Development and validation of an ultrasound-based predictive model for central lymph node metastasis in papillary thyroid carcinoma with peripheral calcification

Introduction

Thyroid cancer, particularly papillary thyroid carcinoma (PTC), has seen a dramatic increase in global incidence over the past decades, with approximately 820,000 new cases reported in 2022 (1). Although most PTC cases have a favorable prognosis, lymph node metastasis (LNM) remains a crucial factor influencing recurrence and survival, with its incidence varying by histological subtype and risk stratification (30–80%) (2).

Currently, the preoperative evaluation of central LNM (CLNM) primarily relies on suspicious ultrasound features of lymph nodes (LNs), such as the absence of a hilum, round shape, hyperechogenicity, cystic changes, microcalcifications, and a peripheral vascular pattern (3). However, the sensitivity of preoperative ultrasound in assessing CLNM for thyroid cancer is notably low (33%) due to significant occlusion and restricted detection angles (4). This diagnostic uncertainty fuels the ongoing controversy surrounding prophylactic central LN dissection (CLND). The 2015 American Thyroid Association (ATA) management guidelines reflect this dilemma, stating that prophylactic CLND may be performed for advanced primary tumors (T3 or T4) or clinically involved lateral nodes, but it is not routinely recommended for small (T1 or T2), noninvasive, clinically node-negative (cN0) tumors (5). This selective recommendation stems from a risk-benefit calculus: while CLND can provide accurate staging and potentially reduce locoregional recurrence, it carries risks of complications such as hypoparathyroidism and recurrent laryngeal nerve injury (6). Consequently, a precise, preoperative tool to identify patients who would truly benefit from CLND is highly desirable.

Notably, peripheral calcification, a characteristic ultrasound finding presenting as a calcification ring surrounding the nodule, is increasingly recognized as a marker of aggressive tumor biology (7,8). Current research on nodules with peripheral calcification primarily focuses on distinguishing between benign and malignant lesions (7,8). Existing studies have insufficiently addressed the aggressiveness of this specific type of thyroid cancer. While predictive models for LNM in thyroid cancer do exist (9-13), they are typically constructed using highly heterogeneous general populations. However, thyroid carcinomas with peripheral calcification represent a distinct high-risk subgroup with unique biological behavior. Applying models designed for all patients directly to this specific high-risk population may significantly compromise predictive accuracy and clinical utility. More critically, many of the key sonographic features utilized in these general models (e.g., internal echogenicity, margin characteristics) are precisely the features that are often obscured or become unreliable in the presence of a dense peripheral calcific rim due to acoustic shadowing. This fundamental limitation suggests that the direct application of general models to the peripherally calcified subgroup may lead to suboptimal performance.

Therefore, we hypothesize that constructing a specific CLNM model for this thyroid cancer subgroup with distinct high-risk ultrasound features could provide more precise and personalized risk stratification than universal models. This study aims to develop and validate such a model, and to provide clinicians with a precise tool to guide decisions regarding prophylactic CLND, reducing unnecessary surgery while enhancing the efficacy of metastasis monitoring in this unique subgroup. We present this article in accordance with the TRIPOD reporting checklist (available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0152/rc).

Methods

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by institutional review board of Shenzhen People’s Hospital (No.LL-KY-2022466) and individual consent for this retrospective analysis was waived

Patients

This single-center study was performed at a tertiary referral center in Shenzhen. While such centers may manage a higher proportion of complex cases, the study setting in Shenzhen—a major city with substantial nationwide population inflow—contributed to a patient cohort with diverse geographical and demographic characteristics. In this retrospective study, we included 210 consecutive peripherally calcified PTC patients who received treatment at our institution’s Thyroid Surgery Department during the period from January 2017 to July 2025. All data were extracted from the electronic medical records and the ultrasound report database, from which any personally identifiable information had been removed to ensure patient confidentiality. To ensure comprehensiveness, the ultrasound database was searched using the keywords “hyperechoic” AND “thyroid” and “calcification” AND “thyroid”. All extracted data were anonymized, containing no personally identifiable information. Based on the pathological results, patients in each dataset were stratified into CLNM-negative and CLNM-positive groups. Based on preoperative ultrasound assessments, all patients underwent either total thyroidectomy or lobectomy, accompanied by unilateral or bilateral CLND following the ATA management guidelines for adult patients with thyroid nodules and differentiated thyroid cancer [2015] (5).

Since the single-focal model can more purely reveal the association between peripheral calcification and metastasis, avoiding the dilution or amplification effects of multi-focal lesions on the results, our study only included cases with a single malignant nodule with peripheral calcification. All cases must rely on postoperative paraffin pathology (rather than intraoperative frozen section) to avoid missing small cancer foci (<0.1 cm). If there are benign nodules in the same lobe or benign lesions in the opposite lobe, the cancer can still be included as long as it is a single lesion.

Selection criteria: (I) underwent thyroidectomy, central neck LN dissection, and therapeutic lateral neck LN dissection if necessary; (II) histopathological examination confirmed PTC; (III) postoperative pathology confirmed the presence of a single cancerous lesion (no other cancerous lesions in the contralateral lobe or ipsilateral lobe); (IV) peripheral calcification was clearly detected by ultrasound within one month before surgery.

Exclusion criteria: (I) incomplete ultrasound imaging data or poor image quality; (II) history of thyroid surgery or neck radiotherapy; (III) presence of distant metastases or malignant tumors in other organs.

Ultimately, 210 eligible patients were included and randomly assigned to the training and validation cohorts in a 7:3 ratio. The patient selection process is illustrated schematically in Figure 1.

Figure 1 Flowchart of the patient enrollment process. CLNM, central lymph node metastasis; PTC, papillary thyroid carcinoma; US, ultrasound.

Surgery and pathology

All patients were treated by an experienced surgical team. Thyroid nodules suspicious for malignancy underwent ultrasound-guided fine-needle aspiration (FNA). Patients with a cytological diagnosis of PTC proceeded to surgery or active surveillance based on patient preference and surgeon assessment.

The surgical extent was determined as follows: lobectomy was performed for confirmed PTC without evidence of bilateral foci, capsular invasion, or LN/distant metastasis on preoperative evaluation; otherwise, total thyroidectomy was conducted. A comprehensive central neck dissection (level VI) was routinely performed. The dissection boundaries were clearly defined, and all resected lymphatic and adipose tissues from the central compartment were submitted for pathological examination. To ensure detection of occult metastases, all identifiable LNs were entirely embedded, serially sectioned, and stained with hematoxylin and eosin (H&E) for evaluation by an experienced pathologist, thereby establishing a reliable histological gold standard for CLNM.

Ultrasound examination

All patients underwent preoperative high-resolution ultrasound examination using a linear transducer with a frequency of 3–15 MHz. All ultrasound assessments were carried out by radiologists possessing a minimum of two years’ expertise in diagnosing thyroid lesions. During the examination, patients were positioned supine with their necks exposed. Grayscale ultrasound was utilized to evaluate the nodules based on their location, size, and ultrasound characteristics. Both transverse and longitudinal images of the target nodules were captured.

Ultrasound image acquisition and analysis of ultrasound features

The ultrasound features were re-evaluated specifically for this study. Two experienced radiologists (S.B. and L.W.) with over 10 years of experience, who were blinded to the final pathological results and the original clinical ultrasound reports, independently assessed all preoperative ultrasound images. The images were anonymized to conceal all patient identifiers and previous diagnostic impressions. Any discrepancies that arose were addressed through discussions with a senior physician (Y.S.). This blinded review process, was implemented to minimize potential assessment bias in this retrospective study. Recognizing that the diagnostic efficacy of our model depends on the accuracy of the imaging features reported by the operators, we assessed the interobserver reproducibility of these ultrasound features. In our investigation, the selection of ultrasound features was primarily guided by the criteria established by the American College of Radiology Thyroid Imaging Reporting and Data System (ACR TI-RADS); additionally, we incorporated tumor vascularity as a variable of interest.

Clinical variables such as age, gender, and Hashimoto’s thyroiditis (HT) were also collected. Ultrasound features mainly included nodule size (maximum diameter, ≤1 cm, >1 to ≤2 cm or >2 cm), location (upper, middle, lower or isthmus), abutment-to-lesion perimeter ratio (A/P) (≤0.25 or >0.25), margin (regular, irregular/lobulated or invisible), shape [longitudinal diameter/transverse diameter (L/T) ratio, ≤1 or >1], composition (solid-cystic, solid or invisible), internal echogenicity (extremely hypoechoic/hypoechoic, isoechoic/hyperechoic compared to the background thyroid gland or invisible because of posterior acoustic attenuation), internal calcification (none or large comet-tail, macrocalcification, microcalcification or invisible), internal vascularity (poorly perfused, well perfused or invisible), and US-reported central LN (CLN) status. The characteristics of the calcification rim were also analyzed, including the type of peripheral calcification, continuity (interrupted or continuous), range of peripheral calcification (circular or arc), halo sign (present or absent), and extrusion beyond calcification (yes or no).

Tumor location was classified into the upper, middle, or lower third of either thyroid lobe or the isthmus. Due to the lack of clear anatomical divisions or guidelines, the thyroid gland was divided into three equal volumetric parts (upper, middle, and lower poles) based on the consensus of most medical centers. If the dominant lesion spanned two adjacent areas, the tumor location was designated by the site that constituted more than two-thirds of the tumor volume.

Tumor vascularity was classified according to the Adler criterion (14), ranging from Grade 0 to 3, and assessed using color Doppler flow imaging (CDFI). Grades 0 and 1 are classified as poorly perfused, while Grades 2 and 3 are classified as well perfused. The relationship of the tumor to the thyroid capsule was evaluated based on ultrasound imaging. Diagnosis of HT primarily relies on laboratory tests, with ultrasonographic findings serving as auxiliary diagnostic criteria. The abutment was defined as direct contact between the edge of a thyroid nodule and the thyroid capsule on ultrasound. To quantify the extent of extra-thyroidal extension, we introduced the concept of A/P.

The cutoff of A/P was set at 1/4 (25%), based on established criteria in prior studies (9,10,15). A tumor was considered suspicious for extrathyroidal extension (ETE) on ultrasonography when more than 25% of its perimeter exhibited capsular abutment (16).

According to the surface characteristics, peripheral calcifications were classified into one of the following three categories: type I, coarse calcification (curvilinear, smooth surface, even thickness); type II, coarse calcification (curvilinear, rough surface, uneven thickness); and type III, stippled (small and non-linear calcification spots). The demarcation between coarse and small calcification is defined as having a diameter of 2 mm. Furthermore, nodules are categorized as either arc or circular based on the extent of calcification surrounding the nodule’s periphery. A nodule is deemed circular when the total length of calcification is equal to or exceeds two-thirds of the nodule’s margin; conversely, it is classified as arc-shaped when this length is less than two-thirds of the margin.

US-reported CLN status

Clinically positive CLNM was suggested if the following ultrasound features were observed: absence of LN hilum, round shape, microcalcifications, peripheral vascular flow, or cystic changes. The decision to perform a lateral LN dissection was based on preoperative imaging findings.

Selection of potential predictive factors

Least absolute shrinkage and selection operator (LASSO) regression performs variable selection and regularization simultaneously, thereby enhancing both the predictive accuracy and interpretability of the resulting model. To mitigate the influence of superfluous variables and prevent overfitting of the model, we employed LASSO regression (17) to identify the most significant predictive features from the entire set of clinical-pathological and ultrasound attributes within the training cohort. The mechanism of the LASSO regression model involves the compression of coefficients for non-essential features to zero, governed by the regularization parameter λ, while retaining those features with non-zero coefficients as selected by LASSO.

The development of the Clinical-ultrasound (Clin-US) nomogram in the training cohort

The potential predictors identified through LASSO regression were utilized in a multivariate logistic regression analysis to establish a predictive model. This model is illustrated as a Clin-US nomogram, which offers a clear representation of the probability of CLNM.

Evaluation of the predictive model

The ability of the model to discriminate was assessed using a receiver operating characteristic (ROC) curve. The nomogram was subjected to bootstrapping validation (1,000 bootstrap resamples) to calculate a relatively robust area under the curve (AUC). We evaluated the model’s calibration by generating calibration curves based on the nomogram-predicted CLNM probabilities and the actual pathological results. The goodness of fit of the model was assessed using the Hosmer-Lemeshow test, a standard statistical method for evaluating model calibration (18). Decision curve analysis (DCA) was further performed to evaluate the clinical utility and quantify the net benefit of the nomogram.

Statistical analysis

The sample size of this study is a consecutive cohort based on all eligible patients treated at our center. In developing clinical prediction models, a key methodological consideration is minimizing overfitting, which is commonly assessed using the events per variable (EPV) criterion. A rule of thumb suggests a minimum of 10 outcome events per candidate predictor parameter (EPV ≥10) included in the model to ensure reliable estimates. In the training set of this study (n=150), there were 86 events. With 7 predictor variables retained in the final model, the EPV was approximately 12.3, meeting this conventional threshold.

Statistical representations of categorical variables are denoted as n (%), while continuous variables are expressed as means with standard errors or medians with corresponding quantile values (Q1, Q3). For the analysis of continuous variables, the Student’s t-test was employed when the variable exhibited a normal distribution; conversely, the Mann-Whitney U test was utilized in cases where normality was not assumed. To compare categorical variables, χ²-tests or Fisher’s exact test were applied.

Missing data occurred in five ultrasonographic features (margin, internal echogenicity, composition, internal calcification, and CDFI), with 8.1–9.0% missing values (Figure S1). As acoustic shadowing may cause non‑random missingness, we used multiple imputation (MI) to retain all cases (n=210) and preserve statistical power, rather than using complete-case analysis (CCA) which could introduce bias. To evaluate the impact of missing data handling, we performed a sensitivity analysis using standard logistic regression on the fixed set of predictors from our primary LASSO model. This approach isolates the effect of imputation by preventing the re-selection of variables, which would occur if LASSO were reapplied. Regression coefficients from both models were directly compared.

Prior to the assessment of the validation dataset, internal validation of the training dataset was conducted through bootstrap resampling. All statistical significance tests were executed as two-tailed, with a significance threshold established at P<0.05.

To evaluate the level of agreement among radiologists regarding the assessment of nodules, Cohen’s kappa statistic was employed. The kappa coefficients were interpreted according to the following scale: 0≤κ≤0.4 indicates poor agreement; 0.41≤κ≤0.75 suggests fair to good agreement; 0.76≤κ<1.0 signifies excellent agreement; and κ=1.0 represents perfect agreement. Analyses were conducted utilizing SPSS Version 23 (IBM, Armonk, NY, USA) along with R software (Ri386 4.0.3, R Foundation for Statistical Computing, Vienna, Austria).

Results

Clinical-pathological data and US features

This study enrolled 210 patients with PTC. These patients were randomly divided into a training dataset and a validation dataset at a ratio of 7:3, with the training dataset including 150 cases and the validation dataset including 60 cases. Table 1 summarizes the detailed clinical-pathological data and US features of patients in the training and validation datasets.

The entire cohort included 131 females (62.4%) and 79 males (37.6%). The median (Q1, Q3) patient age was 39.0 (30.0, 51.0) years for the training cohort and 36.0 (29.5, 49.0) years for the validation cohort. The rates of aggressiveness were comparable between the training (57.3%) and validation (56.7%) cohorts (P=0.93). Furthermore, no significant differences were observed in clinical-pathological characteristics and other ultrasound features (all P>0.05), with the exception of the halo sign, which showed a statistically significant difference (P=0.03).

Selection of potential predictive factors

Ultimately, age, A/P, location, US-reported CLN status, halo sign, extrusion beyond calcification, and type of peripheral calcification exhibited non-zero coefficients when the LASSO logistic regression model was applied (Figure 2A,2B).

Figure 2 Selection of characteristics using the LASSO logistic regression model in the training dataset. (A) The 18 ultrasound and clinical characteristics are profiled by the LASSO coefficient. A coefficient plot against the log (λ) sequence was produced; (B) Selection of the tuning parameter (λ) in the LASSO model via 10-fold cross-validation was based on minimum criteria. The optimal λ value of 0.050 with log (λ) =−2.990 was chosen. Two vertical lines were drawn at the optimal value by using the minimum criteria and the one standard error of the minimum criteria. LASSO, least absolute shrinkage and selection operator.

A sensitivity analysis was performed to evaluate the potential impact of missing data on our primary model’s findings. The pattern of missing data is summarized in showing that five key ultrasound features had missing values ranging from 8.1% to 9.0%. We compared the results of a CCA with those from an MI analysis (MIA) using standard logistic regression.

The comparison revealed a high degree of consistency in both the magnitude of effect estimates and statistical inference. As detailed in Figure S2A, the regression coefficients for the vast majority of predictors showed close agreement before and after imputation. Figure S2B quantifies this agreement by displaying the absolute difference in coefficients (MIA-CCA) for each variable; the differences were minimal for most predictors, with none exceeding 0.7 log-odds units.

Specifically, six variables demonstrated stable and statistically significant associations (P<0.05) in both analytical approaches (Figure S2C). Notably, the seventh variable included in the primary LASSO model (tumor location) may only have marginal relevance, which could be due to its P value fluctuating near the 0.05 threshold, leading to unstable results across different analytical methods; this reflects the expected sensitivity of such marginal associations to different analytical methods. Crucially, the effect estimates and confidence intervals (CIs) for the core predictors driving the model were virtually identical. This comprehensive analysis confirms that the main findings of our study are robust to the method of handling missing data.

Establishment of the predictive model

A nomogram was constructed that integrates seven significant diagnostic factors and is illustrated in Figure 3. Each variable within the model is assigned a specific score according to a defined scoring scale. The risk probability of CLNM for each patient is determined by aggregating the total scores and referencing them on the total score scale.

Figure 3 Clin-US nomogram for estimating the risk of CLNM for PTC nodules with peripheral calcification. A/P, abutment-to-lesion perimeter ratio; Clin-US, Clinical-ultrasound; CLN, central lymph node; CLNM, central lymph node metastasis; PTC, papillary thyroid carcinoma; US, ultrasound.

Evaluation and validation of the nomogram

The internal validation of the nomogram was performed using the bootstrap validation technique. In the training cohort, after 1,000 bootstrap resampling iterations, the C-index reached 0.942 (95% CI: 0.904–0.980) (Figure 4), indicating satisfactory discriminative ability. The calibration curve (Figure 5A) showed good agreement between predicted and observed CLNM risk, and the ROC curve (Figure 4) yielded an AUC of 0.942, further confirming good calibration performance.

Figure 4 ROC curves based on the Clin-US nomogram for CLNM of PTC nodules with peripheral calcification in the training and validation cohorts. AUC, area under the curve; Clin-US, clinical-ultrasound; CLNM, central lymph node metastasis; FPR, false positive rate; PTC, papillary thyroid carcinoma; ROC, receiver operating characteristic; TPR, true positive rate.

Figure 5 Calibration curves of the Clin-US nomogram for the training (A) and validation (B) cohorts. Clin-US, clinical-ultrasound; ROC, receiver operating characteristic.

Using the optimal probability cut-off point of 0.548, the model exhibited a sensitivity of 0.919, specificity of 0.891, positive predictive value (PPV) of 0.919, and negative predictive value (NPV) of 0.891 in the training set.

The model’s performance was further validated in an independent validation cohort. The C-index was 0.870 (95% CI: 0.777–0.962), with a corresponding AUC of 0.870 (Figure 4), and the calibration curve remained well-fitted (Figure 5B). At the same cut-off of 0.548, the sensitivity was 0.735, specificity 0.846, PPV 0.862, and NPV 0.710 in the validation set.

Clinical usefulness

The DCA revealed that the nomogram provided a favorable net benefit for predicting CLNM when threshold probabilities were between 12% and 89% (Figure 6).

Figure 6 Decision curve analysis of the Clin-US nomogram. The x-axis represents the threshold probability, and the y-axis represents the net benefit. Clin-US, clinical-ultrasound.

Agreement among radiologists

The interobserver agreement between the two radiologists for all ultrasound features was substantial to excellent, with kappa coefficients ranging from 0.77 to 0.94 (Table 2).

Examples of the nomogram in use

Figures 7A-7H show exemplary cases of CLNM risk prediction in PTC patients based on the nomogram.

Figure 7 Examples of clinical application of the Clin-US nomogram. The red arrow in the figures indicates the halo sign and the yellow arrow indicates the extrusion beyond calcification. Variables not marked with red arrows belong to the reference category and contribute 0 points to the total score. (A) Image was obtained from a 69-year-old woman with nodule in the left thyroid. (B) The nomogram resulted in a total score of 191 points for age (2 points), halo sign (50 points), lower pole (39 points), and A/P>0.25(100 points). The corresponding risk of CLNM was low (0.07), and the pathological result of the nodule was PTC without CLNM. (C) Image was obtained from a 26-year-old man with nodule in the isthmus part. (D) The nomogram resulted in a total score of 246 points for age (62 points), halo sign (50 points), type II calcification (34 points), and A/P >0.25 (100 points). The corresponding risk of CLNM was relatively low (0.31), and the pathological result of the nodule was PTC without CLNM. (E) Image was obtained from a 35-year-old man with nodule in the right thyroid. (F) The nomogram resulted in a total score of 435 points for age (49 points), middle pole (88 points), halo sign (50 points), US-reported CLN status (+) (61 points), A/P >0.25 (100 points), and type III calcification (87 points). The corresponding risk of CLNM was 0.99, and the pathological result of the nodule was PTC with CLNM. (G) Image was obtained from a 59-year-old woman with nodule in the right thyroid. (H) The nomogram resulted in a total score of 342 points for age (15 points), middle pole (88 points), halo sign (50 points), extrusion beyond calcification (89 points), and and A/P >0.25 (100 points). The corresponding risk of CLNM was 0.91, and the pathological result of the nodule was PTC with CLNM. A/P, abutment-to-lesion perimeter ratio; Clin-US, clinical-ultrasound; CLN, central lymph node; CLNM, central lymph node metastasis; PTC, papillary thyroid carcinoma; US, ultrasound.

Patient 1 was a 69-year-old woman with a left thyroid nodule shown in Figure 7A. The nodule had two high-risk sonographic features: halo sign and A/P >0.25. The nomogram-predicted probability of CLNM was 7% (Figure 7B). Pathology confirmed PTC without CLNM.

Patient 2 was a 26-year-old man with an isthmus nodule shown in Figure 7C, which exhibited four high-risk sonographic features: 26 years old, halo sign, type II calcification, and A/P>0.25. The nomogram-predicted probability of CLNM was 31% (Figure 7D). Pathology confirmed PTC without CLNM.

Patient 3, shown in Figure 7E, was a 35-year-old man with a right thyroid tumor. The tumor had six high-risk sonographic features: age 35 years, mid-pole location, halo sign, US-reported CLN status (+), A/P>0.25, and type III calcification. The nomogram predicted a 99% probability of CLNM (Figure 7F). Postoperative pathology confirmed PTC with CLNM.

Patient 4, shown in Figure 7G, was a 59-year-old woman with a right thyroid tumor. The tumor had four high-risk sonographic features: mid-pole location, halo sign, extrusion beyond calcification, and A/P >0.25. The nomogram predicted a 91% probability of CLNM (Figure 7H). Pathology confirmed PTC with CLNM.

Discussion

This study developed and validated a predictive model for CLNM in patients with thyroid cancer with peripheral calcification. The model focuses on this subgroup of patients who present unique ultrasound features and clinical value. We identified seven independent predictive factors through multivariate logistic regression analysis: age, A/P, location, US-reported CLN status, halo sign, extrusion beyond calcification, and type of peripheral calcification. The model demonstrated excellent discrimination and calibration, providing clinicians with an intuitive and practical quantitative tool for preoperative assessment of the aggressiveness of this special type of thyroid cancer, helping to guide individualized surgical decisions.

The most important finding of this study is the development of a clinically prognostic subclassification of peripheral calcification morphology. We categorized it into coarse calcification (curvilinear, smooth surface, even thickness) (Type I), coarse calcification (curvilinear, rough surface, uneven thickness) (Type II), and stippled calcification (Type III), confirming a significantly increasing trend in CLNM risk across these three types. In our study, we refined the differentiation of coarse calcification smoothness within the peripheral calcification framework. Our findings suggest that smooth calcification may represent a relatively stable, well-encapsulated status. Conversely, irregular coarse calcification with rough margins is more likely related to ongoing, correlating with higher metastatic risk. Stippled calcification, as the highest-risk type, aligns perfectly with the aggressive biology of classic papillary carcinomas (19). These essentially represent sand-like bodies, signifying rapid cellular proliferation and necrosis.

This study proposed and validated extrusion beyond calcification and A/P >0.25 as independent predictors of CLNM. The former indicates that the tumor is not only surrounded by the calcified ring, but also has penetrated the surrounding tissues; the latter is a clear manifestation of thyroid cancer invasion of the membrane. Pathologically, these two features directly correspond to the destruction of the extracellular matrix and the invasion of lymphatic vessels and blood vessels. Specially, the extrusion beyond calcification feature signifies the breach of the calcific rim by soft tissue. This imaging finding is biologically significant, as it likely represents a correlate of local basement membrane destruction and invasive growth. Furthermore, this direct extension facilitates tumor access to and invasion of peritumoral lymphatic channels, providing a mechanistic link to its strong association with LNM. Consequently, they are considered to be potent predictors of CLNM, consistent with our pathological understanding and numerous previous studies. In this study, the A/P ratio was employed as a quantitative measure for ETE and identified as an independent predictor of CLNM in peripherally calcified PTC, a finding consistent with previous reports (20,21).

This study found that the presence of the halo sign is significantly associated with a higher risk of CLNM. Currently, a growing number of studies indicate that the halo sign is not only present in benign thyroid nodules but can also occur in malignant nodules, typically appearing irregular and thickened (22). In the special condition of malignant nodules with peripheral calcification, the halo sign may have a different pathological significance. Invasive tumors may induce more pronounced connective tissue proliferation or inflammatory responses at the interface between the calcification ring and the anterior margin of the invasive tumor.

This study confirms that age is an independent risk factor for CLNM. This finding is consistent with previous research conclusions (23,24). Younger patients (typically those under 40 years old) often have more aggressive tumor behavior and a higher tendency for LNM, which may be related to factors such as more active cellular metabolism, differences in the tumor microenvironment, and hormone levels in younger patients (25). Therefore, for young patients with peripherally calcified PTC, even if other characteristics of the primary lesion are not significant, there should be a high alert for the possibility of CLNM, and prophylactic CLN dissection should be considered when necessary.

This study found that nodules located in the isthmus are a protective factor against CLNM, while those in the mid-region demonstrate the highest risk—a finding that contradicts the conventional conclusion (26,27). We hypothesize this phenomenon may relate to differences in lymphatic drainage pathways across thyroid regions. The dispersed lymphatic drainage in the isthmus may hinder the formation of concentrated central metastatic foci, whereas the mid-region’s more concentrated lymphatic drainage into the pretracheal and anterior laryngeal areas (Zone VI) facilitates cancer cell retention and proliferation. Additionally, clinical diagnostic bias should not be overlooked: superficial isthmus nodules are more likely to be detected and treated early before metastasis occurs, while mid-region nodules are typically identified at later stages with potentially larger tumor burdens. Another possibility involves the inherent biological heterogeneity of tumors at different anatomical locations, though this requires further basic research for confirmation.

This study demonstrates that US-reported CLN status is an independent predictor for CLNM in peripherally calcified thyroid cancer. This finding is closely consistent with clinical expectations. Given the high specificity of ultrasound in detecting cervical LNM (27), suspicious LNs identified on ultrasound carry a high probability of malignancy. In line with this, early studies have incorporated US-reported CLN status into radiomics-based nomograms for predicting both central and lateral cervical LNM in patients with PTC, demonstrating favorable predictive performance (28). These results strongly support the significance of detailed preoperative scanning of CLN regions and suggest that suspicious LNs should be actively recommended for concurrent central neck dissection during surgery.

Although male gender has been reported as a potential risk factor for LNM in thyroid cancer (24,28), our study demonstrated that gender was not an independent predictor of CLNM in thyroid cancer with peripheral calcification (P>0.05), and thus was excluded from the final predictive model. This discrepancy may be attributed to the specific population and nodule features analyzed in this study. Our study specifically focused on thyroid cancer with peripheral calcification. Calcification itself serves as a strong imaging marker indicating malignancy and aggressive behavior. Other included factors—such as extrusion beyond calcification, halo sign, and A/P are more direct and powerful morphological indicators reflecting tumor invasiveness and metastatic potential compared to gender.

Previous study has shown that PTC patients with HT were associated with less aggressive diseases (29). In other words, absence of HT is an independent factor in PTC patients who have cervical LNM (30). However, we found that HT is not an independent predictor of CLNM in our study. The possible reason is that we judged the presence or absence of HT by preoperative US but not postoperative pathology.

In our study, CDFI was not significantly different between the CLNM-positive and CLNM-negative groups. Previous study showed that the richer the blood supply is, the higher the probability of CLNM (31). Regardless, the evaluation of the US for internal vascularity is unreliable and easily influenced by the operator and the machine. This may be why CDFI was not included in the model.

Other ultrasound features, such as L/T ratio, continuity, range of peripheral calcification, margin, internal echogenicity, composition, and internal calcification, were excluded from the model, likely due to multicollinearity with the seven selected factors, meaning their predictive information was already captured by more powerful variables. For example, the irregular margin feature is more directly represented by extrusion beyond calcification and A/P, while the risk information of internal calcification is covered by the stippled calcification type. The L/T ratio >1 is typically an important indicator for assessing malignant risk in primary thyroid lesions, but it may be limited for evaluating CLNM. Additionally, this study focused on nodules with peripheral calcification, a category predominantly composed of round or oval-shaped nodules, which resulted in insufficient variability within the cohort, leading to inadequate statistical power to provide effective discrimination in the model. Similarly, most nodules in this cohort exhibited interrupted continuity, lacking sufficient variation to be a predictor. Similarly, the majority of nodules present hypoechogenic and solid components, which is a common characteristic of PTC. Notably, arc-shaped and circular-shaped calcification showed no significant differences in CLNM.

The main innovation of this study is the construction of a CLNM predictive model specifically focused on thyroid cancer with peripheral calcification. Although there are general LNM prediction models, nodules with this calcification pattern have unique characteristics in terms of diagnosis and prognosis assessment. This study fills this gap by identifying predictive factors applicable to this subgroup. The study conducted an in-depth analysis of the predictive value of calcification morphological details (e.g., smoothness, thickness uniformity). Compared to previous studies that primarily focused on the presence or absence of calcification or a simple dichotomy between coarse and punctate calcification, this research provided a more refined classification of calcification and revealed their risk gradient, offering clinicians a more precise assessment tool. This study also proposes and validates the clinical predictive value of extrusion beyond calcification and provides a more refined interpretation of the significance of the halo sign in such nodules.

This study has the following limitations. Firstly, this study is retrospective and inevitably subject to selection bias. Secondly, a limitation of this study is its single-center design. Despite serving a diverse population, our cohort shares homogeneity in clinical protocols, ultrasound equipment, and operator expertise. Consequently, the model’s performance, even after internal validation, may not be directly generalizable to other institutions. Therefore, rigorous external validation in a prospective, multi-center cohort is mandatory before any clinical application can be considered. Furthermore, molecular/tissue subtype information is limited, and this study did not analyze ultrasound features or the correlation between model predictions and specific molecular genes (such as BRAF). This information may further optimize risk prediction, and it is necessary to include this gene in future studies.

Conclusions

This study successfully developed and internally validated a clinical prediction model using conventional US features to preoperatively assess CLNM risk in PTC with peripheral calcification. The model integrates key indicators such as extrusion beyond calcification. The findings provide a foundational tool for risk stratification. A necessary next step is external validation in multi-center cohorts to evaluate its performance in real-world settings, which is crucial for translating this model into precise surgical planning and optimized patient management.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0152/rc

Data Sharing Statement: Available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0152/dss

Peer Review File: Available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0152/prf

Funding: This project was supported by International Science and Technology Cooperation Project of Shenzhen Science and Technology Innovation Committee (No. GJHZ20220913142801003) held by Professor L.W.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0152/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 institutional review board of Shenzhen People’s Hospital (No. LL-KY-2022466) and individual consent for this retrospective analysis was waived.

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. Jiang LH, Yin KX, Wen QL, et al. Predictive Risk-scoring Model For Central Lymph Node Metastasis and Predictors of Recurrence in Papillary Thyroid Carcinoma. Sci Rep 2020;10:710. [Crossref] [PubMed]
3. Bayramoglu Z, Caliskan E, Karakas Z, et al. Diagnostic performances of superb microvascular imaging, shear wave elastography and shape index in pediatric lymph nodes categorization: a comparative study. Br J Radiol 2018;91:20180129. [Crossref] [PubMed]
4. Zhao H, Li H. Meta-analysis of ultrasound for cervical lymph nodes in papillary thyroid cancer: Diagnosis of central and lateral compartment nodal metastases. Eur J Radiol 2019;112:14-21. [Crossref] [PubMed]
5. Haugen BR, Alexander EK, Bible KC, et al. 2015 American Thyroid Association Management Guidelines for Adult Patients with Thyroid Nodules and Differentiated Thyroid Cancer: The American Thyroid Association Guidelines Task Force on Thyroid Nodules and Differentiated Thyroid Cancer. Thyroid 2016;26:1-133. [Crossref] [PubMed]
6. Carmel-Neiderman NN, Mizrachi A, Yaniv D, et al. Prophylactic central neck dissection has no advantage in patients with metastatic papillary thyroid cancer to the lateral neck. J Surg Oncol 2021;123:456-61. [Crossref] [PubMed]
7. Malhi HS, Velez E, Kazmierski B, et al. Peripheral Thyroid Nodule Calcifications on Sonography: Evaluation of Malignant Potential. AJR Am J Roentgenol 2019;213:672-5. [Crossref] [PubMed]
8. Nabahati M, Ghaemian N, Moazezi Z, et al. Different sonographic features of peripheral thyroid nodule calcification and risk of malignancy: a prospective observational study. Pol J Radiol 2021;86:e366-71. [Crossref] [PubMed]
9. Zhu J, Chang L, Li D, et al. Nomogram for preoperative estimation risk of lateral cervical lymph node metastasis in papillary thyroid carcinoma: a multicenter study. Cancer Imaging 2023;23:55. [Crossref] [PubMed]
10. Zou Y, Shi Y, Bi H, et al. A nomogram for risk stratification of central cervical lymph node metastasis in patients with papillary thyroid carcinoma. Quant Imaging Med Surg 2024;14:5084-98. [Crossref] [PubMed]
11. Qiu P, Guo Q, Pan K, et al. Development of a nomogram for prediction of central lymph node metastasis of papillary thyroid microcarcinoma. BMC Cancer 2024;24:235. [Crossref] [PubMed]
12. Shen P, Yang Z, Sun J, et al. Explainable multimodal deep learning for predicting thyroid cancer lateral lymph node metastasis using ultrasound imaging. Nat Commun 2025;16:7052. [Crossref] [PubMed]
13. Zhang J, Zhang M, Xu S, et al. Model for Predicting Central Lymph Node Metastasis in Papillary Thyroid Carcinoma: A Study Based on Ultrasound Viscosity Imaging. Ultrasound Med Biol 2025;51:2032-8. [Crossref] [PubMed]
14. Adler DD, Carson PL, Rubin JM, et al. Doppler ultrasound color flow imaging in the study of breast cancer: preliminary findings. Ultrasound Med Biol 1990;16:553-9. [Crossref] [PubMed]
15. Sun F, Zou Y, Huang L, et al. Nomogram to Assess the Risk of Central Cervical Lymph Node Metastasis in Patients With Clinical N0 Papillary Thyroid Carcinoma. Endocr Pract 2021;27:1175-82. [Crossref] [PubMed]
16. Wei X, Wang M, Wang X, et al. Prediction of cervical lymph node metastases in papillary thyroid microcarcinoma by sonographic features of the primary site. Cancer Biol Med 2019;16:587-94. [Crossref] [PubMed]
17. Sauerbrei W, Royston P, Binder H. Selection of important variables and determination of functional form for continuous predictors in multivariable model building. Stat Med 2007;26:5512-28. [Crossref] [PubMed]
18. Kramer AA, Zimmerman JE. Assessing the calibration of mortality benchmarks in critical care: The Hosmer-Lemeshow test revisited. Crit Care Med 2007;35:2052-6. [Crossref] [PubMed]
19. Huang XP, Ye TT, Zhang L, et al. Sonographic features of papillary thyroid microcarcinoma predicting high-volume central neck lymph node metastasis. Surg Oncol 2018;27:172-6. [Crossref] [PubMed]
20. Yang Y, Chen C, Chen Z, et al. Prediction of central compartment lymph node metastasis in papillary thyroid microcarcinoma. Clin Endocrinol (Oxf) 2014;81:282-8. [Crossref] [PubMed]
21. Lee YC, Na SY, Chung H, et al. Clinicopathologic characteristics and pattern of central lymph node metastasis in papillary thyroid cancer located in the isthmus. Laryngoscope 2016;126:2419-21. [Crossref] [PubMed]
22. Cao D, Zou R, Zhang M, et al. Sonographic characteristics of thyroid nodules with a Halo. Thyroid Res 2024;17:20. [Crossref] [PubMed]
23. Xu SY, Yao JJ, Zhou W, et al. Clinical characteristics and ultrasonographic features for predicting central lymph node metastasis in clinically node-negative papillary thyroid carcinoma without capsule invasion. Head Neck 2019;41:3984-91. [Crossref] [PubMed]
24. Oh HS, Park S, Kim M, et al. Young Age and Male Sex Are Predictors of Large-Volume Central Neck Lymph Node Metastasis in Clinical N0 Papillary Thyroid Microcarcinomas. Thyroid 2017;27:1285-90. [Crossref] [PubMed]
25. Ito Y, Miyauchi A, Kihara M, et al. Patient age is significantly related to the progression of papillary microcarcinoma of the thyroid under observation. Thyroid 2014;24:27-34. [Crossref] [PubMed]
26. Zhan S, Luo D, Ge W, et al. Clinicopathological predictors of occult lateral neck lymph node metastasis in papillary thyroid cancer: A meta-analysis. Head Neck 2019;41:2441-9. [Crossref] [PubMed]
27. Lee YS, Shin SC, Lim YS, et al. Tumor location-dependent skip lateral cervical lymph node metastasis in papillary thyroid cancer. Head Neck 2014;36:887-91. [Crossref] [PubMed]
28. Wang F, Zhao S, Shen X, et al. BRAF V600E Confers Male Sex Disease-Specific Mortality Risk in Patients With Papillary Thyroid Cancer. J Clin Oncol 2018;36:2787-95. [Crossref] [PubMed]
29. Wang Y, Zheng J, Hu X, et al. A retrospective study of papillary thyroid carcinoma: Hashimoto’s thyroiditis as a protective biomarker for lymph node metastasis. Eur J Surg Oncol 2023;49:560-7. [Crossref] [PubMed]
30. Ye F, Gong Y, Tang K, et al. Contrast-enhanced ultrasound characteristics of preoperative central cervical lymph node metastasis in papillary thyroid carcinoma. Front Endocrinol (Lausanne) 2022;13:941905. [Crossref] [PubMed]
31. Jang JY, Kim DS, Park HY, et al. Preoperative serum VEGF-C but not VEGF-A level is correlated with lateral neck metastasis in papillary thyroid carcinoma. Head Neck 2019;41:2602-9. [Crossref] [PubMed]