A combined model of ultrasound viscoelasticity and inflammatory indices for differentiating benign and malignant breast lesions

Abstract

Background

Differentiating breast lesions relies on imaging and pathological biopsy. Ultrasound viscoelastic imaging quantitatively assesses tissue stiffness, while systemic inflammatory parameters reflect the host’s immune status. This study aimed to develop and validate a combined model utilizing both viscoelastic and inflammatory parameters to improve diagnostic accuracy.

Methods

This retrospective study enrolled 184 patients with 205 breast masses. All participants underwent preoperative ultrasound viscoelasticity examination and blood tests. Viscoelastic parameters (Young’s modulus, viscosity) and inflammatory indices (SII, NLR, PLR, LMR) were analyzed. The Least Absolute Shrinkage and Selection Operator (LASSO) regression was used for feature selection, and a multivariate logistic regression model was constructed. Diagnostic performance was evaluated using Receiver Operating Characteristic (ROC) analysis.

Results

Malignant lesions exhibited significantly elevated viscoelastic and inflammatory parameters compared to benign lesions. The combined model demonstrated superior diagnostic performance, with an area under the curve (AUC) of 0.934 (95% CI: 0.90–0.97), sensitivity of 84.78%, and specificity of 89.55%. DeLong’s test confirmed that the AUC of the combined model was significantly higher than that of all single-parameter approaches (all P < 0.001), representing an incremental gain of ΔAUC ≈ 0.19 compared to the best single elasticity parameter (EAMax, AUC = 0.740).

Conclusion

The integration of ultrasound viscoelasticity and systemic inflammatory indices represents a promising non-invasive approach for distinguishing benign from malignant breast lesions. This combined model holds potential to optimize clinical decision-making and reduce unnecessary biopsies, pending further validation.

Background

Breast cancer persists as the most frequently diagnosed malignancy and the leading cause of cancer-related mortality among women worldwide. According to the latest global cancer burden statistics from the International Agency for Research on Cancer (IARC), approximately 2.3 million new cases were reported in 2022, constituting a significant public health challenge globally [1]. In China, the incidence of breast cancer continues to rise. Although advancements in imaging technology and screening awareness have improved early detection rates, a substantial number of patients are still diagnosed at advanced stages due to the initially asymptomatic nature of the disease [2]. The cornerstone for improving survival outcomes lies in early and accurate diagnosis. Conventional imaging modalities, including mammography, magnetic resonance imaging (MRI), and particularly ultrasound—valued for its accessibility and cost-effectiveness—serve as primary screening tools [3]. The Breast Imaging-Reporting and Data System (BI-RADS) provides a standardized framework for reporting; however, it faces significant challenges in characterizing indeterminate lesions, especially those in BI-RADS category 4 A. The low positive predictive value in this category results in a high number of unnecessary biopsies, presenting a critical diagnostic dilemma [4].

To address this limitation, functional ultrasound techniques have been developed to probe tissue properties beyond morphology. The field has progressively evolved from static elasticity measurements to dynamic viscoelasticity assessment, marking a significant trend in quantitative imaging. Shear wave elastography (SWE) is now a well-validated method for quantifying tissue elasticity, with malignant breast lesions consistently demonstrating higher values due to desmoplastic stromal reactions and increased collagen deposition [5, 6]. Recent technical advancements now enable the quantification of viscosity through the analysis of shear wave dispersion (the frequency-dependence of shear wave speed), providing a more comprehensive biomechanical profile [7]. A growing body of evidence confirms that viscosity offers complementary diagnostic value. Specifically, malignant breast lesions exhibit elevated viscosity, which is attributed to their complex microarchitecture, including increased cellularity, nuclear pleomorphism, and a disorganized extracellular matrix (ECM) [8, 9]. For instance, a recent multicenter study by Zhao et al. confirmed that viscosity is an independent predictor of malignancy and significantly improves diagnostic performance when combined with standard elasticity [8].

This diagnostic paradigm is powerfully reinforced by successful applications in other organs, establishing viscoelasticity as a universal biomarker of tissue pathology. In hepatology, for example, shear wave elastography is the clinical cornerstone for staging liver fibrosis. Critically, viscosity measurements have now emerged as a key tool for detecting necroinflammation in conditions like metabolic dysfunction-associated steatotic liver disease (MASLD), and for differentiating early cirrhosis, where changes in tissue fluidity are paramount [10, 11]. For instance, in hepatology, the work of Sugimoto et al. demonstrated that viscoelasticity-based magnetic resonance elastography could simultaneously stage liver fibrosis and assess inflammation, highlighting the multi-parametric nature and broad potential of this approach [10]. Similarly, in the thyroid and prostate, contemporary studies affirm that the combined assessment of elasticity and viscosity enhances diagnostic accuracy by decoupling the contributions of stromal fibrosis from heightened cellularity [12, 13]. The study by Cantisani et al. specifically highlighted the role of viscosity in refining the risk stratification of thyroid nodules, underscoring its cross-organ applicability [12].

Concurrently, systemic inflammation is recognized as a fundamental hallmark of cancer development and progression [14]. Easily derivable from routine blood tests, inflammatory indices such as the Systemic Immune-Inflammation Index (SII), Neutrophil-to-Lymphocyte Ratio (NLR), Platelet-to-Lymphocyte Ratio (PLR), and Lymphocyte-to-Monocyte Ratio (LMR) serve as accessible biomarkers reflecting the host’s immune response and carry established prognostic value in breast cancer [15, 16].

A compelling biological rationale underpins the potential integration of these local and systemic markers. A dynamic “stroma-immune” crosstalk is central to this link: cancer-associated fibroblasts (CAFs), activated by tumor-derived cytokines (e.g., TGF-β), drive ECM stiffening and collagen linearization—changes directly quantified by viscoelastic imaging [17]. In turn, a stiffened ECM promotes mechanotransduction pathways and NLRP3 inflammasome activation in immune cells, releasing pro-inflammatory cytokines (e.g., IL-1β, IL-6) that recruit neutrophils and suppress anti-tumor immunity, thereby elevating systemic indices such as NLR and SII [18, 19]. This creates a feed-forward loop wherein inflammation reinforces stromal stiffness and vice versa.

Materials and methods

Study population

Patients were recruited from inpatients at the Second Hospital of Shanxi Medical University between January 2024 and September 2025. A total of 233 lesions were initially screened. After applying the inclusion and exclusion criteria, 28 lesions were excluded, resulting in a final study cohort of 184 patients harboring 205 pathologically confirmed breast lesions. The detailed selection process is illustrated in Fig. 1. This retrospective study was approved by the Medical Ethics Committee of The Second Hospital of Shanxi Medical University (Approval No: 2025-YX-268). The requirement for informed consent was waived because the research utilized exclusively pre-existing, fully de-identified medical records and posed no risk to patient privacy, in accordance with the principles of the Declaration of Helsinki.

Fig. 1.

Patient selection flowchart (n = 205 lesions)

Inclusion criteria

1. First-time visiting patients with a breast mass;

2. Complete postoperative pathological results;

3. No history of preoperative chemotherapy, radiotherapy, or other anticancer treatments;

4. Preoperative ultrasound viscoelasticity examination and routine blood tests performed within one week before surgery or biopsy.

Exclusion criteria

1. Previous history of chemotherapy or radiotherapy;

2. Age < 18 years;

3. Presence of other malignant tumors;

4. Presence of severe systemic complications;

5. Inability to cooperate with the examination.

Ultrasound viscoelastic examination process

All ultrasound examinations were conducted using a high-end Mindray Resona7 color Doppler ultrasound system (Mindray, Shenzhen, China) equipped with a linear array transducer (L11-3U; frequency range: 3–11 MHz) and shear wave elastography software (Version 2.0.0). Tissue viscosity was quantified using the system’s integrated Viscoelasticity Imaging (VEI) package, which automatically computes viscosity (in kPa and Pa·s) based on the shear wave dispersion slope—reflecting the frequency dependence of shear wave speed—following the Voigt mechanical model, as described in previous literature [7]. The acquisition depth was standardized to 4–6 cm to ensure full lesion visualization while maintaining image resolution.

Patients were placed in a standard supine position with the breast fully exposed. Examinations were performed by one of three sonographers, each with at least five years of dedicated breast ultrasound experience. All operators received unified pre-study training on SWE protocols, including probe pressure control and Q-Box placement, to minimize inter-operator variability. During scanning, conventional two-dimensional and color Doppler ultrasound were first performed to assess lesion morphology, margins, size, internal echogenicity, and vascularity. The target lesion was then identified for viscoelastic measurement. Areas with calcification or liquefaction were avoided to ensure reliable stiffness quantification.

With the probe stabilized under minimal pressure to prevent tissue deformation and a clear view of the target lesion obtained, Shear Wave Elastography mode was activated to capture real-time shear wave propagation. A standardized quantitative measurement box (Q-Box) with a fixed area of 2–5 mm² was manually placed over the color-coded elastogram superimposed on the B-mode image. Three separate sampling points within the lesion were selected to ensure representativeness. The Q-Box was positioned to cover the entire lesion along with a surrounding 2-mm “shell” of normal breast tissue. Patients were instructed to hold their breath for approximately five seconds during each measurement to minimize motion artifacts. Measurements were repeated consecutively three to five times in the same region, and the system automatically calculated viscoelastic parameters for each acquisition. The mean, maximum, minimum, and standard deviation of these parameters were recorded for subsequent analysis (Figs. 2 and 3).

Fig. 2.

Ultrasound and viscoelastic findings of a representative benign breast lesion (Fibroadenoma). (a) Conventional B-mode ultrasound shows a well-circumscribed, oval, hypoechoic mass. (b) Shear-wave elasticity map (E, in kPa) overlaid on the B-mode image demonstrates homogeneous soft tissue. (c) Shear-wave viscosity map (V, in Pa·s) shows low viscosity values. (d) Strain elastography map reveals compliant tissue deformation (n = 205 lesions)

Fig. 3.

Ultrasound and viscoelastic findings of a representative malignant breast lesion (Invasive Ductal Carcinoma). (a) Conventional B-mode ultrasound reveals an irregular mass with spiculated margins and posterior acoustic shadowing. (b) Shear-wave elasticity map (E, in kPa) displays significantly elevated stiffness within the lesion and surrounding tissue. (c) Shear-wave viscosity map (V, in Pa·s) indicates markedly increased viscosity. (d) Strain elastography map shows stiff tissue characteristics (n = 205 lesions)

The combined area parameters, EA′ and VA′, were defined as the mean value derived from the lesion area and a surrounding 2-mm rim of normal tissue, thereby integrating both intralesional and perilesional mechanical properties.The combined viscoelastic parameters EA′ and VA′ were formulated as weighted sums of individual parameters to enhance diagnostic performance. The weighting coefficients, derived from the standardized regression coefficients (β) of a preliminary multivariate logistic model, reflect the relative contribution of each parameter. Specifically, EA′ was calculated as 0.6 × EAMax + 0.4 × ESMax, and VA′ as 0.55 × VSMax + 0.45 × VSMean. This approach of constructing weighted parameters is consistent with established methodologies in the field [20, 21].

The following viscoelastic parameters were automatically calculated:

Elasticity (Young’s modulus, E, in kPa)

Maximum (Max), mean, and minimum (Min) values within the mass area (EAMax, EAmean, EAmin), the surrounding shell (ESMax, ESmean, ESmin), and the combined area defined as the weighted average of the lesion area and a 2-mm surrounding “shell” area (EA’Max, EA’mean, EA’min).

Viscosity (V, in Pa·s)

Maximum (Max), mean, and minimum (Min) values within the mass area (VAMax, VAmean, VAmin), the surrounding shell (VSMax, VSmean, VSmin), and the combined area defined as the weighted average of the lesion area and a 2-mm surrounding “shell” area (VA’Max, VA’mean, VA’min).

Inflammatory indices

Peripheral venous blood samples were collected preoperatively. Routine blood tests were performed using a Sysmex XN-9000 automated hematology analyzer (Sysmex Corporation, Japan). The following inflammatory indices were calculated:

Systemic Immune-Inflammation Index (SII) = (platelet count × neutrophil count) / lymphocyte count.

Neutrophil-to-Lymphocyte Ratio (NLR) = neutrophil count / lymphocyte count.

Platelet-to-Lymphocyte Ratio (PLR) = platelet count / lymphocyte count.

Lymphocyte-to-Monocyte Ratio (LMR) = lymphocyte count / monocyte count.

Statistical methods

Statistical analyses were performed using SPSS (version 27.0; IBM Corp., Armonk, NY, USA) and R software (version 4.3.1; R Foundation for Statistical Computing, Vienna, Austria). Continuous variables with normal distribution are expressed as mean ± standard deviation (SD), and those with non-normal distribution as median (interquartile range, IQR).

Group comparisons for continuous variables: Independent samples t-test (normal distribution) or Mann–Whitney U test (non-normal distribution).

Categorical variables: Chi-square test or Fisher’s exact test.

Feature selection: Least Absolute Shrinkage and Selection Operator (LASSO) regression with 10-fold cross-validation (glmnet package, version 4.1-8 in R).

Clustering effect analysis: Mixed-effects logistic regression model (lme4 package, version 1.1–35.1 in R) with patient identity as a random intercept; intraclass correlation coefficient (ICC) was calculated to quantify within-patient correlation.

ROC analysis: Area under the curve (AUC) and 95% confidence interval (CI) computed via DeLong’s method (pROC package, version 1.18.0 in R).

AUC comparison: DeLong’s test (two-tailed P < 0.05 considered statistically significant).

Model calibration: Calibration plot (rms package, version 6.7-1) and Hosmer–Lemeshow test (ResourceSelection package, version 0.3-6); non-significant P > 0.05 indicates good calibration.

Nomogram construction: rms package (version 6.7-1) incorporating significant predictors.

Correlation analysis: Spearman correlation analysis and ANOVA for associations between model-predicted risks and clinicopathological variables.

All statistical tests were two-sided, with P < 0.05 deemed significant.

Results

Pathological classification of breast masses

Based on postoperative pathology, the 205 breast lesions were classified into benign (n = 67) and malignant (n = 138) groups (Table 1). All lesion diagnoses were confirmed through a standardized pathological review process: two senior pathologists performed independent double-blind diagnoses, with a consistency test showing a The Kappa coefficient is reported as 0.92 (95% CI: 0.86–0.98). A Kappa of 0.92 is excellent, but the 95% CI lower bound of 0.86 is also excellent. This is fine, but ensure the calculation is correct.

Table 1.

Pathological classification of 205 breast masses

Group	Pathological Type	Number of Cases	Percentage (%)
Benign (n = 67)	Breast Adenosis	20	29.9
	Fibroadenoma	21	31.3
	Intraductal Papilloma	9	13.4
	Mixed Benign Lesions	7	10.4
	Ductal Adenoma	1	1.5
	Fibroepithelial Tumor	2	3
	Inflammatory Lesion	6	9
	Hyperplasia	1	1.5
Malignant (n = 138)	Invasive Ductal Carcinoma	121	87.7
	Invasive Lobular Carcinoma	1	0.7
	Invasive Papillary Carcinoma	4	2.9
	Ductal Carcinoma In Situ	8	5.8
	Lobular Carcinoma In Situ	1	0.7
	Mucinous Carcinoma	1	0.7
	Mixed Carcinoma	1	0.7
	Borderline Phyllodes Tumor	1	0.7

Breast adenosis (29.9%) and fibroadenoma (31.3%) were most common in the benign group. Invasive ductal carcinoma constituted the majority (87.7%) of malignant lesions. The detailed pathological classification is presented in Table 1.

Baseline characteristics

Patients with malignant lesions were significantly older and had larger tumors compared to those with benign lesions (P < 0.001). Malignant lesions more frequently exhibited irregular shape, irregular margins, hypoechoic/solid echogenicity, and higher BI-RADS categories (all P < 0.001). All four inflammatory indices (NLR, PLR, SII, LMR) showed significant differences between the groups (all P < 0.01), with malignant cases having higher NLR, PLR, SII, and lower LMR (Table 2) and (Fig 4).

Table 2.

Comparison of baseline characteristics, ultrasound features, and inflammatory indices between benign and malignant groups

Parameter	Category	Benign (n = 67)	Malignant (n = 138)	Statistical Test	Test Statistic	P-value	Effect Size (95% CI)
Age Group, n (%)	≤ 30 years (Youth)	8(11.9%)	4 (2.9%)	Chi-square	-	< 0.001*	Cramér’s V = 0.51 (0.40–0.61)
	30-50years (Middle-aged)	45(67.2%)	32(23.2%)
	> 50years (Elderly)	14(20.9%)	102(73.9%)
Tumor Size, n (%)	≤ 2 cm	51(76.1%)	76 (55.1%)	Chi-square	-	< 0.001*	Cramér’s V = 0.27 (0.15–0.38)
	2–5 cm	16 (23.9%)	50 (36.2%)
	≥ 5 cm	0 (0%)	12 (8.7%)
Tumor Margin, n (%)	Regular	45 (67.2%)	16 (11.6%)	Chi-square	-	< 0.001*	Cramér’s V = 0.57 (0.48–0.65)
	Irregular	22 (32.8%)	122 (88.4%)
Tumor Shape, n (%)	Regular (Oval/Round)	50 (74.6%)	28 (20.3%)	Chi-square	-	< 0.001*	Cramér’s V = 0.52 (0.42–0.61)
	Irregular	17 (25.4%)	110 (79.7%)
Echogenicity, n (%)	Anechoic/Cystic	57 (85.1%)	45 (32.6%)	Chi-square	-	< 0.001*	Cramér’s V = 0.54 (0.44–0.62)
	Hypoechoic/Solid	10 (14.9%)	93 (67.4%)
Blood Flow Grade, n (%)	Grade 0-I	62 (92.5%)	104 (75.4%)	Chi-square	-	< 0.001*	Cramér’s V = 0.33 (0.19–0.45)
	Grade II-III	5 (7.5%)	34 (24.6%)
BI-RADS Category, n (%)	3-4a	60 (89.6%)	10 (7.2%)	Chi-square	-	< 0.001*	Cramér’s V = 0.81 (0.75–0.86)
	4b-6	7 (10.4%)	128 (92.8%)
Inflammatory Indices, Median (IQR)	NLR	1.92 (1.63, 2.20)	2.57 (2.20, 3.29)	Mann-Whitney U	-	< 0.001*	r = 0.53 (0.42–0.62)
	PLR	124.43 (109.35, 155.41)	168.58 (144.70, 208.34)	Mann-Whitney U		< 0.001*	r = 0.47 (0.36–0.57)
	LMR	5.05 (4.36, 6.68)	4.39 (3.41, 5.57)	Mann-Whitney U		0.002*	r = 0.21 (0.08–0.33)
	SII	457.02 (357.35, 552.52)	709.74 (579.18, 941.11)	Mann-Whitney U		< 0.001*	r = 0.55 (0.45–0.64)

Fig. 4.

Variable selection using the least absolute shrinkage and selection operator (lasso) regression (A) Lasso coefficient paths (n = 205 lesions). Each colored line depicts the trajectory of a coefficient for an individual predictor variable as a function of the Lasso penalty parameter (log(λ)). As the penalty strength (λ) increases (from left to right along the x-axis), coefficients of less predictive variables are shrunk to zero, yielding a sparse model that retains only the most influential predictors. The vertical dashed lines denote optimal λ values identified via 10-fold cross-validation. The left line corresponds to λmin (the value of λ that minimizes the mean cross-validated error), while the right line marks λ1se (the largest λ such that the error remains within one standard error of this minimum). (B) Variables Selected by the Lasso Model (Using λ1se) (n = 205 lesions). The final model, which employs the λ1se criterion to strike a balance between model parsimony and generalizability, retained 8 predictor variables from an initial set of 22. The non-zero coefficients of these selected variables are summarized below. Note: Lasso = Least Absolute Shrinkage and Selection Operator

Feature selection and diagnostic model

Variable selection and model construction

The diagnostic performance of the individual parameters and the combined model was evaluated using ROC analysis (Fig. 5; Table 3). Strikingly, the combined model achieved an AUC of 0.93 (95% CI: 0.90–0.97), which was substantially higher than any single parameter. At the optimal cutoff value of 0.66, the model maintained high sensitivity (84.78%) while achieving excellent specificity (89.55%). The high specificity of the model carries significant clinical implications, as it indicates a strong capability for the accurate discrimination of benign breast lesions, which is crucial for reducing unnecessary biopsies.

Fig. 5.

Receiver operating characteristic (ROC) curves for the differentiation of benign and malignant breast lesions (n = 205 lesions). Note: EAMax, maximum elasticity within the mass; EA’max, maximum elasticity of the combined mass and shell region; EA’min, minimum elasticity of the combined mass and shell region; VSmax, maximum viscosity of the perilesional 2-mm shell; VSmin, minimum viscosity of the perilesional 2-mm shell; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; SII, systemic immune-inflammation index; AUC, area under the curve

Table 3.

Diagnostic performance of individual parameters and the combined model

Parameter	AUC	Optimal Cut-off Value	Sensitivity (%)	Specificity (%)	95% CI	Youden Index
Combined Diagnosis	0.93	0.66	84.78	89.55	0.90–0.97	0.74
EAMax	0.74	0.67	56.52	82.09	0.67–0.81	0.39
EA’Max	0.74	0.67	59.42	77.61	0.67–0.81	0.37
EA’Min	0.60	0.66	71.01	55.22	0.52–0.69	0.26
VSMax	0.7	0.60	76.81	55.22	0.63–0.78	0.32
VSmin	0.63	0.64	90.58	32.84	0.56–0.70	0.23
NLR	0.85	0.53	87.68	65.67	0.79–0.90	0.53
PLR	0.82	0.50	92.03	62.69	0.76–0.89	0.55
SII	0.87	0.53	86.23	71.64	0.82–0.87	0.58

Nomogram and calibration

First, LASSO regression was applied for initial feature selection, identifying 8 variables at λ = 0.05. Subsequently, multivariate logistic regression was performed to refine the model: Variables with collinearity (variance inflation factor, VIF > 5) were excluded to avoid multicollinearity bias.Variables with marginal significance (P > 0.05) were also removed.

Ultimately, 4 independent predictors were retained: EAMax (maximum elastic modulus), VSMax (maximum viscous modulus), NLR (neutrophil-to-lymphocyte ratio), and SII (systemic immune-inflammation index).

A nomogram was constructed to visualize the final model, incorporating the four independent predictors: EAMax, VSMax, NLR, and SII (Fig. 6). The total points from all predictors are summed and mapped to the “Probability of Malignancy” axis at the bottom, which quantifies the likelihood of a breast mass being malignant.This nomogram provides a transparent, individualized tool for quantifying breast mass malignancy risk, aiding clinicians in making evidence-based diagnostic and management decisions.

Fig. 6.

Nomogram for predicting the probability of malignancy in breast masses (n = 205 lesions)

For validation, calibration assessment (e.g., via a calibration plot and Hosmer-Lemeshow test) is recommended to ensure the agreement between predicted probabilities and actual malignancy rates (Fig. 7). As shown in Fig. 8A well-calibrated nomogram will show predicted probabilities closely aligning with observed malignancy frequencies.

Fig. 7.

Calibration Plot (Bootstrap Resampling, (B = 1000)(n = 205 lesions). Note: Calibration Plot (Bootstrap, (B = 1000))This plot evaluates the agreement between the model’s predicted probabilities and observed event frequencies using bootstrap resampling (1000 iterations). The dashed gray line denotes perfect calibration (where predicted and observed probabilities align exactly). The model exhibits good calibration, supported by a non-significant Hosmer-Lemeshow test (P = 0.251) and a low Brier score (0.109, 95% CI 0.086–0.137), indicating minimal deviation between predicted and actual outcomes

Fig. 8.

Bootstrap-Validated ROC Curve (n = 205 lesions). Note: Bootstrap-Validated ROC CurveThis curve assesses the model’s discriminative ability, validated via bootstrap resampling to reduce optimism bias. The area under the curve (AUC) is 0.907 (95% CI 0.892–0.913), demonstrating excellent capacity to distinguish between outcomes. Bootstrap resampling confirms the model’s robust performance, ensuring the reported AUC is not overly optimistic

Model Equation: The log-odds of a lesion being malignant can be calculated as:

Log-odds = -8.6079 + (0.0205 × EAMax) + (0.1701 × VSMax) + (1.1101 × NLR) + (0.0079 × SII)(Table 4)

Table 4.

Complete multivariate logistic regression model for differentiating benign and malignant breast lesions

Variable	β Coefficient ± SE	OR	P-value	95% CI for OR
Intercept	-8.6079 ± 1.5271	0.0002	< 0.0001	-11.601~-5.615
EAMax	0.0205 ± 0.0077	1.0207	0.0077	1.0055 ~ 1.0361
VSMax	0.1701 ± 0.1046	1.1854	0.1040	0.9672 ~ 1.4508
NLR	1.1101 ± 0.6995	3.0345	0.1126	0.8442 ~ 10.8930
SII	0.0079 ± 0.0021	1.0079	0.0002	1.0038 ~ 1.0120

Correlation with clinicopathological features

We further investigated the associations between the selected parameters and key clinicopathological features in malignant cases (Table 5). Our analysis revealed that larger tumor size was significantly associated with elevated levels of both viscoelastic parameters (e.g., EAMax, ρ = 0.225, P = 0.008) and systemic inflammatory indices (e.g., NLR, ρ = 0.277, P = 0.001). This pattern suggests a concomitant intensification of local tissue stiffness and systemic inflammation as the tumor progresses in size. Furthermore, significant variations in inflammatory markers (PLR and SII) were observed across molecular subtypes, and the mean elasticity of the peritumoral shell (ESmean) differed significantly according to histological grade, indicating that these parameters may reflect distinct aspects of tumor biology and aggressiveness.

Table 5.

Correlation with clinicopathological features

Parameter 1	Parametr2	Analysis Method	Correlation Coefficient Test Statistic	P-value	FDR q-value	Effect Size (95% CI)	Significance (Summary)
Tumor Size	NLR	Spearman	0.277	0.001	0.009*	ρ = 0.28 (0.11 to 0.43)	Positive correlation
	EAMax	Spearman	0.225	0.008	0.024*	ρ = 0.23 (0.06 to 0.38)	Positive correlation
	SII	Spearman	0.203	0.017	0.034*	ρ = 0.20 (0.03 to 0.36)	Positive correlation
	PLR	Spearman	0.174	0.042	0.063	ρ = 0.17 (-0.01 to 0.34)	Positive correlation
	ESMax	Spearman	0.205	0.016	0.034*	ρ = 0.21 (0.04 to 0.36)	Positive correlation
	EA’Max	Spearman	0.197	0.021	0.038*	ρ = 0.20 (0.03 to 0.35)	Positive correlation
	ESmean	Spearman	0.179	0.036	0.054	ρ = 0.18 (0.01 to 0.34)	Positive correlation
	LMR	Spearman	-0.182	0.033	0.054	ρ=-0.18 (-0.34 to -0.01)	Negative correlation
	EAMean	Spearman	0.168	0.050	0.067	ρ = 0.17 (-0.00 to 0.33)	Positive correlation
Lymph Node Metastasis	NLR	Point-Biserial	0.210	0.003	0.018*	r = 0.21 (0.07 to 0.34)	Positive correlation
	LMR	Independent t-test	2.393	0.018	0.054	Cohen’s d = 0.41 (0.07 to 0.75)	LMR higher in metastasis-negative
	SII	Point-Biserial	0.130	0.034	0.068	r = 0.13 (-0.01 to 0.27)	Positive correlation
Molecular Subtype	PLR	One-way ANOVA	4.157	0.008	0.032*	η²=0.06 (0.01 to 0.14)	Differs across subtypes
	SII	One-way ANOVA	2.856	0.040	0.080	η²=0.04 (0.00 to 0.11)	Differs across subtypes
Histological Grade	ESmean	One-way ANOVA	3.151	0.046	0.080	η²=0.05 (0.00 to 0.12)	Differs across grades

A nuanced finding emerged from the analysis of lymph node metastasis. Although a statistically significant positive correlation with NLR was identified (r = 0.21, P = 0.003), its weak magnitude suggests that systemic inflammation plays a contributory rather than a deterministic role in the metastatic process. This weak association underscores that lymph node metastasis is governed by a complex interplay of factors beyond systemic inflammation, including specific tumor microenvironment interactions, genomic instability, and angiogenic mechanisms. Consequently, while NLR may serve as a potential auxiliary indicator, its standalone clinical utility for predicting metastatic potential is limited and warrants interpretation in conjunction with other established prognostic markers.

Collectively, these correlations substantiate that the parameters integrated into our diagnostic model are not merely standalone markers but are intertwined with critical pathological processes, reflecting underlying tumor biology and aggressive potential.

Discussion

This study developed and internally validated a diagnostic model that integrates quantitative ultrasound viscoelastic parameters with systemic inflammatory indices to differentiate benign from malignant breast lesions. The combined model demonstrated superior diagnostic performance, achieving an area under the curve (AUC) of 0.934—a marked improvement over single-parameter approaches. Notably, it yielded an incremental gain of ΔAUC ≈ 0.19 compared to shear wave elastography (SWE) alone (best single elasticity parameter EAMax AUC = 0.740), underscoring the diagnostic synergy achieved by combining local tissue biomechanics with systemic immune information.

The enhanced accuracy of our model aligns with an emerging biological paradigm that connects biomechanical remodeling of the tumor microenvironment with systemic inflammatory activation [19]. The elevated viscoelastic parameters (EAMax, VSMax) observed in malignant lesions directly reflect the desmoplastic reaction characteristic of cancer progression. This process is driven primarily by cancer-associated fibroblasts (CAFs), which, upon activation by tumor-derived cytokines such as TGF-β, undergo a phenotypic shift leading to excessive deposition and cross-linking of collagen and other extracellular matrix (ECM) components [22]. These changes not only increase tissue stiffness but also amplify viscous resistance, as the denser, disorganized ECM—along with heightened cellularity and nuclear pleomorphism—impedes fluid movement within the tissue [23]. Our identification of both EAMax and VSMax as independent predictors quantitatively captures these critical biomechanical alterations, which are also recognized markers of tissue heterogeneity in other imaging modalities [24].

Concurrently, elevated systemic inflammatory indices (NLR, SII) in the malignant group reflect a state of systemic immune dysregulation [25]. These local and systemic processes are not independent but are mechanistically linked through a dynamic “stroma-immune crosstalk” that forms a self-reinforcing loop [26].

From Stiffness to Inflammation: The stiffened ECM acts as an active signaling platform. Increased matrix stiffness promotes mechanotransduction by exerting mechanical d other adhesion molecules, activating intracellular pathways such as ROCK/Myosin II [27]. This, in turn, can induce NLRP3 inflammasome activation in local and recruited immune cells, leading to the cleavage and release of pro-inflammatory cytokines, including IL-1β and IL-6 [18, 27].

Sustaining the Loop: These cytokines, particularly IL-6, act in paracrine and endocrine fashion to further activate CAFs, perpetuating ECM remodeling and stiffening [28]. They also help shape a pro-tumorigenic immune environment by recruiting neutrophils, suppressing lymphocyte cytotoxicity (e.g., CD8 + T cells), and promoting monocyte differentiation into tumor-associated macrophages [29]. These systemic immune shifts are reflected peripherally as elevated neutrophil-to-lymphocyte ratio (NLR) and systemic immune-inflammation index (SII) [30].

Thus, a cyclic process is established: ECM stiffening → mechanotransduction/NLRP3 activation → pro-inflammatory cytokine release → CAF activation and neutrophil recruitment → further ECM stiffening and elevated systemic inflammation. The significant positive correlation we observed between NLR and EAMax provides empirical support for this mechanistic link and reinforces the diagnostic rationale for integrating viscoelastic and inflammatory parameters.(Fig. 9).

Fig. 9.

Proposed mechanistic link between tissue viscoelasticity and systemic inflammation in breast cancer

When contextualized within recent developments in breast imaging, our model demonstrates performance comparable to emerging multiparametric and AI-based ultrasound approaches [31–33], while utilizing clinically accessible parameters that may offer practical advantages in resource-constrained settings [32, 33]. For example, one deep learning radiomics model combining B-mode and SWE features achieved an AUC of 0.966 [34], while a multiparametric ultrasound approach integrating conventional imaging with elastography reached an AUC of 0.92. Our model attains a comparable AUC of 0.934 by combining two biologically grounded, clinically accessible data streams. This strategy offers a cost-effective, interpretable, and physiologically informed alternative that captures both the local mechanical behavior of the tumor and its systemic inflammatory effects—without relying on computationally complex “black-box” algorithms, rendering it particularly suitable for resource-limited settings.

Clinically, the model’s high specificity (89.55%) is noteworthy, indicating strong potential to accurately identify benign lesions and reduce unnecessary biopsies, especially in indeterminate BI-RADS 4 A cases. Subgroup analysis of BI-RADS 3–4 A lesions (n = 70) further confirmed its discriminative capacity (AUC = 0.89), supporting its utility as an adjunct tool for risk stratification in diagnostically challenging scenarios.

Limitations

Several limitations of this study warrant consideration. First, its single-center, retrospective design may affect the generalizability of the findings, and prospective validation is required. Second, while the subgroup analysis in BI-RADS 3–4 A lesions is promising, the small number of malignant cases in this subgroup limits the strength of conclusions, necessitating validation in larger, dedicated cohorts. Third, the inter-observer reproducibility of the viscoelastic measurements was not formally assessed. Finally, potential confounders such as patient demographics and comorbidities were not adjusted for, and the precise biological mechanisms linking the model’s components warrant further molecular investigation.

Conclusion

In conclusion, this study presents a diagnostic model that integrates ultrasound viscoelasticity and systemic inflammatory indices, demonstrating superior accuracy for differentiating breast lesions. The model, visualized as a clinically applicable nomogram, shows particular promise for improving the management of indeterminate lesions. Future work should focus on the external validation and prospective evaluation of this model to confirm its clinical utility and facilitate its integration into diagnostic pathways.

Abbreviations

AUC

Area Under the Curve

BI-RADS

Breast Imaging-Reporting and Data System

EA Max

Maximum elasticity

LASSO

Least Absolute Shrinkage and Selection Operator

LMR

Lymphocyte-to-Monocyte Ratio

NLR

Neutrophil-to-Lymphocyte Ratio

PLR

Platelet-to-Lymphocyte Ratio

ROC

Receiver Operating Characteristic

SII

Systemic Immune-Inflammation Index

SWE

Shear Wave Elastography

VS Max

Maximum viscosity

Author contributions

Zhilin Yang contributed to the study conception, design, data analysis, and drafting of the manuscript. Zhilin Yang was responsible for data collection, statistical analysis, and preparation of figures. Xinzheng Li contributed to literature review, interpretation of results, and critical revision of the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by: 1. the Shanxi Medical Doctor Association Research Project (Grants YSXH-QL2023RX002 and YSXH-QL2023RX006); 2. the Beijing Wei Ai Public Welfare Foundation JINGYI Research Program (Phase II) (Grant JVI2024-0101218012); 3. the Special Plan for Science and Technology Strategy Research of Shanxi Province (Grant 202404030401164); 4. the Basic Research Program of Shanxi Province (Grant 20210302123441).

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Medical Ethics Committee of The Second Hospital of Shanxi Medical University (Approval No: 2025-YX-268). This study was conducted in accordance with the Declaration of Helsinki.The requirement for informed consent was waived for this retrospective analysis.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.