing voxels within the ROI, defined as those showing a signal intensity increase greater than 50% compared to the pre-contrast scan; (II) classifying voxel enhancement patterns based on the change in signal intensity from the first to the last post-contrast phase, persistent (>10% increase, marked in blue), washout (>10% decrease, marked in red), and plateau (signal change between −10% and +10%, marked in green) (Figure 2B); (III) subdividing the tumor into three sub-regions according to voxel enhancement types; and (IV) calculating kinetic parameters based on these sub-regions and their relative proportions. These kinetic parameters provide a quantitative and comprehensive evaluation of tumor perfusion and intratumoral heterogeneity, reflecting distinct hemodynamic characteristics, with detailed computational methods summarized in Table 1.

Inter- and intra-observer consistency analysis

For kinetic heterogeneity parameters, the two radiologists independently delineated the whole-tumor ROIs and extracted the corresponding kinetic parameters to assess inter-observer agreement. To evaluate intra-observer agreement, the senior radiologist repeated the ROI delineation and parameter extraction 1 week later.

For breast MRI features, the same two radiologists independently evaluated all imaging features to assess inter-observer consistency. One week later, the senior radiologist re-evaluated all images to determine intra-observer consistency. The consistency of categorical variables between observers was assessed using Cohen’s Kappa test. For continuous variables, inter-class/intra-class correlation coefficient (ICC) analyses were applied to evaluate agreement both between and within observers under identical imaging conditions. An ICC value greater than 0.75 was considered indicative of good consistency, and a Kappa value above 0.75 indicated excellent agreement.

Statistical analysis

All statistical analyses were performed using SPSS version 26.0 (IBM, Armonk, NY, USA) and R software version 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria). The Shapiro-Wilk test was applied to assess the normality of continuous variables. Data conforming to a normal distribution were presented as mean ± standard deviation (SD), and comparisons between groups were conducted using independent sample t-tests. Data not following a normal distribution were expressed as median with interquartile range (IQR), and the Mann-Whitney U test was employed for between-group comparisons.

Categorical variables were expressed as counts and percentages [n (%)]. Comparisons of nominal categorical variables between groups were performed using the Chi-squared (χ²) test, whereas ordinal categorical variables were analyzed using the rank-sum test (Wilcoxon rank test). The diagnostic performance of each parameter was assessed using receiver operating characteristic (ROC) curve analysis, and the area under the curve (AUC) was calculated to evaluate discriminatory ability.

Univariate and multivariate logistic regression analyses were conducted to identify independent predictors of malignancy in BI-RADS 4 lesions. Based on these predictors, predictive models were established and visualized as nomograms. Internal validation of the models was performed via bootstrap resampling (1,000 iterations) to evaluate model stability and reduce overfitting risk. Model calibration was assessed using the Hosmer-Lemeshow goodness-of-fit (GOF) test. To evaluate the clinical utility and net benefit of the models, decision curve analysis (DCA) and clinical impact curve (CIC) were employed. All statistical tests were two-sided, and a P value <0.05 was considered indicative of statistical significance.

Results

Inter- and intra-observer consistency

The inter-observer agreement between Observer A and Observer B for quantitative measurements, including ADC values, maximum tumor diameter, and DCE-MRI kinetic heterogeneity parameters, peak, volume, persistent component (%), plateau component (%), washout component (%), and heterogeneity, was excellent, with ICCs all exceeding 0.8, indicating high reliability.

In the intra-observer analysis, measurements obtained by Observer B in two separate assessments (1 week apart) demonstrated even higher agreement, with ICCs exceeding 0.9, and most ICC values were higher than those observed in the inter-observer analysis, reflecting superior consistency within the same observer.

For qualitative MRI features, including enhancement pattern, BPE, lesion morphology, tumor margins, internal enhancement characteristics, TIC type, peritumoral edema, MRI-assessed axillary lymph node status, as well as kinetic parameters (predominant and worst), the Kappa values for inter-observer agreement were all greater than 0.9, indicating outstanding consistency.

Similarly, intra-observer agreement assessed from the repeated evaluations by Observer B also showed Kappa values above 0.9 across all evaluated features. These results demonstrated excellent reproducibility and robust reliability of both the kinetic parameter extraction and MRI feature evaluation processes.

Comparison of clinical and radiological characteristics

A total of 271 breast lesions, comprising 84 benign and 187 malignant cases, were included in this study. The benign lesions consisted of 42 fibroadenomas, 13 adenoses, 14 intraductal papillomas, six inflammatory lesions, two sclerosing adenoses, and seven other benign entities, while the malignant lesions included 146 invasive ductal carcinomas (IDCs), 10 ductal carcinomas in situ (DCIS), five mucinous carcinomas, six invasive lobular carcinomas (ILCs), and 20 other malignant tumors.

All patients were randomly divided into a training set (n=192) and a validation set (n=79) using a 7:3 ratio. Comparative analyses revealed that TIC type, peritumoral edema, ADC values, and maximum tumor diameter were significantly different between benign and malignant lesions in both the training and validation sets (P<0.05), indicating strong discriminatory potential. In contrast, age and MRI-assessed axillary lymph node status showed significant differences between benign and malignant lesions in the training set (P<0.05) but not in the validation set (P>0.05), suggesting possible variability between cohorts. Moreover, menstrual status, enhancement pattern, tumor morphology, tumor margins, internal enhancement characteristics, and BPE did not significantly differ between benign and malignant lesions in either set (P>0.05), indicating limited value of these features for lesion differentiation. Detailed comparisons of clinical and imaging characteristics are presented in Table 2.

Comparison of kinetic heterogeneity parameters

The comparison of DCE-MRI kinetic heterogeneity parameters between benign and malignant BI-RADS 4 lesions is summarized in Table 3. Mann-Whitney U tests revealed that the peak, plateau component, washout component, and heterogeneity values were significantly higher in the malignant group compared to the benign group (P<0.05). Conversely, the persistent component was significantly lower in malignant lesions than in benign ones (P<0.05), indicating that malignant lesions were more likely to exhibit rapid washout and greater intratumoral heterogeneity. In contrast, volume and predominant did not show significant differences between the benign and malignant groups (P>0.05), suggesting these parameters might have limited value in distinguishing lesion malignancy within BI-RADS 4. These findings highlighted that certain kinetic parameters, particularly those reflecting dynamic enhancement patterns and heterogeneity, were more strongly associated with malignancy, and could serve as valuable indicators for risk stratification in BI-RADS 4 lesions.

The results of univariate logistic regression analysis for the selection of predictive factors

The univariate logistic regression analysis of factors associated with the malignancy of BI-RADS 4 lesions is presented in Table 4. Malignancy status (malignant =1, benign =0) was used as the dependent variable, and independent variables included peak, volume, heterogeneity, predominant, worst, age, menstrual status, enhancement type, background enhancement, tumor shape, maximum tumor diameter, tumor margin, ADC value, TIC type, internal enhancement characteristics, peritumoral edema, and MRI axillary lymph node status. The results indicated that peak, heterogeneity, age, maximum tumor diameter, ADC value, TIC type, peritumoral edema, and MRI axillary lymph node status were all significantly associated with the malignant potential of BI-RADS 4 lesions (P<0.05).

Results of multivariate logistic regression analysis for the selection of predictive factors

The multivariate logistic regression analysis identifying factors associated with malignancy in BI-RADS 4 lesions is presented in Table 5. Malignancy status (malignant =1, benign =0) was set as the dependent variable, while variables with P<0.05 from univariate analysis, including peak, heterogeneity, age, maximum tumor diameter, ADC value, TIC pattern, peritumoral edema, and MRI axillary lymph node status, were included as independent variables. A binary logistic regression model was constructed using the forward conditional method (entry criterion: P=0.05; removal criterion: P=0.10). The analysis revealed that peak intensity, heterogeneity, ADC value, TIC pattern, and peritumoral edema were independently associated with malignancy (P<0.05). Specifically, higher peak intensity, greater heterogeneity, plateau or washout TIC patterns, and the presence of peritumoral edema emerged as significant risk factors for malignancy in BI-RADS 4 lesions. Conversely, higher ADC values were identified as a protective factor favoring benign pathology. Collectively, these findings indicated that elevated peak and heterogeneity values, adverse TIC patterns, peritumoral edema, and reduced ADC values substantially increased the likelihood of malignancy in BI-RADS 4 lesions.

Evaluation of the predictive performance of peak, heterogeneity, and ADC values

The predictive performance of peak intensity, heterogeneity, and ADC values for distinguishing malignant BI-RADS 4 lesions was assessed using ROC curve analysis, as summarized in Table 6 and Figure 3. The AUC for peak, heterogeneity, and ADC value were 0.793 [95% confidence interval (CI): 0.723–0.863], 0.816 (95% CI: 0.750–0.881), and 0.773 (95% CI: 0.704–0.842), respectively, with all corresponding P values less than 0.05, indicating that each parameter possessed significant discriminatory power for malignancy. The optimal threshold for peak in predicting malignancy was 2.099, yielding a sensitivity of 77.4% and specificity of 71.2%. For heterogeneity, the optimal cut-off was 0.097, with a sensitivity of 91.0% and specificity of 61.0%. The optimal threshold for ADC value was 1.088×10⁻³ mm²/s, corresponding to a sensitivity of 74.6% and specificity of 72.9%. These results highlighted that peak, heterogeneity, and ADC value were valuable imaging biomarkers for predicting malignancy in BI-RADS 4 lesions, with heterogeneity demonstrating the highest sensitivity and ADC showing a balanced sensitivity-specificity profile.

Figure 3 Receiver operating characteristic curves of individual indicators for predicting malignancy in BI-RADS 4 lesions. ADC, apparent diffusion coefficient; AUC, area under the curve; BI-RADS, Breast Imaging Reporting and Data System.

Construction of a combined nomogram

A combined nomogram was developed by integrating dynamic analysis parameters (peak intensity and heterogeneity) along with key MRI features (TIC pattern, ADC value, and peritumoral edema). Each variable corresponded to a specific score on the “Points” scale within the nomogram. The sum of these individual scores generated a total score, which reflected the predicted risk of malignancy in BI-RADS 4 lesions, as illustrated in Figure 4.

Figure 4 Nomogram for predicting malignancy in BI-RADS 4 lesions. ADC, apparent diffusion coefficient; BI-RADS, Breast Imaging Reporting and Data System; TIC, time-intensity curve; Y, malignancy.

Evaluation of predictive performance across different models

The predictive performance of the three models for identifying malignancy in BI-RADS 4 lesions was assessed using ROC curve analysis, as presented in Tables 7,8 and shown in Figure 5. The AUC values for the combined nomogram, dynamic analysis parameter model, and clinical-radiological parameter model were 0.928 (95% CI: 0.893–0.963), 0.863 (95% CI: 0.807–0.920), and 0.819 (95% CI: 0.754–0.885), respectively. All corresponding P values were less than 0.05, indicating that each model demonstrated significant predictive value for distinguishing malignant from benign BI-RADS 4 lesions.

Figure 5 Receiver operating characteristic curves of the combined nomogram, dynamic analysis parameter model, and clinical-radiological parameter model for predicting malignancy in BI-RADS 4 lesions. AUC, area under the curve; BI-RADS, Breast Imaging Reporting and Data System.

Furthermore, Z-test comparisons of AUCs revealed that the nomogram’s AUC was significantly higher than that of both the dynamic analysis parameter model and the clinical-radiological parameter model. The differences between the nomogram and the other two models were statistically significant (P<0.05), underscoring the superior discriminatory power of the nomogram. Representative case examples are shown in Figures 6,7.

Figure 6 A 50-year-old female patient with a benign lesion (fibroadenoma). (A) DCE-MRI images. (B) ROI delineation. (C) Kinetic analysis map derived from DCE-MRI, where persisten, plateau, and vashout voxels are represented in blue, green, and red, respectively. (D) Nomogram. The lesion demonstrated a peak enhancement ratio of 2.49, heterogeneity index of 0.41, and ADC value of 1.38×10⁻³ mm²/s, with no evidence of peritumoral edema, and exhibited a plateau-type TIC curve. The predicted probability of malignancy was approximately 67.1%, which was below the diagnostic threshold of 76.4%, and the nomogram-based prediction was fully consistent with the pathological diagnosis. ADC, apparent diffusion coefficient; DCE-MRI, dynamic contrast-enhanced magnetic resonance imaging; ROI, region of interest; TIC, time-intensity curve; Y, malignancy.

Figure 7 A 67-year-old female patient with an invasive breast carcinoma. (A) DCE-MRI images. (B) ROI delineation. (C) Kinetic analysis map derived from DCE-MRI, where persisten, plateau, and vashout voxels are represented in blue, green, and red, respectively. (D) Nomogram. The lesion exhibited a peak enhancement ratio of 2.26, heterogeneity index of 0.964, and ADC value of 0.68×10⁻³ mm²/s, accompanied by peritumoral edema and a washout-type TIC curve. The predicted probability of malignancy was approximately 98.8%, substantially exceeding the diagnostic threshold of 76.4%, and fully concordant with the pathological diagnosis. ADC, apparent diffusion coefficient; DCE-MRI, dynamic contrast-enhanced magnetic resonance imaging; ROI, region of interest; TIC, time-intensity curve; Y, malignancy.

Validation of the nomogram

The ROC curves for the nomogram in both the training and validation cohorts are presented in Figure 8. In the training cohort, the nomogram achieved an AUC of 0.928 (95% CI: 0.893–0.963), with an optimal diagnostic threshold of 0.764, corresponding to a sensitivity of 77.4% and a specificity of 91.5%, indicating a high predictive value for identifying malignancy in BI-RADS 4 lesions.

Figure 8 Receiver operating characteristic curves of the nomogram for predicting malignancy in BI-RADS 4 lesions. (A) Training set. (B) Validation set. AUC, area under the curve; BI-RADS, Breast Imaging Reporting and Data System.

To further assess its generalizability, the model was validated in an independent cohort, yielding an AUC of 0.906 (95% CI: 0.841–0.971). The optimal diagnostic threshold in the validation cohort was 0.706, with a sensitivity of 73.6% and a specificity of 86.3%. These results demonstrated that the nomogram maintained robust discriminatory power in distinguishing malignant from benign BI-RADS 4 lesions across different patient populations.

Calibration curve of the nomogram

The calibration curves for the nomogram in both the training and validation cohorts are shown in Figure 9. The model’s GOF was evaluated using 1,000 bootstrap resamples, and the Hosmer-Lemeshow test in the training cohort yielded a χ² value of 4.583 with a P value of 0.801, indicating an excellent fit to the data. Additionally, the C-index of the model was 0.928, exceeding the threshold of 0.7, which reflected strong discriminative capability.

Figure 9 Calibration curves of the predictive model. (A) Training set. (B) Validation set.

In the validation cohort, the Hosmer-Lemeshow test demonstrated a χ² value of 8.022 with a P value of 0.431, further confirming the good calibration and stability of the nomogram when applied to an independent patient population.

Evaluation of the clinical utility of the nomogram

The DCA and CIC of the nomogram are presented in Figures 10,11, respectively. Both analyses demonstrated that the nomogram exhibited favorable clinical performance and robust predictive ability, indicating its potential value in guiding clinical decision-making for the differentiation of malignant and benign BI-RADS 4 lesions.

Figure 10 Decision curve analysis of the nomogram. (A) Training set. (B) Validation set.

Figure 11 Clinical impact curve of the nomogram. (A) Training set. (B) Validation set.

Discussion

This study systematically evaluated the utility of DCE-MRI kinetic parameters in distinguishing malignant from benign BI-RADS 4 breast lesions. Our findings demonstrated that peak intensity and heterogeneity, as derived from DCE-MRI kinetic analysis, were independent risk factors for predicting malignancy in BI-RADS 4 lesions. The kinetic analysis model based solely on these parameters achieved an AUC of 0.863 (95% CI: 0.807–0.920) for malignancy prediction. Furthermore, by integrating kinetic parameters with key clinical-radiological features, including peritumoral edema, ADC value, and TIC pattern, we developed a comprehensive nomogram that significantly enhanced predictive performance, reaching an AUC of 0.928 (95% CI: 0.893–0.963) in the training cohort, which was successfully validated in an independent cohort (AUC =0.906; 95% CI: 0.841–0.971). This nomogram offered a non-invasive and individualized preoperative tool for accurately stratifying the risk of malignancy in BI-RADS 4 breast lesions.

Our analysis identified peak intensity and heterogeneity as robust predictors of malignancy, with significantly higher values observed in malignant compared to benign BI-RADS 4 lesions. When analyzed independently, peak and heterogeneity demonstrated respectable discriminative abilities, with AUCs of 0.793 (95% CI: 0.723–0.863) and 0.816 (95% CI: 0.750–0.881), respectively. Peak intensity reflects the maximum relative enhancement ratio of the tumor during the early phase following contrast administration, indicating the rate and extent of contrast agent uptake. This parameter captures key vascular features of malignant tumors, such as rich neovascularization, disrupted endothelial barriers, and increased permeability, all of which facilitate rapid and substantial contrast inflow and accumulation, resulting in markedly higher signal intensities and enhancement rates (14).

On the other hand, heterogeneity quantifies the spatial variation in voxel-wise kinetic patterns (persistent, plateau, and washout components), providing a direct measure of the internal complexity and heterogeneity of tumor vasculature and histopathology (10). Tumors with extensive angiogenesis, necrosis, and mixed cellular components typically exhibit greater heterogeneity, reflecting more aggressive biological behavior. Notably, Yao et al. (11) have similarly reported that both peak and heterogeneity values are significantly elevated in malignant breast lesions (P<0.05), with individual AUCs of 0.73 (95% CI: 0.66–0.82) and 0.92 (95% CI: 0.88–0.97), respectively, for malignancy prediction.

Compared to previous studies, our research incorporated a larger cohort specifically focusing on BI-RADS 4 lesions, which are notoriously challenging to classify due to their wide spectrum of pathologies. Importantly, by developing a combined nomogram that integrated DCE-MRI kinetic parameters and clinical- radiological features, we achieved superior predictive accuracy, as reflected in both the training (AUC =0.928; 95% CI: 0.893–0.963) and validation (AUC =0.906; 95% CI: 0.841–0.971) cohorts.

Moreover, previous studies have linked peak intensity to poor clinical outcomes in breast cancer. For instance, an earlier study (10) has demonstrated that higher peak values are associated with shorter disease-free survival, highlighting its prognostic relevance. Similarly, Nam et al. (13) have reported that elevated peak values correlate with advanced clinical stages and poorer histological grades of breast cancer, further supporting its role as an indicator of tumor aggressiveness. In addition, Kim et al. (15) have found that increased heterogeneity is significantly associated with triple-negative breast cancer (TNBC) and human epidermal growth factor receptor 2 (HER2)-positive subtypes, both known for their aggressive behavior, indicating that higher intratumoral heterogeneity may reflect the biological complexity of invasive tumor subtypes.

Consistent with these prior findings, our study underscored that higher peak intensity and greater heterogeneity were critical risk factors for malignancy in BI-RADS 4 lesions. By incorporating these parameters into a comprehensive predictive model, our nomogram offered a powerful tool for enhancing preoperative decision-making. It might help reduce unnecessary invasive procedures for benign lesions while ensuring timely intervention for malignant cases.

In previous studies, quantitative parameters derived from DCE-MRI have been widely used to assess the hemodynamic characteristics of breast lesions. These quantitative parameters, based on pharmacokinetic modeling, provide a quantitative assessment of contrast agent exchange rates between the intravascular, extravascular, and interstitial spaces, thereby offering critical information on tissue microcirculation and capillary permeability (16). A meta-analysis by Arian et al. (17), which includes 10 studies encompassing 537 patients and 707 lesions (435 malignant and 272 benign), has demonstrated that DCE-MRI quantitative parameters such as Ktrans and Kep differ significantly between benign and malignant lesions. The combined model of these parameters achieves a sensitivity of 93.8% (95% CI: 85.3–97.5%) and a specificity of 68.1% (95% CI: 52.7–80.4%). However, it is important to note that quantitative parameters are typically derived from specific tumor regions and therefore may not fully represent the overall hemodynamic characteristics of the tumor (18). Furthermore, their calculation requires high temporal resolution imaging and complex post-processing, which may reduce spatial resolution and prolong examination time.

In contrast, our study demonstrated that the kinetic analysis model based on DCE-MRI parameters achieved an AUC of 0.863 (95% CI: 0.807–0.920), with a sensitivity of 72.91% and a specificity of 88.1%, indicating that DCE-MRI kinetic parameters provided comparable predictive performance to quantitative parameters. Notably, DCE-MRI kinetic analysis is less reliant on complex post-processing, requires lower temporal resolution, preserves high spatial resolution, and is easily accessible with standard imaging software (12). Moreover, beyond capturing overall tumor hemodynamics, kinetic analysis can also offer valuable insights into intratumoral heterogeneity (19).

Among the clinical-radiological features analyzed in this study, ADC value, peritumoral edema, and TIC pattern were identified as independent predictors for distinguishing malignant from benign BI-RADS 4 lesions. The model incorporating these MRI features yielded an AUC of 0.819 (95% CI: 0.754–0.885), outperforming each feature when considered individually. The ADC value serves as a quantitative biomarker reflecting microscopic structural alterations in biological tissues and is widely recognized as an effective indicator for differentiating benign and malignant breast lesions (20). Previous studies (21) have shown that, at b =800 s/mm², ADC values typically follow the pattern: normal breast tissue > benign lesions > malignant lesions. Additionally, Clauser et al. (5) have reported that ADC value effectively differentiates benign from malignant BI-RADS 4 lesions, with a diagnostic threshold of 1.5×10⁻³ mm²/s. In line with these findings, our study identified an optimal ADC threshold of 1.088×10⁻³ mm²/s, yielding an AUC of 0.773 (95% CI: 0.704–0.842), with a sensitivity of 74.6% and a specificity of 72.9% for predicting malignancy in BI-RADS 4 lesions.

The biological mechanism underlying peritumoral edema and its association with malignancy remains incompletely understood. It is hypothesized that the release of vascular endothelial growth factor (VEGF) and tumor-associated cytokines, which increase vascular permeability, may induce localized interstitial fluid accumulation around the tumor (22). Furthermore, peritumoral edema has been reported to correlate strongly with lymphovascular invasion in invasive breast cancer, particularly in IDC (23). In our multivariate logistic regression analysis, peritumoral edema emerged as a significant predictor of malignancy, supporting its clinical relevance in risk stratification.

Finally, TIC pattern, which reflects the hemodynamic behavior of lesions, also served as an independent predictor in our study. Typically, benign lesions exhibit a type I (persistent) TIC, whereas malignant lesions show a type III (washout) pattern (21). Consistent with this, our results identified type II (plateau) and type III (washout) TIC patterns as significant risk factors for malignancy in BI-RADS 4 lesions, which is in agreement with previous literature.

Compared with individual imaging parameters, the integration of multiple complementary indicators provides superior predictive value. Therefore, in this study, we combined both DCE-MRI kinetic parameters and clinical-radiological features into a unified model for comprehensive analysis. The results demonstrated that this integrated model significantly improved the predictive performance for identifying malignancy in BI-RADS 4 lesions, achieving an AUC of 0.928 (95% CI: 0.893–0.963) in the training cohort, which notably surpassed the performance of models based solely on kinetic parameters or clinical-radiological features.

To facilitate clinical interpretation and ensure the intuitive presentation of these findings, we developed a nomogram based on the integrated model. As a powerful graphical tool, the nomogram visually represents the contribution of each variable through a system of aligned vertical lines and point scales, clearly illustrating the relationships among variables. This visual approach aims to provide clinicians and researchers with an accessible and practical reference, enabling a deeper understanding of our findings and supporting the translation of these insights into clinical decision-making.

There are also several limitations in this study. First, as a single- center retrospective study performed on the same MRI platform, the relatively small sample size might introduce selection bias. Second, although internal calibration was performed using 1,000 bootstrap resamples, external validation with independent datasets was not conducted, limiting the assessment of model generalizability. Third, manual delineation of ROIs inevitably introduced a degree of observer subjectivity, potentially affecting the consistency of ROI definition. Finally, this study did not utilize commercial CAD systems (e.g., CADstream, Confirma), but instead relied on MATLAB and SPM software for image processing. Although flexible, these platforms might lack the standardization and user-friendliness of commercial CAD solutions. Therefore, large-scale, multicenter prospective studies incorporating standardized CAD systems are necessary to validate and extend our findings in broader clinical contexts.

Conclusions

In summary, this study demonstrated that the integration of DCE-MRI kinetic parameters (peak intensity and heterogeneity) with critical clinical-radiological features (ADC value, TIC pattern, and peritumoral edema) offered a robust and non-invasive approach for differentiating malignant from benign BI-RADS 4 breast lesions. By combining these complementary imaging biomarkers, we developed a highly accurate predictive nomogram, which significantly outperformed models based on individual parameters, achieving an AUC of 0.928 (95% CI: 0.893–0.963) in the training cohort and 0.906 (95% CI: 0.841–0.971) in the validation cohort.

Importantly, compared to traditional DCE-MRI quantitative analysis, kinetic parameters provide a more accessible, efficient, and clinically practical solution, circumventing the need for complex pharmacokinetic modeling and time-consuming post-processing. Moreover, the integration of kinetic heterogeneity measures allows for a more comprehensive assessment of tumor vascular dynamics and microenvironmental complexity, capturing intratumoral variability that might underlie aggressive biological behavior.

By visualizing this model through an easy-to-use nomogram, it offers clinicians a powerful decision-support tool that can guide individualized risk assessment and management strategies in patients with BI-RADS 4 lesions. This approach has the potential to reduce unnecessary biopsies in benign cases and facilitate early diagnosis and timely intervention for malignant lesions.

Nevertheless, large-scale multicenter prospective validation studies are warranted further to confirm the clinical utility and generalizability of this model. Future work should also explore the integration of automated CAD systems to standardize feature extraction and improve reproducibility in routine clinical workflows.

Collectively, our findings highlight the value of advanced multiparametric MRI analysis in enhancing breast cancer diagnosis and represent a promising step toward precision imaging and personalized clinical decision-making in breast cancer care.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-314/rc

Data Sharing Statement: Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-314/dss

Peer Review File: Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-314/prf

Funding: This study was supported by Research topic of Nantong City Health Commission (Nos. QN2022029 and QNZ2024041).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-314/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of the Affiliated Tumor Hospital of Nantong University (No. 2018-003-01). Informed consent was waived in this retrospective study.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229-63. [Crossref] [PubMed]
2. Hong R, Xu B. Breast cancer: an up-to-date review and future perspectives. Cancer Commun (Lond) 2022;42:913-36. [Crossref] [PubMed]
3. Edwards SD, Lipson JA, Ikeda DM, et al. Updates and revisions to the BI-RADS magnetic resonance imaging lexicon. Magn Reson Imaging Clin N Am 2013;21:483-93. [Crossref] [PubMed]
4. Strigel RM, Burnside ES, Elezaby M, et al. Utility of BI-RADS Assessment Category 4 Subdivisions for Screening Breast MRI. AJR Am J Roentgenol 2017;208:1392-9. [Crossref] [PubMed]
5. Clauser P, Krug B, Bickel H, et al. Diffusion-weighted Imaging Allows for Downgrading MR BI-RADS 4 Lesions in Contrast-enhanced MRI of the Breast to Avoid Unnecessary Biopsy. Clin Cancer Res 2021;27:1941-8. [Crossref] [PubMed]
6. Ao F, Yan Y, Zhang ZL, et al. The value of dynamic contrast-enhanced magnetic resonance imaging combined with apparent diffusion coefficient in the differentiation of benign and malignant diseases of the breast. Acta Radiol 2022;63:891-900. [Crossref] [PubMed]
7. Zhou Y, Li Y, Liu Y, et al. The value of contrast-enhanced energy-spectrum mammography combined with clinical indicators in detecting breast cancer in Breast Imaging Reporting and Data System (BI-RADS) 4 lesions. Quant Imaging Med Surg 2024;14:8272-80. [Crossref] [PubMed]
8. Han C, Chen J, Hong M, et al. MRI radiomics for diagnosing small BI-RADS 4 breast lesions: an interpretable model. Quant Imaging Med Surg 2025;15:5060-72. [Crossref] [PubMed]
9. Zhang B, Guo Z, Chen X, et al. Combining a breast apparent diffusion coefficient category system with Breast Imaging Reporting and Data System assessment improves specificity of breast lesions diagnosis. Br J Radiol 2025;98:1080-9. [Crossref] [PubMed]
10. Kim JJ, Kim JY, Kang HJ, et al. Computer-aided Diagnosis-generated Kinetic Features of Breast Cancer at Preoperative MR Imaging: Association with Disease-free Survival of Patients with Primary Operable Invasive Breast Cancer. Radiology 2017;284:45-54. [Crossref] [PubMed]
11. Yao Y, Mou F, Kong J, et al. Kinetic Heterogeneity Improves the Specificity of Dynamic Enhanced MRI in Differentiating Benign and Malignant Breast Tumours. Acad Radiol 2024;31:812-21. [Crossref] [PubMed]
12. Kim JY, Kim JJ, Hwangbo L, et al. Kinetic Heterogeneity of Breast Cancer Determined Using Computer-aided Diagnosis of Preoperative MRI Scans: Relationship to Distant Metastasis-Free Survival. Radiology 2020;295:517-26. [Crossref] [PubMed]
13. Nam SY, Ko ES, Lim Y, et al. Preoperative dynamic breast magnetic resonance imaging kinetic features using computer-aided diagnosis: Association with survival outcome and tumor aggressiveness in patients with invasive breast cancer. PLoS One 2018;13:e0195756. [Crossref] [PubMed]
14. Xiao J, Rahbar H, Hippe DS, et al. Dynamic contrast-enhanced breast MRI features correlate with invasive breast cancer angiogenesis. NPJ Breast Cancer 2021;7:42. [Crossref] [PubMed]
15. Kim JJ, Kim JY, Suh HB, et al. Characterization of breast cancer subtypes based on quantitative assessment of intratumoral heterogeneity using dynamic contrast-enhanced and diffusion-weighted magnetic resonance imaging. Eur Radiol 2022;32:822-33. [Crossref] [PubMed]
16. Yang Z, Chen X, Zhang T, et al. Quantitative Multiparametric MRI as an Imaging Biomarker for the Prediction of Breast Cancer Receptor Status and Molecular Subtypes. Front Oncol 2021;11:628824. [Crossref] [PubMed]
17. Arian A, Seyed-Kolbadi FZ, Yaghoobpoor S, et al. Diagnostic accuracy of intravoxel incoherent motion (IVIM) and dynamic contrast-enhanced (DCE) MRI to differentiate benign from malignant breast lesions: A systematic review and meta-analysis. Eur J Radiol 2023;167:111051. [Crossref] [PubMed]
18. Fumagalli C, Barberis M. Breast Cancer Heterogeneity. Diagnostics (Basel) 2021;11:1555. [Crossref] [PubMed]
19. Cho HH, Kim H, Nam SY, et al. Measurement of Perfusion Heterogeneity within Tumor Habitats on Magnetic Resonance Imaging and Its Association with Prognosis in Breast Cancer Patients. Cancers (Basel) 2022;14:1858. [Crossref] [PubMed]
20. Meng L, Zhao X, Guo J, et al. Improved Differential Diagnosis Based on BI-RADS Descriptors and Apparent Diffusion Coefficient for Breast Lesions: A Multiparametric MRI Analysis as Compared to Kaiser Score. Acad Radiol 2023;30:S93-S103. [Crossref] [PubMed]
21. Liu C, Liang C, Liu Z, et al. Intravoxel incoherent motion (IVIM) in evaluation of breast lesions: comparison with conventional DWI. Eur J Radiol 2013;82:e782-9. [Crossref] [PubMed]
22. Cheon H, Kim HJ, Kim TH, et al. Invasive Breast Cancer: Prognostic Value of Peritumoral Edema Identified at Preoperative MR Imaging. Radiology 2018;287:68-75. [Crossref] [PubMed]
23. Park NJ, Jeong JY, Park JY, et al. Peritumoral edema in breast cancer at preoperative MRI: an interpretative study with histopathological review toward understanding tumor microenvironment. Sci Rep 2021;11:12992. [Crossref] [PubMed]