Clinical Value of Contrast-Enhanced Ultrasound and Ultrasound Elastography in the Pathological Classification and Treatment Response Prediction of Cervical Tuberculous Lymphadenitis

Introduction

Tuberculosis remains a major global health burden, with approximately 10.6 million new cases and 1.3 million deaths reported in 2022 (WHO). Although pulmonary tuberculosis is the most common form, extrapulmonary tuberculosis accounts for approximately 15–20% of all tuberculosis cases and poses significant diagnostic challenges due to its nonspecific clinical presentations.¹ Cervical tuberculous lymphadenitis (CTL), the most common form of extrapulmonary tuberculosis, exhibits marked clinical and pathological heterogeneity. At present, diagnosis and classification mainly rely on invasive biopsy, which limits the ability to dynamically assess internal pathological changes such as granuloma formation and necrosis. These pathological subtypes are closely associated with treatment response.² Therefore, achieving noninvasive and accurate pre-treatment classification and therapeutic response prediction is of critical clinical importance.

Ultrasonography, owing to its noninvasiveness, real-time capability, and good reproducibility, has become the first-line imaging modality for superficial lymph node diseases.³ Conventional grayscale ultrasonography and color Doppler ultrasonography primarily provide morphological and vascular distribution information; however, they have limitations in distinguishing specific internal pathological changes, such as caseous necrosis, and in quantifying tissue mechanical properties. In recent years, advances in functional ultrasound imaging techniques have provided new approaches to overcome these limitations. Contrast-enhanced ultrasound (CEUS), performed by intravenous injection of microbubble contrast agents, enables real-time dynamic visualization of microvascular perfusion patterns within lymph nodes and clearly delineates non-perfusion areas, which may indicate necrosis or ischemia.⁴ Ultrasound elastography, in contrast, evaluates tissue deformation under external compression or acoustic radiation force and quantitatively reflects tissue stiffness; necrotic or fibrotic regions typically exhibit increased stiffness.⁵ Theoretically, different pathological subtypes of tuberculous lymphadenitis should demonstrate distinct microcirculatory perfusion characteristics and tissue elasticity features. These differences may constitute the biological basis for imaging-based classification and may also be potentially associated with therapeutic response.⁶ Existing studies have largely focused on differentiating benign from malignant lymph nodes or on overall analyses of tuberculous lymphadenitis, with limited systematic evaluation of imaging characteristics among different pathological subtypes. Furthermore, studies investigating the prediction of treatment response based on imaging parameters remain scarce.⁷ Noninvasive identification of pathological subtypes before treatment matters because granulomatous lesions typically respond well to anti-tuberculosis therapy, whereas caseous necrotic lesions are associated with poorer response and higher relapse risk. The combined application of contrast-enhanced ultrasound and ultrasound elastography (UE) allows simultaneous acquisition of vascular and mechanical information, potentially providing a more comprehensive assessment.

The research questions guiding this study were: (1) Can CEUS and UE parameters differentiate pathological subtypes of CTL? (2) Can pre-treatment imaging parameters predict treatment response at 6 months? We hypothesized that Type II (caseous necrosis) would show higher non-perfusion area ratios and strain ratios, and that these parameters would be associated with poorer treatment response.

Accordingly, this retrospective study systematically evaluated the imaging characteristics of contrast-enhanced ultrasound and ultrasound elastography in differentiating pathological subtypes of CTL. In addition, the associations between imaging parameters and lymph node volume changes at 6 months after treatment were explored. The aim was to provide imaging evidence for noninvasive pathological classification and early prediction of treatment response in CTL, thereby facilitating individualized and precision-based clinical management.

Participants and Methods

Study Population

This study was designed as a retrospective cohort study. A total of 120 patients pathologically diagnosed with cervical tuberculous lymphadenitis between January 2021 and October 2025 were consecutively enrolled. All patients had complete CEUS and UE imaging data.

Inclusion criteria were as follows: (1) Pathologically confirmed CTL; (2) Complete CEUS and UE imaging data obtained during the same period; (3) Receipt of standardized anti-tuberculosis therapy with available ultrasound follow-up data before treatment and at 6 months after treatment; (4) Complete clinical records.

Exclusion criteria included: (1) Concomitant cervical malignant tumors or active infections; (2) Previous surgical intervention or local treatment for CTL; (3) Poor ultrasound image quality affecting imaging feature evaluation; (4) Loss to follow-up or poor treatment adherence precluding efficacy assessment.

All histopathological slides were independently re-evaluated using a double-blind approach by two senior pathologists in our institution, according to widely accepted pathological classification criteria for tuberculous lymphadenitis.⁸ The lesions were classified into three subtypes: Type I (granulomatous hyperplasia), Type II (caseous necrosis), and Type III (mixed type). In cases of disagreement, a third senior pathologist adjudicated until consensus was reached.

Based on ultrasound-measured lymph node volume changes at 6 months after anti-tuberculosis therapy, patients were categorized into a treatment-sensitive group (volume reduction rate ≥ 50%) and a treatment-insensitive group (volume reduction rate < 50% or volume increase).⁹ The specific numbers in each group are presented in the Results section.

At the study design stage, to detect a moderate effect size (Cohen’s f = 0.25) difference in key imaging features among the three pathological subtypes (α = 0.05, power = 80%), sample size estimation was performed using analysis of variance (ANOVA). The calculated minimum total sample size required was 99 patients. A total of 120 patients were ultimately included in this study, meeting and exceeding the estimated requirement.

Ultrasound Examination

Patients were examined in the supine position with full exposure of the neck. All cervical lymph nodes were routinely scanned. The target lymph node was defined as the largest and most morphologically representative lymph node confirmed by needle biopsy or surgical pathology and was selected for analysis. All examinations were performed using a color Doppler ultrasound diagnostic system equipped with CEUS and UE functions (Canon Aplio i900 TUA-A500, linear array probe, frequency 5–12 MHz). Before CEUS and UE, conventional ultrasound assessment was performed, and the basic characteristics of the target lymph node (pathologically confirmed) were recorded.

For CEUS, the contrast agent used was sulfur hexafluoride microbubbles for injection (SonoVue, Bracco Suisse SA). A bolus of 2.4mL microbubble suspension (reconstituted with 5 mL of 0.9% sodium chloride solution before use), followed by a 5 mL saline flush. From the moment of injection, dynamic images of the target lymph node region were continuously acquired for ≥ 180 seconds. Enhancement patterns (eg, homogeneous enhancement, annular enhancement) were observed and recorded. At the peak enhancement phase, the non-perfusion area ratio was calculated as: Non-perfusion area ratio = (non-perfusion area / maximum cross-sectional area of the lymph node) × 100%.

For UE, the system was switched to real-time tissue elastography mode. Patients were instructed to breathe calmly and avoid swallowing. The sampling frame covered the target lymph node and part of the surrounding normal tissue. The probe was held perpendicular to the skin with stable and gentle pressure (monitored using the system’s pressure indicator). Measurements were repeated at least three times, and the most stable and reproducible images were selected for analysis. A semi-quantitative five-point scoring system was applied for elasticity grading: Score 1 (Entire lesion deformable [uniformly green]); Score 2 (Mostly deformable [green > blue]); Score 3: (Approximately equal deformable and non-deformable areas [green ≈ blue]); Score 4 (Mostly non-deformable [blue > green]); Score 5 (Entire lesion non-deformable [uniformly blue]). The strain ratio between the lesion and adjacent normal tissue at the same depth (eg, muscle) was also measured.

All CEUS and UE dynamic images were stored in Digital Imaging and Communications in Medicine format within the Picture Archiving and Communication System. To ensure consistency, all image analyses were independently performed by two physicians with more than 5 years of experience in superficial lymph node ultrasonography, blinded to pathological subtype and treatment outcome. Discrepancies were resolved by discussion or adjudicated by a third senior physician.

Data Collection

Clinical, pathological, and imaging data were retrieved from the hospital electronic medical record system and the Picture Archiving and Communication System. Two uniformly trained investigators independently extracted the data. Cross-verification was conducted after data collection, and discrepancies were resolved by consensus or adjudicated by a third senior investigator.

Collected variables included: (1) Baseline characteristics: age, sex, disease duration (time from first symptom onset to ultrasound examination, in months), previous tuberculosis history, diabetes mellitus, and smoking history. (2) Pathological subtype: Type I (granulomatous hyperplasia), Type II (caseous necrosis), and Type III (mixed type). (3) Treatment information: standardized anti-tuberculosis regimen, treatment duration, and treatment initiation time. (4) Imaging features: CEUS enhancement pattern (annular or non-annular), non-perfusion area ratio, UE elasticity score (≥ 4 vs < 4), and strain ratio.

All imaging features were independently analyzed by two physicians with more than 5 years of experience in thyroid and cervical lymph node ultrasonography, blinded to pathological classification and treatment outcome. In case of disagreement, consensus was achieved through discussion or adjudicated by a third senior physician.

Treatment Response Assessment

The primary endpoint was defined as the change in target lymph node volume at 6 months after anti-tuberculosis therapy. Lymph node volumes before treatment and at 6 months were obtained from standardized ultrasound follow-up reports. Volume was calculated using the ellipsoid formula based on three orthogonal diameters measured by ultrasound (long diameter L, short diameter S, and thickness T): V = (π/6) × L × S × T. The volume reduction rate (VRR) was calculated as: VRR (%) = [(V_pre − V_post) / V_pre] × 100%, where V_pre represents pre-treatment volume and V_post represents volume at 6 months after treatment.⁹ Patients were classified into two groups according to the predefined threshold.

Diagnostic Criteria and Performance Analysis

To evaluate the diagnostic value of CEUS and UE for differentiating pathological subtypes, pathological diagnosis was used as the reference standard. Diagnostic criteria were defined as follows:⁸ (1) For Type II (caseous necrosis): CEUS positive: non-perfusion area ratio ≥ 20%; UE positive: strain ratio ≥ 3.0 or elasticity score ≥ 4. (2) For Type I (granulomatous hyperplasia): CEUS positive: non-perfusion area ratio < 10% and non-annular enhancement (enhancement pattern = 0); UE positive: strain ratio < 2.0 and elasticity score < 4. (3) For combined diagnosis: When diagnosing Type II, positivity in either modality was considered combined positive; When diagnosing Type I, both modalities needed to be positive to be considered combined positive.

Based on these criteria, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy were calculated for each modality and their combination. Due to the heterogeneous pathological composition of Type III (mixed type), no independent positive imaging diagnostic criteria were established.

Statistical Analysis

All statistical analyses were performed using SPSS (version 26.0, IBM Corp.) and R software. All tests were two-sided, and P < 0.05 was considered statistically significant. Normally distributed data were expressed as mean ± standard deviation and compared using the independent samples t-test or one-way ANOVA. Non-normally distributed data were expressed as median (interquartile range) [M (Q1, Q3)] and compared using the Mann–Whitney U-test or Kruskal–Wallis H-test. Categorical variables were expressed as number (percentage) [n (%)] and compared using the chi-square test or Fisher’s exact test. All potential influencing factors were analyzed against treatment sensitivity in univariate analysis. In multivariate analysis, variables with P < 0.05 in univariate analysis were entered into a multivariate logistic regression model (backward elimination method) to identify independent risk factors for treatment insensitivity. Odds ratios (OR) and 95% confidence intervals (CI) were calculated. To assess potential multicollinearity among the variables included in the final multivariate model, the variance inflation factor was calculated for each variable. A VIF value < 5 was considered indicative of no significant multicollinearity, while VIF > 10 suggested serious collinearity requiring variable selection or model adjustment.

Based on identified independent risk factors, a combined prediction model was constructed in the training set. To compare the diagnostic performance between single-modality and combined models, the DeLong test was used to evaluate whether differences in the area under the receiver operating characteristic curve were statistically significant.

Results

Comparison of Baseline and Imaging Characteristics Among Different Pathological Subtypes

A total of 120 patients with CTL were included and classified according to pathological subtype into Type I (granulomatous hyperplasia, n = 40), Type II (caseous necrosis, n = 40), and Type III (mixed type, n = 40). Regarding baseline characteristics, there were no significant differences among the three groups in age, sex, disease duration, history of tuberculosis, diabetes mellitus, or smoking history (all P > 0.05). Comparison of imaging features revealed significant differences among the subtypes. Type II and Type III lesions more frequently demonstrated annular enhancement on CEUS (52.5% and 45.0%, respectively; P = 0.008). The non-perfusion area ratio was significantly higher in Type II and Type III than in Type I (P < 0.001), with the highest median value observed in Type II (33.95%). For UE parameters, the proportion of lesions with elasticity score ≥ 4 was highest in Type II (57.5%, P = 0.003). The strain ratio was also significantly higher in Type II than in the other two groups, with a statistically significant difference among the three groups (P < 0.001). These findings indicate that CEUS and UE characteristics differ significantly across pathological subtypes of CTL, particularly with Type II showing more pronounced necrotic and sclerotic features (Table 1).

Table 1 Comparison of Baseline Characteristics Among Different Pathological Subtypes

Diagnostic Performance of CEUS and UE for Different Pathological Subtypes

Diagnostic performance analysis is presented in Table 2. For differentiating Type II from non-Type II lesions, both CEUS and UE demonstrated high diagnostic value. CEUS achieved a sensitivity of 87.5% and a specificity of 87.5%. UE yielded a sensitivity of 85.0% and a specificity of 76.3%. When combined, the sensitivity increased to 95.0%, and the negative predictive value reached 97.1%. For differentiating Type I from non-Type I lesions, CEUS showed a sensitivity of 70.0% and a specificity of 75.0%. UE demonstrated a high specificity of 92.5% but a low sensitivity of 27.5%. After combination, specificity further increased to 98.8%, and the positive predictive value reached 90.9%, indicating high diagnostic certainty for Type I; however, a substantial proportion of cases would still be missed.

Table 2 Diagnostic Performance of CEUS and UE for Different Pathological Subtypes

Comparison of Baseline and Imaging Characteristics Between Treatment-Sensitive and Treatment-Insensitive Groups

There were no significant differences between the treatment-sensitive and treatment-insensitive groups in age, sex, history of tuberculosis, diabetes mellitus, or smoking history (all P > 0.05). However, the treatment-insensitive group had a significantly longer disease duration (P < 0.05). In terms of imaging features, the treatment-insensitive group more frequently exhibited annular enhancement on CEUS and had a significantly higher non-perfusion area ratio (P < 0.001). On UE, lesions in the treatment-insensitive group more often had elasticity score ≥ 4, and the strain ratio was significantly higher compared with the treatment-sensitive group (P < 0.001). These results suggest that poor treatment response is closely associated with longer disease duration, more extensive necrotic areas (non-perfusion), and increased tissue stiffness (Table 3).

Table 3 Comparison of Baseline and Imaging Characteristics Between Treatment-Sensitive and Treatment-Insensitive Groups

Multivariate Prediction Model for Treatment Sensitivity

Variables with P < 0.05 in univariate analysis were entered into a multivariate logistic regression model using a backward stepwise method (removal criterion P > 0.10). The final model retained five variables. The results demonstrated that strain ratio, non-perfusion area ratio, and disease duration were independent risk factors for treatment insensitivity. CEUS enhancement pattern and UE elasticity score did not reach statistical significance. The Hosmer–Lemeshow goodness-of-fit test showed χ² = 5.790 and P = 0.671, indicating good model calibration. Multicollinearity diagnostics revealed no significant collinearity among variables, with all VIF values below 5 (range: 1.029–1.204), confirming the robustness of the regression estimates.(Table 4).

Table 4 Multivariate Logistic Regression Model for Treatment Sensitivity

Comparison of Predictive Performance Among Imaging-Based Models

In the multivariate logistic regression analysis (Table 4), CEUS enhancement pattern (P = 0.100) and UE elasticity score category (P = 0.060) showed certain trends but did not reach statistical significance (P < 0.05). To construct a parsimonious and robust prediction model, only statistically significant core imaging parameters (non-perfusion area ratio and strain ratio) were included in the model comparison (Table 5). ROC analysis demonstrated that the AUC of CEUS was 0.782; the AUC of UE was 0.751; and the combined prediction model (CEUS + UE) achieved an AUC of 0.841.

Table 5 Predictive Performance of Different Imaging-Based Models

DeLong testing indicated that the area under the curve of the combined model was significantly higher than that of UE alone (P = 0.030), whereas the difference between the combined model and CEUS alone did not reach statistical significance (P = 0.066). There was no statistically significant difference between the area under the curve values of CEUS and UE as single predictors (P = 0.638) (Figure 1).

Figure 1 ROC Curves of Different Predictive Models.

Discussion

In this retrospective cohort of 120 patients with CTL, we systematically evaluated the dual value of CEUS and UE in pathological classification and treatment prediction. The main findings can be summarized as follows: first, quantitative and semi-quantitative parameters derived from CEUS and UE demonstrated distinct patterns among pathological subtypes, and imaging-based diagnostic criteria showed excellent performance in identifying Type II (caseous necrosis); second, imaging features represented by CEUS-derived non-perfusion area ratio and UE-derived strain ratio, together with disease duration, were independent risk factors for treatment insensitivity. These findings provide practical imaging tools to facilitate the transition of CTL management from empirical and homogeneous approaches toward precision and individualized strategies.

Value and Mechanisms of CEUS and UE in Pathological Subtype Differentiation

Our results showed that Type II lymph nodes exhibited significantly higher non-perfusion area ratios and a greater proportion of annular enhancement on CEUS, as well as higher strain ratios and elasticity scores on UE.¹⁰ These imaging characteristics have a clear pathophysiological basis. CEUS, by evaluating microvascular perfusion, directly reflects tissue vascular supply.¹¹ Caseous necrosis represents coagulative necrosis characterized by destruction of microvascular architecture and complete interruption of blood flow; consequently, CEUS demonstrates non-enhancing “black hole” regions.¹² In this study, a non-perfusion area ratio ≥ 20% yielded a sensitivity and specificity of 87.5% for diagnosing Type II, which is highly consistent with the findings of Zhang X et al¹³ regarding CEUS in detecting necrosis in tuberculous lymph nodes. Annular enhancement may correspond to residual granulomatous tissue or inflammatory reaction surrounding the necrotic core and represents a typical perfusion pattern of necrotic lesions.¹⁴

UE provides complementary information from the perspective of tissue biomechanics. Necrotic tissue loses normal cellular architecture and is often accompanied by fibrin deposition and, in later stages, calcification, resulting in increased stiffness.¹⁵ In this study, the strain ratio of Type II lymph nodes was significantly higher than that of the other subtypes, confirming increased overall stiffness. In contrast, pure granulomatous inflammation, although characterized by dense cellular infiltration, may not yet have formed extensive fibrotic stroma; therefore, its stiffness may overlap with that of normal soft tissue or reactive hyperplastic lymph nodes, limiting the discriminative ability of UE.¹⁶ This observation is consistent with previous reports describing controversy regarding the diagnostic value of UE in benign hyperplastic lymph node lesions.¹⁷

Nevertheless, when combined with CEUS features, the specificity and PPV for diagnosing Type I reached 98.8% and 90.9%, respectively, indicating that the combined criteria may serve as a high-certainty exclusion tool. This may assist clinicians in identifying cases dominated by granulomatous hyperplasia, which are theoretically more sensitive to pharmacotherapy, thereby supporting early refinement of treatment strategies.¹⁸

Imaging-Based Predictive Model for Treatment Response and Clinical Implications

Another key finding of this study is the establishment of a predictive model for treatment response. Both univariate and multivariate analyses confirmed that a larger pre-treatment non-perfusion area ratio and a higher strain ratio were independent risk factors for treatment insensitivity at 6 months. Extensive non-perfusion areas (representing large-scale necrosis) not only imply potentially higher bacterial burden and reduced drug penetration but may also indicate a local microenvironment less favorable for immune-mediated clearance. Meanwhile, increased tissue stiffness (high strain ratio) reflects more severe structural destruction and fibrosis, conditions under which lesion absorption and remodeling tend to be slower.^18,19

The combined prediction model (CEUS + UE) achieved an AUC of 0.841, outperforming single-modality models. This finding underscores the complementary value of multimodal functional imaging.²⁰ CEUS primarily reflects microcirculatory perfusion deficits, whereas UE emphasizes biomechanical properties. From two independent dimensions—hemodynamics and tissue stiffness—these techniques provide a more comprehensive assessment of pathological complexity and together constitute an integrated imaging biomarker for predicting treatment response. This finding has direct clinical translational potential. At the initiation of therapy, clinicians may use this model to stratify patients by risk. For patients predicted to be treatment-insensitive, more aggressive strategies may be considered, such as prolonged intensive therapy, closer imaging follow-up, or early evaluation of local interventional procedures (eg, aspiration or intralesional drug administration), thereby potentially improving long-term outcomes.²¹

Comparisons with previous studies further support our findings. Yu et al reported that post-treatment enlargement of non-enhancement areas was associated with poor prognosis;¹⁸ our study extends this by demonstrating that pre-treatment non-perfusion area ratio independently predicts treatment insensitivity. Zhang et al found that an initial necrosis rate ≥50% predicted unsatisfactory outcomes; our study shows that even a lower non-perfusion area ratio (median 33.95% in Type II) is an independent risk factor, and combining CEUS with UE further improves predictive performance.²²

Limitations

First, the sample composition was somewhat specific. During the inclusion period, the three pathological subtypes were equally represented (40 cases each). Although this facilitated balanced subgroup comparisons, the 1:1:1 ratio may not reflect the true distribution of CTL subtypes in the broader population. Therefore, while the observed differences in imaging characteristics are informative, the absolute diagnostic performance metrics (eg, sensitivity and specificity) require validation and recalibration in larger, multicenter, consecutive, and non-selective cohorts. Second, as a single-center retrospective study, the overall sample size was relatively limited. Specifically, in the multivariate predictive model, 38 events corresponded to five predictor variables. Given the limited number of events, statistical power and model stability may have been affected. Thus, the identified independent predictors (strain ratio, non-perfusion area ratio, and disease duration) and the model itself should be regarded as exploratory findings, requiring validation in large prospective cohorts. Third, although the cutoff values used in this study (eg, non-perfusion area ratio ≥ 20%, strain ratio ≥ 3.0) were defined based on data distribution characteristics and clinical consensus and demonstrated good performance in this cohort, their generalizability requires external validation across different populations, ultrasound systems, and operators for further optimization and standardization. Finally, no independent diagnostic criteria were established for Type III (mixed type) due to its pathological heterogeneity. Its imaging features typically lie between those of Type I and Type II. Future studies may explore artificial intelligence–based texture analysis or more complex multiparametric models to achieve more precise identification of this subtype.

Conclusion

CEUS and UE enable noninvasive assessment of vascular supply and tissue stiffness in CTL. In this cohort, both modalities demonstrated potential for identifying Type II caseous necrotic CTL. Pre-treatment imaging parameters may help identify patients at higher risk of poor therapeutic response. These findings provide preliminary evidence supporting imaging-based classification and individualized treatment strategies for CTL. However, broader clinical application requires validation through large-scale prospective studies.