Construction method and application of thyroid cancer lateral neck lymph node metastasis prediction model

Abstract

The invention relates to a thyroid cancer side neck lymph node metastasis prediction model construction method and application. The thyroid cancer side neck lymph node metastasis prediction model comprises a Logistic regression model and a decision process module. The Logistic regression model is used for calculating an actual risk value based on a value of a primary tumor maximum diameter, a value of a tumor boundary, a value of a BRAF V600E mutation state, a value of micro-calcification existence, a value of an aspect ratio and a value of single / double-side cancerous lesion; the decision process module is used for giving a decision suggestion according to the calculated actual risk value and a set decision process; the model provides a new quantitative tool for preoperative evaluation of papillary thyroid carcinoma patients, can provide individualized and precise treatment schemes for clinic, reduces unnecessary treatment and missed diagnosis, and finally improves the treatment effect and life quality of the patients.

Description

TECHNICAL FIELD

[0001] The present application relates to the medical technical field, more particularly, to a thyroid cancer lateral neck lymph node metastasis prediction model construction method and application. BACKGROUND

[0002] Papillary thyroid carcinoma (PTC) is the most common malignant tumor of the thyroid, although the overall prognosis is good, but part of the patients will appear neck lymph node metastasis, and then affect the surgical decision and patient prognosis.

[0003] Previous studies have shown that PTC lymph node metastasis is closely related to a variety of clinical pathological features. Although preoperative ultrasound technology has made some progress in identifying lymph node metastasis, its sensitivity is still limited.

[0004] In addition, the existing prediction model has not effectively distinguished different regions of neck lymph node metastasis, especially lacking a prediction tool specifically for lateral neck metastasis, which to some extent limits its clinical application value, and there is an urgent need for a specific prediction model and method for lateral neck region that can fuse multi-modal data. SUMMARY

[0005] The technical problem solved by the present application is to provide a thyroid cancer lateral neck lymph node metastasis prediction model construction method and an application of a thyroid cancer lateral neck lymph node metastasis prediction model to solve the above-mentioned defects of the prior art.

[0006] The technical solution adopted by the present application to solve its technical problem is:

[0007] A thyroid cancer lateral neck lymph node metastasis prediction model construction method is constructed, wherein the prediction model comprises a Logistic regression model and a decision flow module;

[0008] The Logistic regression model is used to calculate the actual risk value based on the value of the maximum diameter of the primary tumor, the value of the tumor boundary, the value of the BRAFV600E mutation state, the value of the presence of microcalcification, the value of the aspect ratio, and the value of the single / bilateral cancer focus;

[0009] The decision flow module is used to give a decision suggestion according to the calculated actual risk value and the set decision flow.

[0010] The thyroid cancer lateral neck lymph node metastasis prediction model construction method disclosed by the present application, wherein the decision flow comprises:

[0011] determining whether the actual risk value is greater than the set risk threshold;

[0012] If the actual risk value is greater than the set risk threshold, the sensitivity is calculated and the missed diagnosis risk is evaluated, if the missed diagnosis risk is in the set acceptable range, whether to perform lateral neck lymph node biopsy or cleaning can be decided according to the current guidelines or guideline recommended standard; if the missed diagnosis risk is in the set unacceptable range, it is recommended to strengthen the intervention, and actively perform suspicious lymph node biopsy to avoid missed diagnosis of potential metastatic lesions;

[0013] If the actual risk value is not greater than the set risk threshold, the specificity is used to evaluate the misdiagnosis risk, if the misdiagnosis risk is in the set acceptable range, the model prediction result can be used as supplementary information for preoperative ultrasound and puncture biopsy to guide whether to perform lateral neck lymph node cleaning or biopsy; if the misdiagnosis risk is in the set unacceptable range, it is recommended to take a more cautious strategy, to choose clinical observation, or to consider further intervention only when the image and clinical manifestations are highly suspicious.

[0014] The thyroid cancer lateral neck lymph node metastasis prediction model construction method provided by the application, wherein the actual risk value is calculated based on the value of the maximum diameter of the primary tumor, the value of the tumor boundary, the value of the BRAF V600E mutation state, the value of the microcalcification, the value of the aspect ratio and the value of the single / double cancer focus based on the formula:

[0015] Logit (P) =-5.0589 + 0.1672 x D-0.8839 x boundary + 1.3924 x BRAF + 1.0407 x Ca + 1.0633 x aspect ratio-1.6501 x Unilateral;

[0016] Wherein:

[0017] D is the value of the maximum diameter of the primary tumor, D is a continuous variable;

[0018] Boundary is the value of the tumor boundary, and is valued as 0 or 1, boundary is valued as 0, indicating that it is not clear, and boundary is valued as 1, indicating that it is not clear;

[0019] BRAF is the value of the BRAF V600E mutation state, and is valued as 0 or 1, BRAF is valued as 0, indicating wild type, and BRAF is valued as 1, indicating mutant type;

[0020] Ca is the value of the microcalcification, and is valued as 0 or 1, Ca is valued as 0, indicating that it is not detected, and Ca is valued as 1, indicating that it exists;

[0021] Aspect ratio is the value of the aspect ratio;

[0022] Unilateral is the value of the single / double cancer focus, and is valued as 0 or 1, Unilateral is valued as 0, indicating bilateral, and Unilateral is valued as 1, indicating unilateral.

[0023] The thyroid cancer lateral neck lymph node metastasis prediction model construction method, wherein the risk threshold is set to 0.14.

[0024] The application of a thyroid cancer lateral neck lymph node metastasis prediction model, according to the thyroid cancer lateral neck lymph node metastasis prediction model construction method described above, wherein the thyroid cancer lateral neck lymph node metastasis prediction model construction method is applied to the prediction of the risk of thyroid cancer lateral neck lymph node metastasis.

[0025] The thyroid cancer lateral neck lymph node metastasis prediction model construction method has the advantages that the model provides a new quantitative tool for preoperative evaluation of thyroid papillary carcinoma patients, can provide individualized and precise treatment plans for the clinic, reduces unnecessary treatment and missed diagnosis, and ultimately improves the treatment effect and quality of life of patients. BRIEF DESCRIPTION OF DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the present application will be further described below with reference to the drawings and embodiments. The drawings described below are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor:

[0027] Figure 1 is a principle block diagram of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0028] Figure 2 is a decision flowchart of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0029] Figure 3 is a research population baseline characteristic table of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0030] Figure 4 is a multivariate logistic regression analysis table of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0031] Figure 5 is a variable diagnostic performance table of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0032] Figure 6 is a multivariate logistic regression analysis table of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0033] Figure 7is a quantitative evaluation table of calibration performance of the prediction model of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0034] Figure 8 is a sensitivity, specificity and clinical decision table of different decision thresholds of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0035] Figure 9 is a patient inclusion and exclusion flowchart of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0036] Figure 10 is a forest plot of the association of BRAF mutation, tumor diameter, gender and calcification and CLNM of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0037] Figure 11 is a receiver operating characteristic (ROC) curve of the joint prediction model constructed using multivariate Logistic regression of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0038] Figure 12 is a ROC curve of the PTC patient lateral neck lymph node metastasis prediction model of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0039] Figure 13 is a calibration curve of the PTC patient lateral neck lymph node metastasis prediction model of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0040] Figure 14 is a decision curve of the PTC patient lateral neck lymph node metastasis prediction model of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application;

[0041] Figure 15 is a sensitivity and specificity curve at different thresholds of the thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application. DETAILED DESCRIPTION

[0042] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will be a clear and complete description of the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the protection scope of the present application.

[0043] The thyroid cancer lateral neck lymph node metastasis prediction model construction method of the preferred embodiment of the present application, as shown in Figure 1 , while referring to Figures 2-15 , includes a Logistic regression model 100 and a decision flow module 101;

[0044] The Logistic regression model 100 is used to calculate the actual risk value P based on the value of the primary tumor maximum diameter, the value of the tumor boundary, the value of the BRAFV600E mutation status, the value of the microcalcification presence, the value of the aspect ratio, and the value of the unilateral / bilateral cancer focus;

[0045] The calculation is based on the formula:

[0046] Logit(P) = -5.0589 + 0.1672 x D - 0.8839 x boundary + 1.3924 x BRAF + 1.0407 x Ca + 1.0633 x aspect ratio - 1.6501 x Unilateral;

[0047] Wherein the variable definitions are shown in the following table:

[0048]

[0049] Intercept = -5.0589;

[0050] D: The parameter value is based on the original measurement unit (mm) of preoperative ultrasound, without standardization processing. The regression coefficient of D 0.1672 is estimated in centimeters (mm) as the measurement unit, that is, for every 1mm increase in the maximum diameter of the tumor, the odds ratio (OR) of lateral neck lymph node metastasis increases by about 18.20% (OR = exp(0.1672) ≈ 1.18199);

[0051] boundary: The determination result of the tumor boundary definition is independently evaluated by two ultrasound physicians with the title of deputy chief physician and above, and the results are checked. If the results cannot be determined, or there is a difference in the determination results of the two physicians, the case data is not included in the scope of application of the model. In the model, the regression coefficient of the boundary variable is -0.8839, that is, under the condition that other conditions are the same, compared with boundary = 0, the odds ratio (OR) of boundary = 1 for lateral neck lymph node metastasis is about 58.7% lower (OR = exp(-0.8839) ≈ 0.413);

[0052] BRAF: If there is a disagreement between the preoperative needle biopsy and postoperative pathological examination, the case is not included in the scope of application of the model. In the model, the regression coefficient of the BRAF variable is 1.3924, that is, under the same conditions, when BRAF=1, the odds ratio (OR) of lateral neck lymph node metastasis increases by about 302% (OR=exp(1.3924)≈4.02);

[0053] Ca: This binary variable is defined as whether there is microcalcification. Microcalcification refers to the presence of fine hyperechoic spots less than 2 mm in diameter within or around the thyroid tumor on ultrasound examination. The regression coefficient of this variable in the model is 1.0407, that is, under the same conditions, when Ca=1, the odds ratio (OR) of lateral neck lymph node metastasis increases by about 183% (OR=exp(1.0407)≈2.83);

[0054] aspectratio: A binary variable defined as the aspect ratio, i.e., the ratio of the anteroposterior diameter to the lateral diameter of the thyroid nodule. When the aspect ratio is close to 1, it is assigned to the "less than 1" group and coded as 0. If the patient has thyroid nodules on both sides, the aspect ratio of each nodule must be determined separately, and the final determination is made: if the aspect ratio of any nodule is greater than 1, it is coded as 1, otherwise it is coded as 0. In the model, the regression coefficient of this variable is 1.0633, that is, under the same conditions, when aspectratio=1, the odds ratio (OR) of lateral neck lymph node metastasis increases by about 189.6% (OR=exp(1.0633)≈2.90);

[0055] Unilateral: A binary variable defined as the unilateral nature of the tumor, i.e., coded as 1 when the tumor is located only on one side of the thyroid, and coded as 0 when there are tumors on both sides of the thyroid. In the model, the regression coefficient of this variable is -1.6501, that is, under the same conditions, when Unilateral=1, the odds ratio (OR) of lateral neck lymph node metastasis decreases by about 80.8% (OR=exp(-1.6501)≈0.192);

[0056] The decision flow module 101 is used to give a decision suggestion according to the calculated actual risk value P and the set decision flow;

[0057] As shown in Figure 2 , the decision flow includes:

[0058] determining whether the actual risk value is greater than the set risk threshold;

[0059] If the actual risk value is greater than the set risk threshold, the sensitivity is calculated and the risk of missed diagnosis is assessed. If the risk of missed diagnosis is within the set acceptable range, the decision on whether to perform lateral cervical lymph node biopsy or dissection can be made based on the current guidelines or guidelines. If the risk of missed diagnosis is within the set unacceptable range, it is recommended to strengthen intervention and actively perform suspicious lymph node biopsy to avoid missing potential metastatic lesions.

[0060] If the actual risk value is not greater than the set risk threshold, the risk of misdiagnosis is assessed based on specificity. If the risk of misdiagnosis is within the set acceptable range, the model prediction results can be used as supplementary information for preoperative ultrasound and puncture biopsy to guide whether to perform lateral cervical lymph node dissection or biopsy. If the risk of misdiagnosis is within the set unacceptable range, it is recommended to adopt a more cautious strategy, choose clinical observation, or only consider further intervention when imaging and clinical manifestations are highly suspicious.

[0061] More specifically: Within different predicted probability ranges, this model can provide targeted auxiliary judgments for specific clinical scenarios. The model calculates the actual risk value P for each individual patient and combines it with… Figure 1 The decision-making process illustrated can assist in comprehensive clinical decision-making. When the risk value P > 0.14, sensitivity should be further calculated and the risk of missed diagnosis assessed. If the risk of missed diagnosis is within an acceptable range, the decision to perform lateral neck lymph node biopsy or dissection can be made based on current guidelines or guideline recommendations. "Acceptable" refers not only to the clinician's subjective assessment but also to the patient's informed acceptance of the predicted risk value, probability of missed diagnosis, and potential consequences. If the risk is unacceptable, it is recommended to strengthen intervention and actively perform suspicious lymph node biopsies to avoid missing potential metastatic lesions. Conversely, when the risk value P ≤ 0.14, the risk of misdiagnosis should be assessed based on specificity. If the risk of misdiagnosis is within an acceptable range, the model prediction results can serve as supplementary information to preoperative ultrasound and puncture biopsy to guide whether to perform lateral neck lymph node dissection or biopsy. If the risk of misdiagnosis is considered unacceptable, a more cautious strategy is recommended, prioritizing clinical observation or considering further intervention only when imaging and clinical manifestations are highly suspicious.

[0062] The model proposed in this application provides a new quantitative tool for the preoperative assessment of patients with papillary thyroid carcinoma, enabling clinicians to provide individualized and precise treatment plans, reduce unnecessary treatment and missed diagnoses, and ultimately improve patients' treatment outcomes and quality of life.

[0063] An application of a thyroid cancer lateral cervical lymph node metastasis prediction model, based on the above-mentioned thyroid cancer lateral cervical lymph node metastasis prediction model construction method, wherein the thyroid cancer lateral cervical lymph node metastasis prediction model construction method is applied to the prediction of the risk of thyroid cancer lateral cervical lymph node metastasis.

[0064] The principle data of this application are explained as follows:

[0065] This single-center retrospective study included 450 newly diagnosed papillary thyroid carcinoma patients who underwent surgery at our hospital between January 2022 and December 2024. Information was collected on their clinical characteristics, BRAF gene V600E mutation status, preoperative ultrasound results, and surgical plans to clarify the status of lymph node metastasis in the central and lateral cervical regions. Univariate analysis was used to screen variables related to lymph node metastasis, and a multivariate binary logistic regression model was further incorporated to construct a predictive tool. ROC analysis, calibration curve analysis, and decision curve analysis were used to evaluate its predictive performance and clinical application value.

[0066] The results showed that correlation analysis of central lymph node metastasis indicated statistically significant associations with tumor maximum diameter, BRAF gene V600E mutation, calcification, and gender (male). In the analysis of lateral neck metastasis, tumor maximum diameter, BRAF gene V600E mutation, calcification, aspect ratio, lesion location (unilateral / bilateral), and tumor boundary clarity were closely related to metastasis. In the lateral neck lymph node metastasis prediction model constructed based on these variables, tumor maximum diameter, BRAF gene V600E mutation, calcification, and aspect ratio >1 were positive predictors, while unilateral lesions and clear boundaries were negative predictors. The model showed good fit, with an area under the curve of 0.863, indicating good discriminant ability. Calibration curves and error indices suggested a high degree of consistency between the model's predictions and actual conditions. Decision curve analysis further demonstrated that the model has certain reference value under different actual clinical thresholds.

[0067] Conclusion: This study confirms that lymph node metastasis in patients with PTC is associated with multiple clinicopathological features, and exhibits different association characteristics in the central and lateral cervical regions. Based on this, a statistical model for predicting lateral cervical lymph node metastasis was constructed, aiming to provide auxiliary reference for individualized surgical decisions regarding lateral cervical lymph node dissection.

[0068] Study Design: This was a retrospective, single-center study approved by the Ethics Committee of Shenzhen Second People's Hospital. Patients were informed that their pathology results could be used for study purposes and gave their consent. Enrolled patients were divided into corresponding groups based on the presence of lymph node metastasis in the central and lateral cervical regions.

[0069] Patient selection: The subjects of this study were newly diagnosed patients with papillary thyroid carcinoma who underwent thyroid surgery in the Department of Thyroid and Breast Surgery of our hospital between January 1, 2022 and December 31, 2024 (3,380 patients were initially included in the study, and after screening according to the inclusion and exclusion criteria, 450 patients met the study criteria and were included in the analysis). The inclusion criteria for this study were as follows: (1) aged 18 to 70 years, newly diagnosed with PTC and completed surgical treatment according to the 2025 "Guidelines for Integrated Diagnosis and Treatment of Cancer in China - Thyroid Cancer Section"; (2) clear scope and standardized operation of central and / or lateral cervical lymph node dissection; (3) complete preoperative ultrasound data and postoperative pathological results, and reliable BRAF gene V600E mutation detection results. The exclusion criteria are as follows: (1) History of serious diseases of other important organs or malignant tumors at the time of initial diagnosis, or family history of PTC (first-degree relatives); (2) Multiple primary cancer lesions (referring to the number of cancer lesions >2); (3) Poor quality or missing imaging data, affecting the judgment; (4) No clear lymphoid tissue components were detected in the postoperative lymph node specimen; (5) The results of BRAF gene V600E mutation detection in the preoperative puncture and postoperative pathology were inconsistent; (6) The actual treatment plan received by the patient was inconsistent with the guideline recommendations; (7) Missing clinical data, etc.

[0070] Histopathological data collection: This was conducted jointly by two or more pathologists with over 10 years of experience in thyroid disease pathology. All thyroid cancer tissue samples were examined by them to confirm the PTC diagnosis and determine the BRAF gene V600E mutation status. The detection method used was the BRAF gene V600E mutation detection kit provided by Amoy Diagnostics.

[0071] Clinical data collection: Basic information such as the patient's gender and age was obtained from the hospital's electronic medical record system. The surgical plans for each patient were evaluated by two thyroid surgeons with associate senior professional titles or above, based on the 2025 "Chinese Guidelines for Integrative Diagnosis and Treatment of Cancer—Thyroid Cancer Section," and the decisions were made in conjunction with the patient's specific condition. If the recommended surgical plan did not match the patient's actual situation, the case was excluded to ensure that all enrolled patients received treatment according to the guideline-recommended surgical plan.

[0072] Ultrasound data acquisition: All ultrasound examination results were reviewed by two ultrasound physicians with associate senior professional titles or above to ensure accuracy and consistency. Key indicators determined included: maximum tumor diameter (preoperative ultrasound measurement), lesion distribution (unilateral / bilateral), boundary characteristics (clear / indistinct), presence of calcifications, aspect ratio, and presence of comorbidities.

[0073] Statistical Analysis: Statistical analyses were performed on central and lateral cervical lymph node metastases. The Mann-Whitney U test was used for continuous variables, and the chi-square test for categorical variables for univariate analysis. Statistically significant variables were then included in a binary logistic regression model for multivariate analysis. The goodness of fit of the model was assessed using the Hosmer-Lemeshow test. Diagnostic efficacy was evaluated using receiver operating characteristic (ROC) curve analysis and area under the curve (AUC) comparison. Model calibration curves and decision curve analysis were plotted to determine clinical applicability. The coefficients of significant variables in the multivariate logistic regression model were extracted using the `coef`, `cat`, and `sprintf` functions in R. A predictive formula for lateral cervical lymph node metastasis was constructed, and sensitivity and specificity curves as a function of the prediction threshold were plotted.

[0074] Statistical analysis was performed using R (v 4.5.0) software, and image plotting was performed using R (v 4.5.0) and GraphPad Prism (v 10.4.2) software.

[0075] result:

[0076] Baseline data of study subjects

[0077] The baseline characteristics of patients in each group are as follows: In the non-central lymph node metastasis group (n=209), the median age was 40 years (interquartile range: 34-51 years), with 32 male patients (15.3%); in the central lymph node metastasis group (n=241), the median age was 36 years (interquartile range: 30-45 years), with 80 male patients (33.1%); in the non-lateral cervical lymph node metastasis group (n=389), the median age was 38 years (interquartile range: 32-49 years), with 88 male patients (22.6%); and in the lateral cervical lymph node metastasis group (n=61), the median age was 35 years (interquartile range: 32-40 years), with 24 male patients (39.3%) (see [reference]). Figure 3 ).

[0078] Analysis of the association between factors such as BRAF gene V600E mutation and central lymph node metastasis in papillary thyroid carcinoma:

[0079] Univariate analysis of factors related to central lymph node metastasis: In the comparison between the central lymph node metastasis (CLNM) group and the non-CLNM group, the Kolmogorov-Smirnov test showed that neither age nor maximum tumor diameter followed a normal distribution; therefore, the Mann-Whitney U test was used for analysis. The results showed that patients in the CLNM group were significantly younger than those in the non-CLNM group (median: 36 years vs 40 years, P<0.001), and had larger maximum tumor diameters (median: 11 mm vs 7 mm, P<0.001). These two continuous variables were statistically associated with central lymph node metastasis. Regarding categorical variables, the chi-square test results showed that the CLNM group had a higher proportion of males (33.1% vs 15.3%, P<0.001), more common bilateral lesions (18.7% vs 6.7%, P<0.001), and significantly higher rates of BRAF gene V600E mutation (88.8% vs 71.3%, P<0.001) and calcification (74.3% vs 49.8%, P<0.001) compared to the non-CLNM group. However, there were no significant differences between the two groups in tumor boundary clarity (P=0.191), the proportion of tumors with an aspect ratio >1 (P=0.845), and the presence of thyroiditis (P=0.340) (see [reference]). Figure 3 ).

[0080] Multivariate Logistic Regression Analysis of Central Region Lymph Node Metastasis: Based on the results of univariate analysis, two continuous variables (age and tumor diameter) and four categorical variables (sex, lesion location, BRAF mutation, and calcification) significantly associated with central region lymph node metastasis (CLNM) were included in a multivariate logistic regression model to further explore their relationship with CLNM. The regression analysis results showed that sex (male) (P = 0.006), tumor diameter (larger) (P < 0.001), BRAF gene V600E mutation (mutant) (P < 0.001), and calcification (presence) (P < 0.001) were all significantly associated with central region lymph node metastasis (see [reference]). Figure 4 Forest plots were drawn to more visually demonstrate the predictive value of these variables for central lymph node metastasis (see [reference]). Figure 10The results showed that although age (P = 0.027) and tumor location (P = 0.011) were statistically different in the multivariate analysis, their odds ratios (ORs) were close to 1, indicating that these two variables had a relatively small impact on central lymph node metastasis. Further ROC curve analysis indicated that these two variables contributed limitedly to the final model and were therefore not included in the final predictive model. Finally, the four-factor binary logistic regression predictive model passed the goodness-of-fit test (Hosmer–Lemeshow test), with P = 0.260 (P > 0.05), indicating that the model fit was good and could effectively predict the risk of central lymph node metastasis.

[0081] Diagnostic efficacy assessment of each variable and construction of a joint predictive model: ROC analysis was performed on the four variables included in the logistic regression analysis (see [reference]). Figure 5 The results showed that tumor diameter had relatively high diagnostic efficacy, with an area under the curve (AUC) of 0.700 (95% CI: 0.625–0.748), while calcification had an AUC of 0.623 (95% CI: 0.570–0.675). Sex (AUC = 0.589) and BRAF gene V600E mutation (AUC = 0.588) had relatively low diagnostic efficacy. All variables had p-values ​​< 0.01, indicating statistical significance. These results are consistent with multivariate logistic regression analysis, suggesting that these variables can serve as candidate factors for constructing a predictive model. Based on these four variables, we further established a joint predictive model and plotted the ROC curve (see [reference]). Figure 11 The model's AUC was 0.758 (95% CI: 0.701–0.815), indicating a moderate level of predictive ability.

[0082] Although the central lymph node metastasis prediction model developed in this study still has room for further optimization, considering that missed central lymph node metastases significantly impact patient prognosis, and current clinical guidelines have relatively strict requirements for exemptions from central lymph node dissection, most patients still require central lymph node dissection in actual clinical practice. Therefore, the next focus will be on developing a prediction model for lateral neck lymph node metastasis, aiming to provide stronger supplementary evidence for clinical decisions regarding lateral neck lymph node dissection.

[0083] Association analysis and predictive model construction and evaluation of BRAF gene V600E mutation and other factors with lateral cervical lymph node metastasis in papillary thyroid carcinoma:

[0084] Univariate analysis of factors related to lateral cervical lymph node metastasis: In comparing patients with and without lateral cervical lymph node metastasis, the Kolmogorov-Smirnov test also revealed that age and maximum tumor diameter did not follow a normal distribution. The Mann-Whitney U test showed that the age of patients in the LC-LNM group was significantly lower than that in the Non-LC-LNM group (median: 35 years vs. 38 years, P = 0.015), and the maximum tumor diameter was significantly greater in the LC-LNM group (median: 15 mm vs. 8 mm, P < 0.001). In the categorical variable analysis, the chi-square test results showed that the LC-LNM group had a higher proportion of males (39.3% vs 22.6%, P = 0.007), a lower proportion of tumors with clear boundaries (19.7% vs 34.7%, P = 0.019), and significantly higher rates of BRAF gene V600E mutation (91.8% vs 78.9%, P < 0.001) and calcification (82.0% vs 59.9%, P < 0.001). Furthermore, the proportion of patients with an aspect ratio > 1 was significantly higher (77.0% vs 60.2%, P = 0.011), while the incidence of bilateral lesions was significantly increased in the LC-LNM group (36.1% vs 9.5%, P < 0.001). However, the proportion of patients with concurrent thyroiditis did not differ statistically between the two groups (P = 0.354). In summary, age, tumor diameter, and categorical variables significantly associated with LC-LNM (gender, tumor border, BRAF mutation, calcification, aspect ratio, and lesion location) were included in subsequent multivariate logistic regression analysis to further explore their statistical association with LC-LNM (see [reference]). Figure 3 ).

[0085] Multivariate logistic regression analysis of lateral cervical lymph node metastasis: Eight relevant variables selected from univariate analysis were included in a binary logistic regression model to further identify factors significantly associated with lateral cervical lymph node metastasis in PTC patients. The regression analysis results showed that the maximum tumor diameter (OR = 3.103, 95% CI: 1.983–4.858, P < 0.001) and tumor location (unilateral / bilateral) (OR = 0.202, 95% CI: 0.099–0.413, P < 0.001) had high predictive value in the model; BRAF gene V600E mutation (OR = 3.397, 95% CI: 1.187) was also observed. –9.728, P=0.023), calcification (OR=2.668, 95% CI: 1.197–5.948, P=0.016), aspect ratio (OR=2.833, 95% CI: 1.344–5.972, P=0.006), and tumor margin (OR=0.452, 95% CI: 0.206–0.993, P=0.048) all showed some predictive ability. Gender (P=0.176) and age (P=0.208) did not reach statistical significance in this model (see [reference]). Figure 6 Therefore, it was excluded from the prediction model. The Hosmer-Lemeshow goodness-of-fit test showed that the p-value of the six-factor binary logistic regression model was 0.526 (p>0.05), indicating a good model fit.

[0086] Performance evaluation and clinical application value analysis of the lateral cervical lymph node metastasis prediction model: To evaluate the discriminative efficacy of the prediction model based on binary logistic regression, receiver operating characteristic (ROC) curve analysis was first performed (see [reference]). Figure 12 The results show that the area under the curve for this model is 0.863. Further evaluation of the model's prediction accuracy using calibration curves reveals that the model's predicted values ​​are generally close to the diagonal dashed line (the ideal reference line for y=x), indicating a high degree of consistency between the model's predicted probabilities and the actual observed frequencies (see [reference]). Figure 14 ).

[0087] The mean absolute error (MAE), mean squared error (MSE), and 90th percentile of absolute error (0.9 quantile) of the model are calculated to quantify the model's calibration performance. See also Figure 7The model's MAE was 0.016, MSE was less than 0.001, and the absolute error of 90% of the predicted values ​​did not exceed 0.055. The decision curves of the multi-factor prediction model constructed in this study show that, within the threshold probability range of 0.05-0.65, its net benefit was significantly higher than the reference strategy, indicating that the model has good clinical decision support value. (See reference) Figure 15 .

[0088] Establishment of a predictive model for lateral cervical lymph node metastasis and determination of the optimal diagnostic threshold: The following predictive formula for lateral cervical lymph node metastasis was constructed using R language for clinical application: logit(p)=0.167×Diameter+1.041×Calcification+1.392×BRAF mutation+1.063×Aspect ratio+(-1.650×Tumorlesions)+(-0.884×Tumor boundary)-5.059. The prediction probability P can be calculated using the logit function. Based on receiver operating characteristic (ROC) analysis, the Youden Index was calculated to be 0.14. The sensitivity and specificity corresponding to the optimal prediction threshold (0.14) were extracted, and analytical curves showing the change of sensitivity and specificity with the prediction threshold were plotted. The black dashed line in the figure (x = 0.14) represents the optimal threshold line. At the intersection point, the sensitivity is 0.80 and the specificity is 0.81. At this point, the false negative rate (1 - sensitivity) is 20%, and the false positive rate (1 - specificity) is 19% (see [reference]). Figure 2 ).

[0089] This study aimed to explore the clinicopathological factors of lymph node metastasis in papillary thyroid carcinoma and to construct a predictive model for lateral cervical lymph node metastasis. In the central region metastasis analysis, the maximum tumor diameter, BRAF gene V600E mutation, calcification, and male sex were identified as significant correlates (all P < 0.001). In the lateral cervical metastasis analysis, in addition to the above factors, an aspect ratio > 1 (positive predictor, P = 0.006) and clear unilateral lesion boundaries (negative predictors, P < 0.001 and P = 0.048, respectively) were proven to be significant correlates of lateral cervical lymph node metastasis. The LC-LNM binary logistic regression predictive model established based on the aforementioned six variables showed good fit according to the Hosmer-Lemeshow test (P = 0.526), ​​and its discriminant power was assessed by receiver operating characteristic (ROC) curve analysis, with an area under the curve (AUC) of 0.863. To further evaluate the accuracy of the predictive model, we used the `rms` package in R to plot a calibration curve. The horizontal axis of the curve represents the predicted transition probability, and the vertical axis represents the actual observed transition frequency. The diagonal dashed line (y=x) in the figure represents the ideal reference line, indicating that the predicted value is completely consistent with the actual result; the black solid line is the calibration curve after bias correction, and the thin dashed line represents the one-sided confidence interval. The calibration curve shows that the model's predicted value is largely consistent with the actual observed value. Subsequently, decision curve analysis was used to evaluate the clinical application value of the predictive model. The horizontal axis in the figure represents the decision threshold probability, i.e., the lowest risk probability for initiating clinical intervention; the vertical axis is the net benefit, calculated as: the proportion of true positive cases - the proportion of false positive cases × [threshold probability / (1 - threshold probability)]. The two reference lines in the figure represent: the black horizontal line represents the "intervene all patients" strategy (simulating overtreatment), and the gray diagonal line represents the "do not intervene in any patients" strategy (simulating missed diagnosis). When the threshold probability is between 0.05 and 0.65, the model curve consistently remains above the "All" and "None" lines, indicating that within this risk threshold range, using the model for lateral neck lymph node dissection decisions yields a higher net benefit. However, it is worth noting that when the threshold probability is below 0.05, the model may lead to unnecessary over-dissection; while when the threshold is above 0.65, it may miss some patients requiring intervention. Therefore, this model has good clinical reference value within the moderate risk threshold range.

[0090] Based on the above model, we further constructed a specific prediction formula to achieve a quantitative assessment of the risk of lateral cervical lymph node metastasis. Using the coefficient extraction functions (coef, cat, and sprintf) in R language, we extracted the coefficients of each significant variable in the multivariate logistic regression analysis, and finally established the following logit model: logit(p)=0.167×Diameter+1.041×Calcification+1.392×BRAF mutation+1.063×Aspect ratio+(-1.650×Tumor lesions)+(-0.884×Tumor boundary)-5.059;

[0091] Where logit(p) represents the logarithmic probability of lateral cervical lymph node metastasis. This is calculated using the inverse formula. The predictive probability P for each patient can be obtained. This formula comprehensively considers four positive predictors (tumor diameter, calcification, BRAF mutation, aspect ratio >1) and two negative predictors (unilateral lesion, well-defined borders), providing a quantitative basis for preoperative risk assessment.

[0092] ROC curve analysis yielded an optimal prediction threshold of 0.14 for the model, corresponding to a sensitivity of 0.80, a specificity of 0.81, and a false negative rate of 20% and 19%, respectively. This threshold represents a good balance between sensitivity and specificity, indicating that the model possesses good overall discriminative ability at this probability level and is suitable for standardized treatment decisions based on guidelines. In the curves showing the change in sensitivity and specificity with the prediction threshold, the optimal threshold is indicated by a dashed line at x = 0.14. The intersection of this line with the two performance curves shows a good balance between sensitivity and specificity, indicating that the model has high overall discriminative ability at this threshold. At the optimal threshold level, the model's sensitivity and specificity are balanced, making it suitable for standardized diagnostic and treatment decisions based on the latest clinical guidelines. The region to the left of the threshold exhibits higher sensitivity, suitable for clinical scenarios aimed at screening, helping to reduce the risk of false negatives; the region to the right of the threshold exhibits higher specificity, suitable for the diagnostic stage, reducing the false negative rate and thus improving the accuracy of intervention decisions (see [reference]). Figure 8 ).

[0093] Within different predicted probability ranges, this model can also provide targeted auxiliary judgments for specific clinical scenarios. By calculating the actual risk value P for each patient using the model and combining it with the decision-making process, it can assist in the formulation of comprehensive clinical decisions (see [reference]). Figure 2When the risk value P > 0.14, sensitivity should be further calculated and the risk of missed diagnosis assessed. If the risk of missed diagnosis is within an acceptable range, the decision to perform lateral neck lymph node biopsy or dissection can be made based on current guidelines or Southern Pathway recommendations; "acceptable" refers not only to the clinician's subjective assessment but also to the patient's informed acceptance of the predicted risk value, probability of missed diagnosis, and potential consequences. If the risk is unacceptable, stronger intervention and active biopsy of suspicious lymph nodes are recommended to avoid missing potential metastatic lesions. Conversely, when the risk value P ≤ 0.14, the risk of misdiagnosis should be assessed based on specificity. If the risk of misdiagnosis is within an acceptable range, the model prediction results can serve as supplementary information to preoperative ultrasound and puncture biopsy to guide the decision to perform lateral neck lymph node dissection or biopsy; if the risk of misdiagnosis is considered unacceptable, a more cautious strategy is recommended, prioritizing clinical observation or considering further intervention only when imaging and clinical presentation are highly suspicious.

[0094] In numerous similar studies, BRAF gene V600E mutations and various clinicopathological features play an important role in predicting lymph node metastasis in patients with papillary thyroid carcinoma (PTC). Previous studies have shown that BRAF mutations are significantly associated with lymph node metastasis in PTC and exhibit some interaction effects with clinical factors such as obesity, hypertension, maximum tumor diameter, and calcification. Zhang et al. further indicated that maximum tumor diameter, extra-glandular tumor invasion, and central lymph node metastasis are independent risk factors for lateral cervical lymph node metastasis, and that maximum tumor diameter has some predictive ability for lateral cervical lymph node metastasis (AUC = 0.639). In this study, we validated the importance of maximum tumor diameter in predicting lateral cervical lymph node metastasis. However, this study did not consider clinical variables such as hypertension and smoking history, and the presence or absence of central lymph node metastasis was not included in the lateral cervical lymph node metastasis model. Future research should further explore the combined effects of these factors to improve cervical lymph node metastasis risk assessment models for different regions.

[0095] Currently, contrast-enhanced ultrasound (HEE) has a significant advantage over conventional ultrasound in identifying suspicious lymph nodes, enabling sonographers to make more accurate diagnoses and differential diagnoses. The application of radiomics has also made significant progress in predicting lymph node metastasis in PTC (percutaneous transcranial neoplasia). Some researchers have used trimodal ultrasound images to extract radiomic features of the tumor and surrounding areas, and combined this with clinical variables to construct a highly efficient hybrid model, which showed good predictive efficacy (AUC 0.841-0.883). However, most of these studies did not provide detailed delineation of cervical lymph node metastasis areas, and the imaging identification of lateral cervical lymph node metastasis still requires further exploration.

[0096] The potential of prediction models combining deep learning and ultrasound features in predicting lymph node metastasis has been widely recognized. One study proposed an ensemble model that combined deep learning with ultrasound features such as lymph node pluripotency, location, and microcalcifications, demonstrating high predictive ability (AUC 0.812-0.829). This result indicates that combining deep learning with ultrasound imaging features has a significant advantage in improving predictive accuracy. Similarly, a study combining artificial intelligence technology with ultrasound image analysis showed high diagnostic accuracy, demonstrating its application prospects in clinical decision support. Future research could consider combining this method with other clinical variables to further improve the model's predictive accuracy.

[0097] For the treatment of lateral cervical lymph node metastasis, current clinical guidelines recommend therapeutic dissection, i.e., lateral cervical lymph node dissection is performed when preoperative assessment or intraoperative frozen section pathology confirms N1b stage. However, at present, there is still a lack of effective tools to accurately assess the risk of lateral cervical lymph node metastasis, and reliance on biopsy pathology may lead to false negative results, thus missing some high-risk patients, especially those with lateral cervical metastasis.

[0098] In current clinical practice, preoperative assessment of lateral cervical lymph node metastasis often relies on ultrasound imaging, biopsy pathology, and subjective judgment. When imaging is insufficient to clearly define the location, this judgment can lead to diagnostic errors. Even biopsy pathology can produce false negatives, especially when the lymph nodes are large or numerous. By combining factors such as BRAF gene V600E mutation, maximum tumor diameter, calcification, and aspect ratio, the model developed in this study can provide quantitative evidence for clinical decision-making, helping clinicians to be more objective and scientific in predicting metastasis risk, and reducing the risk of missed diagnoses or overtreatment.

[0099] This model has significant potential for expansion, especially when combined with cutting-edge technologies such as contrast-enhanced ultrasound, radiomics, and deep learning, which could further improve predictive accuracy and intelligence. Incorporating the latest artificial intelligence algorithms can help physicians make faster and more accurate diagnoses in more complex clinical settings. Furthermore, with the accumulation of multi-center, long-term follow-up data, the model is expected to be continuously optimized, thereby providing more universally applicable clinical decision support for thyroid cancer patients.

[0100] In conclusion, the model developed in this study provides a new quantitative tool for the preoperative assessment of patients with papillary thyroid carcinoma, enabling clinicians to provide individualized and precise treatment plans, reduce unnecessary treatments and missed diagnoses, and ultimately improve patients' treatment outcomes and quality of life.

[0101] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for constructing a predictive model for lateral cervical lymph node metastasis in thyroid cancer, characterized in that, The predictive model includes a Logistic regression model and a decision process module; The Logistic regression model is used to calculate the actual risk value based on the values ​​of the maximum diameter of the primary tumor, the tumor boundary, the BRAF V600E mutation status, the presence of microcalcifications, the aspect ratio, and the single / bilateral lesions. The decision-making process module is used to provide decision recommendations based on the calculated actual risk value and the established decision-making process.

2. The method for constructing a lateral cervical lymph node metastasis prediction model for thyroid cancer according to claim 1, characterized in that, The decision-making process includes: Determine whether the actual risk value exceeds the set risk threshold; If the actual risk value is greater than the set risk threshold, the sensitivity is calculated and the risk of missed diagnosis is assessed. If the risk of missed diagnosis is within the set acceptable range, the decision on whether to perform lateral cervical lymph node biopsy or dissection can be made based on the current guidelines or guidelines. If the risk of missed diagnosis is within the set unacceptable range, it is recommended to strengthen intervention and actively perform suspicious lymph node biopsy to avoid missing potential metastatic lesions. If the actual risk value is not greater than the set risk threshold, the risk of misdiagnosis is assessed based on specificity. If the risk of misdiagnosis is within the set acceptable range, the model prediction results can be used as supplementary information for preoperative ultrasound and puncture biopsy to guide whether to perform lateral cervical lymph node dissection or biopsy. If the risk of misdiagnosis is within the set unacceptable range, it is recommended to adopt a more cautious strategy, choose clinical observation, or only consider further intervention when imaging and clinical manifestations are highly suspicious.

3. The method for constructing a lateral cervical lymph node metastasis prediction model for thyroid cancer according to claim 2, characterized in that, The actual risk value is calculated based on the values ​​of the primary tumor's maximum diameter, tumor boundary, BRAF V600E mutation status, presence of microcalcifications, aspect ratio, and unilateral / bilateral lesions, using the following formula: Logit(P)=-5.0589+0.1672×D-0.8839×boundary+1.3924×BRAF+1.0407×Ca+1.0633×aspectratio-1.6501×Unilateral; in: D is the value of the maximum diameter of the primary tumor, and D is a continuous variable; The boundary value is 0 or 1. A boundary value of 0 indicates that the boundary is unclear, and a boundary value of 1 indicates that the boundary is unclear. BRAF is the value of the BRAF V600E mutation status, which is assigned a value of 0 or 1. A BRAF value of 0 indicates wild type, and a BRAF value of 1 indicates mutant type. Ca represents the value for the presence of microcalcifications, and is assigned a value of 0 or 1. A value of 0 for Ca indicates that it is not detected, and a value of 1 for Ca indicates that it is present. Aspectratio is the value of the aspect ratio; Unilateral is the value for unilateral or bilateral cancer lesions, assigned a value of 0 or 1. Unilateral is assigned a value of 0 to indicate bilateral, and Unilateral is assigned a value of 1 to indicate unilateral.

4. The method for constructing a lateral cervical lymph node metastasis prediction model for thyroid cancer according to claim 3, characterized in that, The set risk threshold is 0.

14.

5. An application of a thyroid cancer lateral cervical lymph node metastasis prediction model, comprising the method for constructing a thyroid cancer lateral cervical lymph node metastasis prediction model according to any one of claims 1-4, characterized in that, The proposed method for constructing a lateral cervical lymph node metastasis prediction model for thyroid cancer is applied to the prediction of the risk of lateral cervical lymph node metastasis in thyroid cancer.

Images