A Multimodal Machine Learning Model Integrating Ultrasound and Serological Biomarkers for Non-Invasive Prediction of Gallbladder Polyp Malignancy: Development, Validation, and Clinical Translation

A Multimodal Machine Learning Model Integrating Ultrasound and Serological Biomarkers for Non-Invasive Prediction of Gallbladder Polyp Malignancy: Development, Validation, and Clinical Translation | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A Multimodal Machine Learning Model Integrating Ultrasound and Serological Biomarkers for Non-Invasive Prediction of Gallbladder Polyp Malignancy: Development, Validation, and Clinical Translation Yang Yan, Tu Haibin, Lin Youguo, Wei Jianting This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6744318/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract Background: Differentiating benign from malignant gallbladder polyps (GBPs) is critical for clinical decisions. Pathological biopsy, the gold standard, requires cholecystectomy, underscoring the need for non-invasive alternatives. Methods: This retrospective study included 202 patients (50 malignant, 152 benign) who underwent cholecystectomy (2018–2024) at Fujian Provincial Hospital. Ultrasound features (polyp diameter, stalk presence), serological markers (neutrophil-to-lymphocyte ratio [NLR], CA19-9), and demographics were analyzed. Patients were split into training (70%) and validation (30%) sets. Ten machine learning (ML) algorithms were trained; the model with the highest area under the receiver operating characteristic curve (AUC) was selected. SHapley Additive exPlanations (SHAP) identified key predictors. Models were categorized as Clinical (ultrasound + age), Hematological (NLR + CA19-9), and Combined (all five variables). ROC, Precision-Recall (PR), calibration, and Decision Curve Analysis (DCA) curves were generated. A web-based calculator was developed. Results: The Extra Trees model achieved the highest AUC (0.97 in training, 0.93 in validation). SHAP analysis highlighted polyp diameter, sessile morphology, NLR, age, and CA19-9 as top predictors. The Combined Model outperformed Clinical (AUC 0.89) and Hematological (AUC 0.68) models, with balanced sensitivity (66–54%), specificity (94–93%), and accuracy (87–83%). Conclusion: This ML model integrating ultrasound and serological markers accurately predicts GBP malignancy. The web-based calculator facilitates clinical adoption, potentially reducing unnecessary surgeries. gallbladder cancer ultrasound hematological markers prediction machine learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Introduction Gallbladder polyps (GBPs) are frequently detected during abdominal ultrasonography, with a reported prevalence ranging from 0.3–12% in the general population [1, 2] . While the majority of GBPs are benign, a clinically significant proportion represents premalignant or malignant lesions [3] . Accurate differentiation between benign and malignant GBPs is paramount for guiding appropriate clinical management, as early detection and surgical resection of malignant lesions are essential for improving patient prognosis [4] . Ultrasound imaging plays a crucial role in the initial assessment of GBPs, providing vital information such as polyp size, number, morphology, and the presence of gallstones or gallbladder wall irregularities [5, 6] . Studies have indicated that specific ultrasound features, including polyp diameter exceeding 10 mm, sessile morphology, and single polyp presentation, are associated with an elevated risk of malignancy [7, 8] . However, the diagnostic accuracy of ultrasound alone remains limited, as overlapping features between benign and malignant polyps often lead to diagnostic ambiguity [9, 10] . Conversely, serological markers such as carbohydrate antigen 19 − 9 (CA19-9), carcinoembryonic antigen (CEA), and neutrophil-to-lymphocyte ratio (NLR) have been investigated as potential indicators of gallbladder malignancy [11–13] . Elevated levels of these markers have been correlated with malignant transformation [14] , yet their independent predictive value is insufficient for clinical decision-making. Despite the complementary nature of ultrasound and serological markers, limited research has integrated these modalities to enhance the prediction of GBP malignancy. Although both ultrasound features and serological markers offer valuable insights, their independent application has limitations. Notably, there is a scarcity of studies comprehensively integrating these two modalities for GBP malignancy prediction, representing a significant gap in the current diagnostic approach. The challenge lies in effectively combining these disparate data types to improve diagnostic accuracy. Machine learning (ML) provides a robust solution to this challenge [15] . ML algorithms are specifically designed to manage complex interactions and non-linear relationships among multiple variables, rendering them ideally suited for integrating diverse data sources [16] . In oncology, ML has demonstrated considerable success in enhancing diagnostic accuracy, predicting treatment response, and identifying prognostic factors across various cancer types. For instance, ML models incorporating radiomic features from CT scans have been utilized to predict lymph node metastasis in lung cancer [17] . In breast cancer, ML algorithms integrating genomic data and clinical parameters have improved the prediction of recurrence risk [18] . Furthermore, ML has been applied to differentiate benign from malignant liver lesions using a combination of imaging and serological data, achieving high diagnostic performance [19, 20] . Inspired by these advancements, we hypothesized that an ML model integrating both ultrasound features and serological biomarkers could significantly improve the non-invasive prediction of malignancy in GBPs. By leveraging the capacity of ML to analyze complex interactions between these data types, we aimed to develop a more accurate and reliable diagnostic tool. This study, therefore, sought to develop and validate such a model, potentially transforming the clinical management of GBPs by enabling more informed decision-making, reducing unnecessary cholecystectomies, and facilitating earlier detection of malignant lesions. Patients and Methods Study Design and Population This study employed a retrospective cohort design, reviewing data from patients who underwent cholecystectomy at Fujian Provincial People’s Hospital between January 1, 2018, and January 1, 2024. The study protocol was approved by the Institutional Review Board of Fujian Provincial People’s Hospital and conducted in accordance with the Declaration of Helsinki. Patients were included if they had undergone cholecystectomy and had a confirmed histopathological diagnosis of GBPs. Exclusion criteria were: (1) patients with a pre-operative diagnosis of gallbladder carcinoma; (2) patients who underwent cholecystectomy for reasons other than GBPs (e.g., cholecystitis, biliary dyskinesia without polyps); (3) patients with incomplete clinical data, defined as missing ultrasound reports, serological marker results, or histopathology reports; and (4) patients with a history of prior biliary tract surgery or malignancy. The patient inclusion flow was showed in Fig. 1 . Data Collection Data were retrospectively extracted from electronic medical records, radiology information systems (RIS) for ultrasound reports, and laboratory information systems (LIS) for serological marker results. The following variables were collected for each patient: Ultrasound Features All examinations were performed using high-resolution ultrasound systems (Mindray Pesona 7S/8S, Philips EPIQ5; C5-2 transducer) under standardized protocols. Two board-certified sonographers independently interpreted the findings through consensus-based evaluation. In cases of diagnostic discrepancy (observed in 3.8% of examinations), a third senior sonographer (with > 15 years’ experience) conducted blinded reassessment to achieve definitive interpretation. Finalized parameters were prospectively documented in radiology reports for subsequent analysis. Polyp Diameter (mm): Maximum diameter of the largest polyp, measured in millimeters. Polyp Number: Total number of polyps identified in the gallbladder, categorized as ‘single’ or ‘multiple’. Polyp Stalk: Presence or absence of a stalk, categorized as ‘pedunculated’ (presence of a stalk) or ‘sessile’ (absence of a stalk). In cases of pedunculated polyps, stalk thickness was not routinely measured and thus not included. Polyp Location: Anatomical location of the polyp within the gallbladder, categorized as ‘bottom’, ‘body’, or ‘neck’. Polyp Surface Characteristics: Described as ‘smooth’ or ‘coarse’ based on the ultrasound report. Echo Homogeneity: Polyp echogenicity was categorized as ‘homogeneous’ or ‘heterogeneous’. Hyperechoic Foci: Presence or absence of hyperechoic foci within the polyp, documented as ‘yes’ or ‘no’. Gallbladder Stones: Presence or absence of gallstones within the gallbladder, documented as ‘yes’ or ‘no’. Gallbladder Wall Roughness: Presence or absence of gallbladder wall roughness, described in reports as irregular or thickened gallbladder wall contour, documented as ‘yes’ or ‘no’. This was distinct from gallbladder wall thickness, which was not consistently reported and therefore not included. Serological Markers Pre-operative serum samples were collected within one week prior to cholecystectomy. The following hematological markers were recorded, with units of measurement specified Alanine Aminotransferase (ALT, U/L) Aspartate Aminotransferase (AST, U/L) Total Bilirubin (TBIL, µmol/L) Neutrophil-to-Lymphocyte Ratio (NLR): Calculated as absolute neutrophil count divided by absolute lymphocyte count. Carbohydrate Antigen 19 − 9 (CA19-9, U/mL) Carcinoembryonic Antigen (CEA, ng/mL) Alkaline Phosphatase (ALP, U/L) Adenosine Deaminase (ADA, IU/L) Demographic and Clinical Data The following demographic and clinical variables were collected Age (years): Age at the time of cholecystectomy. Sex: Categorized as ‘male’ or ‘female’. Body Mass Index (BMI, kg/m²): Calculated as weight in kilograms divided by height in meters squared. Waist-to-Hip Ratio: Measured at the time of admission, calculated as waist circumference divided by hip circumference, and categorized as ‘normal’ or ‘fatty’. Comorbidities: Presence or absence of hypertension, diabetes, hyperlipidemia, and viral hepatitis, documented as ‘yes’ or ‘no’. Alcohol History: History of alcohol consumption, documented as ‘yes’ or ‘no’. Data Preprocessing Prior to model development, data preprocessing steps were undertaken. Missing values, present in less than 5% of the dataset and primarily in CEA and ALP measurements, were imputed using median imputation, as these markers were not expected to be highly skewed. Outliers were assessed using boxplots and defined as values exceeding 1.5 times the interquartile range above the 75th percentile or below the 25th percentile. No outliers were removed as clinically implausible, but extreme values were winsorized to the 99th percentile to mitigate undue influence on model training. Continuous variables were standardized using z-score normalization to ensure features were on a comparable scale, improving the performance and convergence of certain ML algorithms. Machine Learning Model Development The dataset was randomly split into a training set (70%) and a validation set (30%) using a fixed random seed (seed = 123456) to ensure reproducibility. Ten ML algorithms were employed for model development using the scikit-learn library in Python (3.6.1). The algorithms included: Logistic Regression, Random Forest, Gradient Boosting, Support Vector Classifier (SVC) with a radial basis function kernel, Decision Tree, K-Nearest Neighbors, Naive Bayes (Gaussian Naive Bayes), LightGBM, Extra Trees, and AdaBoost. Hyperparameter tuning for each algorithm was performed using a grid search approach with 10-fold cross-validation on the training set to optimize model performance and prevent overfitting. Ten-fold cross-validation was implemented to robustly estimate the performance of each model on unseen data within the training set. In each fold, the training data was further divided into 9 folds for training and 1 fold for validation. The average Area Under the Receiver Operating Characteristic Curve (AUC) across the 10 folds was used to evaluate each algorithm’s performance. Model selection was based on the highest average AUC achieved during the 10-fold cross-validation in the training set. The Extra Trees algorithm demonstrated the highest AUC and was selected for further analysis and evaluation on the validation set. Feature importance analysis was conducted using SHapley Additive exPlanations (SHAP) values, SHAP values provide a unified measure of feature importance by quantifying the contribution of each feature to the prediction of individual instances. The top ten most important variables, ranked by mean absolute SHAP value, were visualized to understand their relative influence on the model’s predictions. To explore model parsimony, we iteratively built models with varying numbers of top-ranked features, starting with the single most important feature and incrementally adding features based on their SHAP ranking. The AUC was calculated for each model configuration using 10-fold cross-validation on the training set to identify the optimal number of variables that maximized performance while maintaining model simplicity. Based on the variables included, three model categories were defined: Clinical Model: Included polyp diameter, polyp stalk (sessile/pedunculated), and patient age – variables readily available from routine clinical assessment and basic ultrasound. Hematological Model: Included neutrophil-to-lymphocyte ratio (NLR) and carbohydrate antigen 19 − 9 (CA19-9) – serological markers with potential relevance to inflammation and malignancy. Combined Model: Included all collected variables, representing the integration of clinical, ultrasound, and hematological data. Model Evaluation The performance of the selected Extra Trees model and the three defined model categories (Clinical, Hematological, and Combined) were evaluated on the independent validation set. Performance metrics included: AUC, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy. Receiver Operating Characteristic (ROC) curves and Precision-Recall (PR) curves were plotted to visualize the trade-off between sensitivity and specificity, and precision and recall, respectively. Calibration curves were generated to assess the reliability of the predicted probabilities, using isotonic regression for calibration. Decision Curve Analysis (DCA) was performed to evaluate the clinical utility of the models by quantifying the net benefit across a range of clinically relevant risk thresholds. Web-Based Calculator Development To facilitate clinical translation, a user-friendly web-based calculator was developed using specify web framework. The calculator is hosted at https://gallbladder-kwljafh4ile9dlb9qu9y46.streamlit.app/ ”. The calculator allows clinicians to input patient clinical characteristics, ultrasound features, and serological marker values, and obtain a predicted probability of gallbladder polyp malignancy based on the Combined Model. The calculator interface is designed for ease of use and rapid risk assessment in a clinical setting. Statistical Analysis Statistical analyses were performed using Python (3.6.1) with the libraries scikit-learn, pandas, numpy, matplotlib, seaborn, and statsmodels. Continuous variables are presented as mean ± standard deviation (SD) or median (interquartile range [IQR]) depending on normality, assessed using the Shapiro-Wilk validation. Categorical variables are presented as frequencies and percentages. Differences between the training and validation sets for continuous variables were assessed using the independent samples t-validation or Mann-Whitney U validation as appropriate, and for categorical variables using the chi-square validation. A two-sided p-value < 0.05 was considered statistically significant. Results Basic Information of All Patients This study included a total of 202 patients, of whom 50 were diagnosed with gallbladder cancer. The cohort was divided into a training set (n = 142, including 35 with gallbladder cancer) and a validation set (n = 60, including 15 with gallbladder cancer). Of the entire cohort, 110 patients (54.5%) were male, and the mean age was 59.9 ± 15.89 years. Additional baseline characteristics are presented in Table 1. Crucially, the training and validation sets were well-balanced, with no statistically significant differences observed between the groups for any of the measured variables (all p > 0.05). The information in the training set is shown in Supplementary Table 1. Comparison of Predictive Performance of Different Machine Learning Models Table 2 presents a comprehensive comparison of the predictive performance of various ML models using multiple evaluation metrics, including mean AUC, 95% confidence intervals for AUC, mean accuracy, sensitivity, specificity, PPV, and NPV. Among the models evaluated, the Extra Trees classifier demonstrated the highest predictive performance, with a mean AUC of 0.97 (95% CI: 0.94–0.99), significantly outperforming other models. Additionally, the Extra Trees model achieved the highest mean accuracy (0.95), sensitivity (0.95), specificity (0.94), PPV (0.92), and NPV (0.92), indicating its robust ability to discriminate between positive and negative cases. Given its superior performance across all metrics, the Extra Trees model was selected for further analysis in this study. Selection of the Optimal Number of Variables After selecting the Extra Trees model as the predictive algorithm, we ranked the variables based on their importance and visualized the results using SHAP, as illustrated in Figure 2. The top ten variables identified were: diameter, stalk, NLR, age, CA19-9, AST, BMI, CEA, ALP, and TBIL. To adhere to the principle of model parsimony, we calculated the corresponding AUC values for models incorporating different numbers of variables, as shown inFigure 3. The results demonstrated that the AUC reached a high value when the model included five variables, and further inclusion of additional variables did not significantly improve the AUC. Therefore, for subsequent analyses, we constructed a fusion model incorporating the following five variables: diameter, stalk, NLR, age, and CA19-9. Model Classification and Comparative Analysis Based on the selected variables, we categorized diameter, stalk, and age as the ‘Clinical Model’. CA19-9 and NLR were classified as the ‘Hematological Model’. We then compared the predictive performance of these two models, along with a ‘Combined Model’ incorporating all five variables. ROC curves for each model are presented in Figure 4, and the PR curve is presented in Figure 5. Both the ROC and PR curves demonstrated the superior performance of the Combined model. AUC values and performance metrics are summarized in Table 3. The Combined Model consistently outperformed the other two models, achieving the highest AUC values (0.93 for both training and validation sets) and demonstrating balanced sensitivity, specificity, and accuracy. Calibration Curve Figure 6 displays the calibration curves for the training and validation cohorts. In both datasets, the Combined Model exhibits the best calibration, with its curve tracking closest to the ideal 45-degree diagonal, indicating good agreement between predicted probabilities and observed proportions. The Clinical Model shows a slight tendency towards over-calibration in the higher probability ranges, particularly in the validation set. The Hematological Model demonstrates the least satisfactory calibration, deviating more noticeably from the ideal line in both datasets. These calibration plots visually confirm the superior calibration of the Combined Model compared to the other two models.Table 4 presents the calibration metrics for all models. The Combined model demonstrated good calibration in both the training and validation sets. In the training set, the Combined model had a Brier score of 0.09, an HL p-value of 0.25, a calibration slope of 1.03, and a calibration intercept of <0.01. In the validation set, the Combined model maintained good calibration, with a Brier score of 0.08, an HL p-value of 0.32, a calibration slope of 0.95, and a calibration intercept of 0.11. These values indicate that the Combined model’s predicted probabilities are well-calibrated with the observed outcomes. The Clinical and Hematological models showed less optimal calibration, as detailed in Table 4. DCA Net Reclassification Improvement (NRI), and Integrated Discrimination Improvement (IDI) values comparing the Clinical and Hematological models to the Combined model are also presented inTable 4. Combined model demonstrated superior performance across all other evaluation metrics. DCA; Figure 7 further evaluated the clinical utility of the models. The Combined model exhibited the highest net benefit across a wide range of threshold probabilities in both the training and validation sets, indicating its superior clinical value compared to the Clinical and Hematological models, as well as the strategies of treating all or treating no patients. Individualized Prediction Explanations using Waterfall Plot To provide deeper insight into the decision-making process of the Extra Trees model, we visualized individual predictions using SHAP force plots (Figure 8). These plots reveal how specific feature values contribute to each prediction, shifting it from the average model output towards either malignancy or benignity. For instance, a true negative case (Figure 8A) was correctly classified primarily due to a small polyp diameter, a well-established indicator of low risk. Conversely, a false positive prediction (Figure 8B) was driven by a moderately elevated NLR, suggesting that systemic inflammation can, in some cases, outweigh the influence of favorable ultrasound characteristics. A false negative case (Figure 8C) illustrates the complexity of the interplay; despite an elevated CA19-9, other features, such as a smaller diameter, led to an incorrect benign prediction. Finally, a true positive case (Figure 8D) was correctly identified, largely due to a significantly larger polyp diameter, reinforcing the critical role of size in malignancy risk. These individual case analyses underscore the model’s ability to integrate diverse data types and highlight the nuanced interplay between clinical, ultrasound, and serological features in predicting gallbladder polyp malignancy. They also reveal potential areas for future refinement, such as incorporating additional biomarkers or imaging modalities to improve accuracy in cases where conflicting indicators are present. Web-based Calculator for Gallbladder Polyp Malignancy Prediction To facilitate the clinical translation of our prediction model, we developed a user-friendly, web-based calculator (Figure 9): https://gallbladder-kwljafh4ile9dlb9qu9y46.streamlit.app/. This tool allows clinicians to easily input the five key variables – polyp diameter, stalk presence, age, NLR, and CA19-9 – and obtain an immediate prediction of malignancy risk. The calculator, built upon the Combined Model, provides a readily accessible and practical means of implementing our research findings in a clinical setting. By providing a quantitative risk assessment, the calculator has the potential to aid in clinical decision-making, promoting more personalized management strategies for patients with GBPs. This represents a crucial step towards moving beyond subjective assessments and towards a more data-driven approach to GBP management. Discussion This retrospective study successfully developed and validated a non-invasive ML model for predicting gallbladder polyp malignancy, demonstrating excellent performance in our validation cohort. Our key finding is the robust predictive capability of a Combined Model integrating readily available clinical data, ultrasound features, and serological markers, outperforming models relying on single data modalities. The Extra Trees algorithm emerged as the optimal ML approach, achieving a high AUC of 0.97 in training and robust performance in validation. Furthermore, SHAP analysis provided valuable insights into the key predictors driving model performance, identifying polyp diameter, stalk morphology, NLR, age, and CA19-9 as the most influential variables. Finally, the development of a user-friendly web-based calculator represents a significant step towards clinical translation of our model. Consistent with established literature, polyp diameter emerged as a dominant predictor of malignancy in our model. Larger polyp size has been repeatedly shown to correlate with increased risk of gallbladder cancer [21, 22] . This is biologically plausible as larger polyps are more likely to harbor dysplastic or malignant transformation due to prolonged growth and increased cellular turnover. Our model effectively leveraged this well-established clinical risk factor, further validating its clinical relevance [9] . Sessile polyp morphology, indicated by the absence of a stalk (i.e., a wide-based or sessile polyp), was also identified as a significant predictor. Sessile polyps are known to have a higher malignant potential compared to pedunculated polyps, Song’s research also found that when the base of the polyp widens, it is more likely to be malignant [23] . Wang’s study also found that pedunculated polyps had a higher rate of malignancy compared to sessile polyps [24] . This may be attributed to their broader attachment to the gallbladder wall, potentially facilitating deeper invasion and lymphatic spread in case of malignancy. The model’s ability to incorporate this morphological feature underscores its capacity to capture nuanced ultrasound characteristics beyond simple size. Advanced age was another important predictor identified by SHAP analysis. Increasing age is a well-recognized risk factor for various cancers [25, 26] , including gallbladder cancer [27, 28] . This reflects the cumulative effect of genetic mutations, environmental exposures, and declining immune surveillance over time, increasing the likelihood of malignant transformation in GBPs as well [29, 30] . Elevated levels of CA19-9, a serological marker commonly associated with pancreatobiliary malignancies, also contributed significantly to the model’s predictive power [31] . While CA19-9 is not recommended for routine screening of gallbladder cancer, its elevation can reflect underlying malignant processes and has been shown to correlate with advanced stage and poorer prognosis in gallbladder cancer [32] . Our findings suggest that even within the context of GBPs, pre-operative CA19-9 levels can provide valuable information regarding malignancy risk, enhancing the discriminatory ability of our model. Finally, a higher NLR was identified as a significant predictor. NLR, a readily available marker of systemic inflammation, has gained increasing attention as a prognostic indicator in various cancers [12] . Chronic inflammation is recognized as a key driver of carcinogenesis, and an elevated NLR, reflecting a pro-tumorigenic inflammatory microenvironment, may indicate a higher likelihood of malignancy within a GBP. The inclusion of NLR in our model highlights the potential of readily available systemic inflammatory markers to improve non-invasive risk stratification [33, 34] . Compared to previous studies focusing primarily on ultrasound features or clinical risk factors alone, our study’s novelty lies in the integrated approach, combining ultrasound morphology, serological markers, and clinical data within an ML framework. While some studies have explored radiomics or advanced imaging techniques for gallbladder lesion characterization [35] , our model leverages routinely collected, clinically accessible data, making it readily translatable to real-world practice. Furthermore, the use of Extra Trees, a robust ensemble learning algorithm, and the comprehensive model evaluation using ROC curves, PR curves, calibration curves, and DCA, ensures the rigor and reliability of our findings. A key strength of our study is the visualization of feature importance using SHAP analysis. This not only enhances the interpretability of our “black box” ML model but also provides clinically relevant insights into the relative contribution of each predictor [36, 37] . Understanding which factors are most influential in driving the model’s predictions can improve clinician confidence in adopting and applying the model in practice. Moreover, the development of a user-friendly web-based calculator is a significant step towards facilitating clinical implementation [38] . By providing a readily accessible tool for risk prediction, we aim to bridge the gap between research and clinical practice, enabling clinicians to easily utilize our model to aid in decision-making for patients with GBPs. This calculator has the potential to streamline risk assessment, reduce reliance on subjective interpretation of ultrasound features, and ultimately contribute to more personalized and evidence-based management strategies. Despite these strengths, our study has limitations. The retrospective, single-center design may limit the generalizability of our findings to other populations and healthcare settings. External validation in independent, multi-center cohorts is crucial to confirm the robustness and broad applicability of our model. While our sample size of 202 patients is reasonable for a single-center study, a larger sample size in future studies could further enhance model performance and stability. Furthermore, we acknowledge that our model is based on pre-operative ultrasound and serological markers. Future research could explore the incorporation of more advanced imaging modalities, such as contrast-enhanced ultrasound or endoscopic ultrasound, and novel biomarkers to potentially further improve predictive accuracy. Finally, while SHAP analysis provides valuable insights into feature importance, further investigation into the underlying biological mechanisms linking these predictors to GBP malignancy is warranted. Future research directions should focus on external validation of our model in diverse patient populations and clinical settings. Prospective studies are needed to evaluate the clinical impact of using our model to guide management decisions for patients with GBPs, specifically assessing its ability to reduce unnecessary cholecystectomies and improve patient outcomes. Furthermore, exploring the cost-effectiveness of implementing our model in routine clinical practice is essential for its widespread adoption. Investigating the potential of combining our model with other diagnostic modalities, such as advanced imaging or molecular markers, could further refine risk stratification strategies. Conclusion Our study demonstrates the successful development and validation of a non-invasive ML model for predicting gallbladder polyp malignancy, integrating ultrasound features and serological markers. The Combined Model, leveraging readily available clinical data and facilitated by a user-friendly web calculator, offers a promising tool to improve risk stratification and potentially optimize the management of patients with GBPs. Further validation and prospective clinical implementation studies are warranted to fully realize the clinical potential of this novel approach. Abbreviations GBPs: Gallbladder polyps; CA19-9: Carbohydrate Antigen 19-9; CEA: Carcinoembryonic Antigen; NLR: Neutrophil-to-Lymphocyte Ratio; ML: Machine Learning; RIS: Radiology Information Systems; LIS: Laboratory Information Systems; ALT: Alanine Aminotransferase; AST: Aspartate Aminotransferase; TBIL: Total Bilirubin; ALP: Alkaline Phosphatase; ADA: Adenosine Deaminase; BMI: Body Mass Index; SHAP: SHapley Additive exPlanations; SVC: Support Vector Classifier; AUC: Area Under the Receiver Operating Characteristic Curve; ROC: Receiver Operating Characteristic; PR: Precision-Recall; PPV: Positive Predictive Value; NPV: Negative Predictive Value; DCA: Decision Curve Analysis; SD: Standard Deviation; IQR: Interquartile Range; NRI: Net Reclassification Improvement; IDI: Integrated Discrimination Improvement. Declarations Ethics approval and consent to participate This study was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki. The study protocol was approved by the Institutional Review Board of Fujian Provincial People’s Hospital. Informed consent was waived due to the retrospective nature of the study and the use of anonymized patient data. All data were handled in compliance with hospital policies and data protection regulations to ensure patient confidentiality. Consent for publication Not applicable Availability of data and materials The datasets used and analyzed during the current study are available from the corresponding author upon reasonable request. The data are not publicly available due to privacy and ethical restrictions. Competing interests The authors declare no conflicts of interest related to this study. No financial or non-financial benefits have been received or will be received from any party related directly or indirectly to the subject of this manuscript. Funding No funding. Authors' contributions Yang Yan: Conceptualization, Methodology, Formal analysis, Data curation, Software, Writing – Original Draft, Writing – Review & Editing, Project administration, Funding acquisition. Tu Haibin: Methodology, Data curation, Investigation, Resources, Validation, Visualization, Writing – Review & Editing. Lin Youguo: Data curation, Investigation, Resources, Validation, Writing – Review & Editing.Wei Jianting: Data curation, Investigation, Resources, Validation, Writing – Review & Editing. Acknowledgements We sincerely thank the patients who participated in this study and whose data made this research possible. We are grateful to the medical records staff at Fujian Provincial Hospital for their assistance in data collection. We also appreciate the contributions of the sonographers and pathologists who performed the ultrasound examinations and histopathological analyses, respectively. References Pineros M, Vignat J, Colombet M, Laversanne M, Ferreccio C, Heise K, Mhatre S, Koshiol J, Bray F. Global variations in gallbladder cancer incidence: What do recorded data and national estimates tell us? Int J Cancer 2025; 156(7): 1358-1368 [PMID: 39580808 DOI: 10.1002/ijc.35232] Zhang X, Xu C, Zhang H, Du X, Zhang Q, Lu M, Ma Y, Ma W. Gallbladder cancer incidence and mortality rate trends in China: analysis of data from the population-based cancer registry. BMC Public Health 2024; 24(1): 3122-3130 [PMID: 39529002 PMCID: PMC11555955 DOI: 10.1186/s12889-024-20584-9] Bhalla S, Shabbir N, Yadav K, Kumar M, Gupta N, Chaudhary S, Mithilesh, Sharma A, Agarwal P. Evaluating the Incidence of Incidental Gallbladder Carcinoma in a Tertiary Care Centre: A Retrospective Analysis in North India. Cureus 2024; 16(12): e76217-e76230 [PMID: 39867094 PMCID: PMC11757650 DOI: 10.7759/cureus.76217] Min JH, Choi SY, Kim SH, Kim YK, Hwang JA, Cha DI, Lee JH, Baek SY, Lee JE. Should we suspect gallbladder cancer if which CT finding is observed in patients with localized gallbladder wall thickening? Eur J Radiol 2024; 176: 111505-111518 [PMID: 38796886 DOI: 10.1016/j.ejrad.2024.111505] Son JH. Recent Updates on Management and Follow-up of Gallbladder Polyps. Korean J Gastroenterol 2023; 81(5): 197-202 [PMID: 37226819 DOI: 10.4166/kjg.2023.038] Liu JS, Wang XL, Fang J, Wang A, Yang XM, He B, Zhu WH. Current situation of surgical treatment of gallbladder polyps and some problems that should be paid attention to. Zhonghua Yi Xue Za Zhi 2024; 104(34): 3171-3174 [PMID: 39193604 DOI: 10.3760/cma.j.cn112137-20240415-00879] Liu H, Lu Y, Shen K, Zhou M, Mao X, Li R. Advances in the management of gallbladder polyps: establishment of predictive models and the rise of gallbladder-preserving polypectomy procedures. BMC Gastroenterol 2024; 24(1): 7-19 [PMID: 38166603 PMCID: PMC10759486 DOI: 10.1186/s12876-023-03094-7] Wang K, Xu Q, Xia L, Sun J, Shen K, Liu H, Xu L, Li R. Gallbladder polypoid lesions: Current practices and future prospects. Chin Med J (Engl) 2024; 137(14): 1674-1683 [PMID: 38420780 PMCID: PMC11268823 DOI: 10.1097/CM9.0000000000003019] Jiang D, Qian Y, Gu Y, Wang R, Yu H, Wang Z, Dong H, Chen D, Chen Y, Jiang H, Li Y. Establishing a radiomics model using contrast-enhanced ultrasound for preoperative prediction of neoplastic gallbladder polyps exceeding 10 mm. Eur J Med Res 2025; 30(1): 66-72 [PMID: 39901203 PMCID: PMC11789348 DOI: 10.1186/s40001-025-02292-1] Zhu L, Li N, Zhu Y, Han P, Jiang B, Li M, Luo Y, Clevert DA, Fei X. Value of high frame rate contrast enhanced ultrasound in gallbladder wall thickening in non-acute setting. Cancer Imaging 2024; 24(1): 7-19 [PMID: 38191513 PMCID: PMC10775603 DOI: 10.1186/s40644-023-00651-x] Dhingra S, Raman P, Ramsaroop T, Harrison I, Bergsten T, Nusbaum E, Feldman LE. Elevated serum CA 19-9 level mimicking pancreaticobiliary carcinoma from a hepatic abscess: case report and literature review. Front Med (Lausanne) 2024; 11: 1470046-1470058 [PMID: 39876872 PMCID: PMC11772410 DOI: 10.3389/fmed.2024.1470046] Liu F, Yin P, Jiao B, Shi Z, Qiao F, Xu J. Detecting the preoperative peripheral blood systemic immune-inflammation index (SII) as a tool for early diagnosis and prognosis of gallbladder cancer. BMC Immunol 2025; 26(1): 7-18 [PMID: 39966731 PMCID: PMC11834489 DOI: 10.1186/s12865-025-00683-x] Huang L, Deng X, Fan RZ, Hao TT, Zhang S, Sun B, Xu YH, Li SB, Feng YF. Coagulation and fibrinolytic markers offer utility when distinguishing between benign and malignant gallbladder tumors: A cross-sectional study. Clin Chim Acta 2024; 560: 119751-119762 [PMID: 38830523 DOI: 10.1016/j.cca.2024.119751] Huang EY, Reeves JJ, Broderick RC, Serra JL, Goldhaber NH, An JY, Fowler KJ, Hosseini M, Sandler BJ, Jacobsen GR, Horgan S, Clary BM. Distinguishing characteristics of xanthogranulomatous cholecystitis and gallbladder adenocarcinoma: a persistent diagnostic dilemma. Surg Endosc 2024; 38(1): 348-355 [PMID: 37783778 DOI: 10.1007/s00464-023-10461-8] Ruffle JK, Gray RJ, Mohinta S, Pombo G, Kaul C, Hyare H, Rees G, Nachev P. Computational limits to the legibility of the imaged human brain. Neuroimage 2024; 291(6): 120600-120614 [PMID: 38569979 DOI: 10.1016/j.neuroimage.2024.120600] Wang LF, Wang Q, Mao F, Xu SH, Sun LP, Wu TF, Zhou BY, Yin HH, Shi H, Zhang YQ, Li XL, Sun YK, Lu D, Tang CY, Yuan HX, Zhao CK, Xu HX. Risk stratification of gallbladder masses by machine learning-based ultrasound radiomics models: a prospective and multi-institutional study. Eur Radiol 2023; 33(12): 8899-8911 [PMID: 37470825 DOI: 10.1007/s00330-023-09891-8] Yuan K, Zhang X, Yang Q, Deng X, Deng Z, Liao X, Si W. Risk prediction and analysis of gallbladder polyps with deep neural network. Comput Assist Surg (Abingdon) 2024; 29(1): 2331774-2331785 [PMID: 38520294 DOI: 10.1080/24699322.2024.2331774] Kovacs KA, Kerepesi C, Rapcsak D, Madaras L, Nagy A, Takacs A, Dank M, Szentmartoni G, Szasz AM, Kulka J, Tokes AM. Machine learning prediction of breast cancer local recurrence localization, and distant metastasis after local recurrences. Sci Rep 2025; 15(1): 4868-4875 [PMID: 39929942 PMCID: PMC11811162 DOI: 10.1038/s41598-025-89339-9] Deng X, Liao Z. A machine-learning model based on dynamic contrast-enhanced MRI for preoperative differentiation between hepatocellular carcinoma and combined hepatocellular-cholangiocarcinoma. Clin Radiol 2024; 79(6): e817-e825 [PMID: 38413354 DOI: 10.1016/j.crad.2024.02.001] Long X, Zeng H, Zhang Y, Lu Q, Cao Z, Shu H. Development of a Reliable GADSAH Model for Differentiating AFP-negative Hepatic Benign and Malignant Occupying Lesions. J Hepatocell Carcinoma 2024; 11(6): 607-618 [PMID: 38549786 PMCID: PMC10973090 DOI: 10.2147/JHC.S452628] Fujiwara K, Abe A, Masatsugu T, Hirano T, Sada M. Effect of gallbladder polyp size on the prediction and detection of gallbladder cancer. Surg Endosc 2021; 35(9): 5179-5185 [PMID: 32974780 DOI: 10.1007/s00464-020-08010-8] Sarici IS, Duzgun O. Gallbladder polypoid lesions >15mm as indicators of T1b gallbladder cancer risk. Arab J Gastroenterol 2017; 18(3): 156-158 [PMID: 28958638 DOI: 10.1016/j.ajg.2017.09.003] Liu XS, Chen T, Gu LH, Guo YF, Li CY, Li FH, Wang J. Ultrasound-based scoring system for differential diagnosis of polypoid lesions of the gallbladder. J Gastroenterol Hepatol 2018; 33(6): 1295-1299 [PMID: 29280187 DOI: 10.1111/jgh.14080] Wang Y, Peng J, Liu K, Sun P, Ma Y, Zeng J, Jiang Y, Tan B, Cao J, Hu W. Preoperative prediction model for non-neoplastic and benign neoplastic polyps of the gallbladder. Eur J Surg Oncol 2024; 50(2): 107930 [PMID: 38159390 DOI: 10.1016/j.ejso.2023.107930] Satoh H, Okuma Y, Shinno Y, Masuda K, Matsumoto Y, Yoshida T, Goto Y, Horinouchi H, Yamamoto N, Ohe Y. Evolving treatments and prognosis in Stage IV non-small cell lung cancer: 20 years of progress of novel therapies. Lung Cancer 2025; 202: 108453-108462 [PMID: 40020466 DOI: 10.1016/j.lungcan.2025.108453] Li J, Zeng J, Yang Y, Huang B. Trend of skin cancer mortality and years of life lost in China from 2013 to 2021. Front Public Health 2025; 13(2): 1522790-1522801 [PMID: 40013033 PMCID: PMC11861555 DOI: 10.3389/fpubh.2025.1522790] Bozer A, Durgun N. Radiological Findings for Distinguishing Between Xanthogranulomatous Cholecystitis and Gallbladder Cancer. Arch Iran Med 2024; 27(12): 674-682 [PMID: 39891455 PMCID: PMC11786208 DOI: 10.34172/aim.31710] Pavlidis ET, Galanis IN, Pavlidis TE. Current considerations for the surgical management of gallbladder adenomas. World J Gastrointest Surg 2024; 16(6): 1507-1512 [PMID: 38983335 PMCID: PMC11229988 DOI: 10.4240/wjgs.v16.i6.1507] Yokoyama A, Kakiuchi N, Yoshizato T, Nannya Y, Suzuki H, Takeuchi Y, Shiozawa Y, Sato Y, Aoki K, Kim SK, Fujii Y, Yoshida K, Kataoka K, Nakagawa MM, Inoue Y, Hirano T, Shiraishi Y, Chiba K, Tanaka H, Sanada M, Nishikawa Y, Amanuma Y, Ohashi S, Aoyama I, Horimatsu T, Miyamoto S, Tsunoda S, Sakai Y, Narahara M, Brown JB, Sato Y, Sawada G, Mimori K, Minamiguchi S, Haga H, Seno H, Miyano S, Makishima H, Muto M, Ogawa S. Age-related remodelling of oesophageal epithelia by mutated cancer drivers. Nature 2019; 565(7739): 312-317 [PMID: 30602793 DOI: 10.1038/s41586-018-0811-x] Guan X, Meng X, Zhong G, Zhang Z, Wang C, Xiao Y, Fu M, Zhao H, Zhou Y, Hong S, Xu X, Bai Y, Kan H, Chen R, Wu T, Guo H. Particulate matter pollution, polygenic risk score and mosaic loss of chromosome Y in middle-aged and older men from the Dongfeng-Tongji cohort study. J Hazard Mater 2024; 471: 134315 [PMID: 38678703 DOI: 10.1016/j.jhazmat.2024.134315] Kim SH, Lee MJ, Hwang HK, Lee SH, Kim H, Paik YK, Kang CM. Prognostic potential of the preoperative plasma complement factor B in resected pancreatic cancer: A pilot study. Cancer Biomark 2019; 24(3): 335-342 [PMID: 30829612 DOI: 10.3233/CBM-181847] Dou C, Han Y, Lin L, Wen J, Zhao W, Yang Y, Guan S, Li X, Gao M, Lu J. Development and Validation of a Comprehensive Risk Prediction Model for Polypoid Lesions of the Gallbladder. Clin Exp Pharmacol Physiol 2025; 52(4): e70028-e70038 [PMID: 39929712 DOI: 10.1111/1440-1681.70028] Sikora-Skrabaka M, Walkiewicz KW, Waniczek D, Strzelczyk JK, Nowakowska-Zajdel E. Relationship Between Systemic Inflammatory Response Exponents, Levels of ADAM10, ADAM17 Proteins and Selected Clinical Parameters in Patients with Colorectal Cancer: Original Research Study. Int J Mol Sci 2025; 26(3): 68-79 [PMID: 39940871 PMCID: PMC11817235 DOI: 10.3390/ijms26031104] Maeda T, Shirakami Y, Taguchi D, Miwa T, Kubota M, Sakai H, Ibuka T, Mori K, Tomita H, Shimizu M. Glyburide Suppresses Inflammation-Related Colorectal Tumorigenesis Through Inhibition of NLRP3 Inflammasome. Int J Mol Sci 2024; 25(21): 68-79 [PMID: 39519191 PMCID: PMC11546087 DOI: 10.3390/ijms252111640] Wang Y, Qu C, Zeng J, Jiang Y, Sun R, Li C, Li J, Xing C, Tan B, Liu K, Liu Q, Zhao D, Cao J, Hu W. Establishing a preoperative predictive model for gallbladder adenoma and cholesterol polyps based on machine learning: a multicentre retrospective study. World J Surg Oncol 2025; 23(1): 27-38 [PMID: 39875897 PMCID: PMC11773841 DOI: 10.1186/s12957-025-03671-y] Fan Z, Jiang J, Xiao C, Chen Y, Xia Q, Wang J, Fang M, Wu Z, Chen F. Construction and validation of prognostic models in critically Ill patients with sepsis-associated acute kidney injury: interpretable machine learning approach. J Transl Med 2023; 21(1): 406-422 [PMID: 37349774 PMCID: PMC10286378 DOI: 10.1186/s12967-023-04205-4] Jiang A, Li J, Wang L, Zha W, Lin Y, Zhao J, Fang Z, Shen G. Multi-feature, Chinese-Western medicine-integrated prediction model for diabetic peripheral neuropathy based on machine learning and SHAP. Diabetes Metab Res Rev 2024; 40(4): e3801-e3815 [PMID: 38616511 DOI: 10.1002/dmrr.3801] Brenner T, Kuo A, Sperna Weiland CJ, Kamal A, Elmunzer BJ, Luo H, Buxbaum J, Gardner TB, Mok SS, Fogel ES, Phillip V, Choi JH, Lua GW, Lin CC, Reddy DN, Lakhtakia S, Goenka MK, Kochhar R, Khashab MA, van Geenen EJM, Singh VK, Tomasetti C, Akshintala VS. Development and validation of a machine learning-based, point-of-care risk calculator for post-ERCP pancreatitis and prophylaxis selection. Gastrointest Endosc 2025; 101(1): 129-138 [PMID: 39147103 DOI: 10.1016/j.gie.2024.08.009] Tables Table 1 Basic information of all patients Variable Total Training Validation Statistic P_Value Gallbladder polyps 152 (75.2%) 107 (75.4%) 45 (75%) 0 1 Gallbladder cancer 50 (24.8%) 35 (24.6%) 15 (25%) female 92 (45.5%) 67 (47.2%) 25 (41.7%) 0.32 0.57 male 110 (54.5%) 75 (52.8%) 35 (58.3%) stalk(sessile) 112 (55.4%) 85 (59.9%) 27 (45%) 3.19 0.07 stalk(pedunculated) 90 (44.6%) 57 (40.1%) 33 (55%) Waist_Hip_Ratio(Normal) 95 (47%) 67 (47.2%) 28 (46.7%) 0 1 Waist_Hip_Ratio(fatty) 107 (53%) 75 (52.8%) 32 (53.3%) Hypertension(no) 152 (75.2%) 104 (73.2%) 48 (80%) 0.7 0.4 Hypertension(yes) 50 (24.8%) 38 (26.8%) 12 (20%) Diabetes(no) 173 (85.6%) 124 (87.3%) 49 (81.7%) 0.69 0.41 Diabetes(yes) 29 (14.4%) 18 (12.7%) 11 (18.3%) Hyperlipidemia(no) 157 (77.7%) 115 (81%) 42 (70%) 2.34 0.13 Hyperlipidemia(yes) 45 (22.3%) 27 (19%) 18 (30%) Alcohol history(no) 192 (95%) 134 (94.4%) 58 (96.7%) 0.11 0.74 Alcohol history(yes) 10 (5%) 8 (5.6%) 2 (3.3%) Polyp_Location(bottom) 79 (39.1%) 54 (38%) 25 (41.7%) 0.42 0.81 Polyp_Location(body) 74 (36.6%) 54 (38%) 20 (33.3%) Polyp_Location(neck) 49 (24.3%) 34 (23.9%) 15 (25%) Polyp_Number(single) 100 (49.5%) 68 (47.9%) 32 (53.3%) 0.31 0.58 Polyp_Number(multiple) 102 (50.5%) 74 (52.1%) 28 (46.7%) Echo_Homogeneity 75 (37.1%) 52 (36.6%) 23 (38.3%) 0.01 0.94 Echo_Heterogeneous 127 (62.9%) 90 (63.4%) 37 (61.7%) Polyp Surface Smoothness 170 (84.2%) 122 (85.9%) 48 (80%) 0.71 0.4 Polyp Surface Coarse 32 (15.8%) 20 (14.1%) 12 (20%) Hyperechoic_Foci(no) 101 (50%) 77 (54.2%) 24 (40%) 2.87 0.09 Hyperechoic_Foci(yes) 101 (50%) 65 (45.8%) 36 (60%) Gallstones(no) 61 (30.2%) 43 (30.3%) 18 (30%) 0 1 Gallstones(yes) 141 (69.8%) 99 (69.7%) 42 (70%) Gallbladder_Smooth 155 (76.7%) 108 (76.1%) 47 (78.3%) 0.03 0.87 Gallbladder_Roughness 47 (23.3%) 34 (23.9%) 13 (21.7%) Viral_Hepatitis(no) 81 (40.1%) 59 (41.5%) 22 (36.7%) 0.24 0.62 Viral_Hepatitis(yes) 121 (59.9%) 83 (58.5%) 38 (63.3%) age(years) 59.9 ± 15.89 59.45 ± 15.95 60.97 ± 15.84 0.44 0.5 diameter(mm) 13.99 ± 3.32 13.75 ± 3.43 14.55 ± 2.99 2.51 0.11 NLR 2.97 ± 1.82 3.03 ± 1.82 2.82 ± 1.83 0.66 0.42 BMI 22.1 ± 4.01 22.1 ± 3.96 22.09 ± 4.18 0 0.99 TBIL（mg/dL） 16.9 ± 8.71 16.72 ± 8.79 17.33 ± 8.6 0.29 0.59 ADA（IU/L） 12.1 ± 8.03 12.58 ± 8.14 10.98 ± 7.71 2.14 0.14 CEA（ng/mL) 4.07 ± 3.87 4.21 ± 3.96 3.74 ± 3.66 0.78 0.38 CA199（U/mL） 24.24 ± 20.19 23.54 ± 19.9 25.89 ± 20.94 0.97 0.33 ALP（IU/L） 104.67 ± 50.22 106.18 ± 52.89 101.1 ± 43.45 0 0.96 AST（IU/L） 29.74 ± 16.07 28.88 ± 15.58 31.78 ± 17.15 1.56 0.21 ALT（IU/L） 34.82 ± 21.73 33.48 ± 21.56 37.98 ± 21.99 2.71 0.1 NLR: Neutrophil-to-Lymphocyte Ratio; BMI: Body Mass Index; TBIL: Total Bilirubin; ADA: Adenosine Deaminase; CEA: Carcinoembryonic Antigen; CA199: Carbohydrate Antigen 19-9; ALP: Alkaline Phosphatase; AST: Aspartate Aminotransferase; ALT: Alanine Aminotransferase. Table 2 Comparison of Predictive Performance of Different Machine learning Model Mean AUC AUC 95% CI Lower AUC 95% CI Upper Mean Accuracy Sensitivity Specificity PPV NPV Logistic Regression 0.85 0.75 0.9 0.83 0.92 0.75 0.9 0.8 Random Forest 0.9 0.81 0.99 0.91 0.95 0.87 0.92 0.88 Gradient Boosting 0.88 0.8 0.99 0.89 0.88 0.91 0.89 0.9 SVC 0.76 0.69 0.85 0.78 0.81 0.82 0.79 0.84 Decision Tree 0.89 0.84 0.94 0.91 0.9 0.85 0.91 0.86 K-Nearest Neighbors 0.88 0.83 0.94 0.83 0.54 0.95 0.94 0.86 Naive Bayes 0.85 0.81 0.88 0.82 0.7 0.79 0.85 0.83 LightGBM 0.8 0.76 0.85 0.76 0.85 0.92 0.72 0.77 Extra Trees 0.97 0.94 0.99 0.95 0.95 0.94 0.92 0.92 AdaBoost 0.86 0.84 0.89 0.83 0.7 0.95 0.72 0.91 Table 3 Comparison of Predictive Performance of Different Models Dataset Model AUC AUC 95% CI Lower AUC 95% CI Upper Sensitivity Specificity PPV NPV Accuracy Train Combined Model 0.93 0.88 0.98 0.66 0.94 0.8 0.89 0.87 Clinical Model 0.89 0.83 0.95 0.54 0.93 0.72 0.86 0.83 Hematological Model 0.68 0.58 0.77 0 1 0 0.75 0.75 Validation Combined Model 0.93 0.86 1 0.54 0.93 0.75 0.85 0.83 Clinical Model 0.89 0.8 0.98 0.54 0.96 0.85 0.86 0.86 Hematological Model 0.68 0.54 0.82 0 1 0 0.74 0.74 AUC: Area Under the Curve; CI: Confidence Interval; PPV: Positive Predictive Value; NPV: Negative Predictive Value. Table 4 Calibration Metrics and NRI/IDI of Different Models in Different Cohorts Dataset Model Brier Score HL p-value Calibration Slope Calibration Intercept NRI IDI Train Combined 0.09 0.25 1.03 <0.01 Clinical 0.12 0.2 1.16 -0.02 0.11 0.1 Hematological 0.17 <0.01 1.76 -0.17 0.62 0.41 Validation Combined 0.08 0.32 0.95 0.11 Clinical 0.15 0.03 0.85 0.25 0.15 0.11 Hematological 0.19 <0.01 0.25 0.32 0.44 0.29 HL: Hosmer-Lemeshow; NRI: Net Reclassification Improvement; IDI: Integrated Discrimination Improvement. Additional Declarations No competing interests reported. Supplementary Files Supplementarytable1.xlsx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 20 Feb, 2026 Reviews received at journal 12 Jul, 2025 Reviews received at journal 06 Jul, 2025 Reviewers agreed at journal 02 Jul, 2025 Reviewers agreed at journal 29 Jun, 2025 Reviewers invited by journal 20 Jun, 2025 Editor invited by journal 28 May, 2025 Editor assigned by journal 27 May, 2025 Submission checks completed at journal 27 May, 2025 First submitted to journal 25 May, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6744318","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":474627684,"identity":"6b23d0ec-ce44-47d8-8139-cbdf2d20da51","order_by":0,"name":"Yang Yan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAs0lEQVRIiWNgGAWjYLCCDwY2cvzMzIcfEK2DcUZBmrFkO1uaAdFamHk+HE7ccJ5HQYIo5brtzccezjBIY9x8mIfBgKHGJpqgFrMzx9INgH5hNjvMe+ABw7G03AaCWm7kmEkCbWEzO8yXYMDYcJg4LdI8Bod5jJt5DCRI0iJhwEy0ljPH0kAOM5A4DAzkBKL8crz5mMSHPzb1/f2HDz/4UGNDWAsqSCBN+SgYBaNgFIwCXAAAs/g+yXFIdeoAAAAASUVORK5CYII=","orcid":"","institution":"Preventive Treatment of Disease in Traditional Chinese Medicine，The Affiliated People’s Hospital of Fujian University of Traditional Chinese Medicine，Fuzhou","correspondingAuthor":true,"prefix":"","firstName":"Yang","middleName":"","lastName":"Yan","suffix":""},{"id":474627685,"identity":"f551111e-6d87-427f-a63c-f6783060a5c0","order_by":1,"name":"Tu Haibin","email":"","orcid":"","institution":"Mengchao Hepatobiliary Hospital of Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Tu","middleName":"","lastName":"Haibin","suffix":""},{"id":474627686,"identity":"323eab06-6b74-4f9d-bb48-c54a0e64ec6c","order_by":2,"name":"Lin Youguo","email":"","orcid":"","institution":"Preventive Treatment of Disease in Traditional Chinese Medicine，The Affiliated People’s Hospital of Fujian University of Traditional Chinese Medicine，Fuzhou","correspondingAuthor":false,"prefix":"","firstName":"Lin","middleName":"","lastName":"Youguo","suffix":""},{"id":474627687,"identity":"4a8e6490-7270-4759-ab5d-9e9e455f81a7","order_by":3,"name":"Wei Jianting","email":"","orcid":"","institution":"Preventive Treatment of Disease in Traditional Chinese Medicine，The Affiliated People’s Hospital of Fujian University of Traditional Chinese Medicine，Fuzhou","correspondingAuthor":false,"prefix":"","firstName":"Wei","middleName":"","lastName":"Jianting","suffix":""}],"badges":[],"createdAt":"2025-05-25 14:38:05","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6744318/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6744318/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":85349654,"identity":"db80217b-389d-4385-8e0d-3041060b4218","added_by":"auto","created_at":"2025-06-25 02:38:58","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":70812,"visible":true,"origin":"","legend":"\u003cp\u003ePatitent inclusion flow\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6744318/v1/16b101efa4d16c6e3cf79e48.png"},{"id":85348956,"identity":"24a8c073-1a25-4d3c-8772-f4d3d65bdfe0","added_by":"auto","created_at":"2025-06-25 02:30:58","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":99918,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP Summary Plot of Feature Importance for Gallbladder Polyp Malignancy Prediction.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6744318/v1/c546b3c944d2ebb5cba8a722.png"},{"id":85349657,"identity":"f8353489-6cc6-469c-a925-92a3727c6ba4","added_by":"auto","created_at":"2025-06-25 02:38:58","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":34170,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAUC vs. Number of Features for Polyp Prediction\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6744318/v1/0114818e4af6461becf8de19.png"},{"id":85348969,"identity":"a26f2349-fff9-4ca7-b2ae-692871c02eff","added_by":"auto","created_at":"2025-06-25 02:30:58","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":72153,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eReceiver Operating Characteristic (ROC) Curves of Different Models on the Training and Validation Sets\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6744318/v1/acf10ece569d13d983ddd6b2.png"},{"id":85348961,"identity":"b32905a4-5078-4022-b88c-567d3320cb58","added_by":"auto","created_at":"2025-06-25 02:30:58","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":70024,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePrecision-Recall (PR) Curves of the Combined, Clinical, and Hematological Models on the Training and Validation Sets\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6744318/v1/e03c30b36caa9e6b82ae1306.png"},{"id":85350087,"identity":"e4cf10fe-3806-4015-9935-1140dd30490a","added_by":"auto","created_at":"2025-06-25 02:46:58","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":75676,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCalibration Curves of the Combined, Clinical, and Hematological Models on the Training and Validation Sets\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-6744318/v1/6bdd764c523079bdec1ec2fa.png"},{"id":85349658,"identity":"27516e1b-aa71-4b61-a895-4db33c21cc9c","added_by":"auto","created_at":"2025-06-25 02:38:58","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":98392,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDecision Curve Analysis (DCA) of the Combined, Clinical, and Hematological Models on the Training and Validation Sets\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-6744318/v1/d1e821af42ef213f54b31fa6.png"},{"id":85348970,"identity":"e3f8cc2f-22c0-412b-83c4-66f200799226","added_by":"auto","created_at":"2025-06-25 02:30:58","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":61359,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIndividualized Prediction Explanations using Waterfall Plots.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-6744318/v1/3ad3bc21f89e6f9e71c5ba7a.png"},{"id":85348979,"identity":"20347820-7709-4a12-905e-e47e2aedb216","added_by":"auto","created_at":"2025-06-25 02:30:59","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":18193,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eWeb-based Calculator for Gallbladder Polyp Malignancy Prediction.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-6744318/v1/f419911c7fda25e7e3bc1fbe.png"},{"id":85350796,"identity":"f4661aec-086e-429b-b017-0c919e828168","added_by":"auto","created_at":"2025-06-25 02:55:01","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1817446,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6744318/v1/be5b6bf9-b6ab-4712-92a4-fcd750d4db22.pdf"},{"id":85348954,"identity":"15c3b9de-cfc7-4b04-b716-2282535fd581","added_by":"auto","created_at":"2025-06-25 02:30:58","extension":"xlsx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":11693,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarytable1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-6744318/v1/327ec6a77863469d6f682242.xlsx"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Multimodal Machine Learning Model Integrating Ultrasound and Serological Biomarkers for Non-Invasive Prediction of Gallbladder Polyp Malignancy: Development, Validation, and Clinical Translation","fulltext":[{"header":"Introduction","content":"\u003cp\u003eGallbladder polyps (GBPs) are frequently detected during abdominal ultrasonography, with a reported prevalence ranging from 0.3\u0026ndash;12% in the general population\u003csup\u003e[1, 2]\u003c/sup\u003e. While the majority of GBPs are benign, a clinically significant proportion represents premalignant or malignant lesions\u003csup\u003e[3]\u003c/sup\u003e. Accurate differentiation between benign and malignant GBPs is paramount for guiding appropriate clinical management, as early detection and surgical resection of malignant lesions are essential for improving patient prognosis\u003csup\u003e[4]\u003c/sup\u003e .\u003c/p\u003e \u003cp\u003eUltrasound imaging plays a crucial role in the initial assessment of GBPs, providing vital information such as polyp size, number, morphology, and the presence of gallstones or gallbladder wall irregularities\u003csup\u003e[5, 6]\u003c/sup\u003e. Studies have indicated that specific ultrasound features, including polyp diameter exceeding 10 mm, sessile morphology, and single polyp presentation, are associated with an elevated risk of malignancy\u003csup\u003e[7, 8]\u003c/sup\u003e. However, the diagnostic accuracy of ultrasound alone remains limited, as overlapping features between benign and malignant polyps often lead to diagnostic ambiguity\u003csup\u003e[9, 10]\u003c/sup\u003e. Conversely, serological markers such as carbohydrate antigen 19\u0026thinsp;\u0026minus;\u0026thinsp;9 (CA19-9), carcinoembryonic antigen (CEA), and neutrophil-to-lymphocyte ratio (NLR) have been investigated as potential indicators of gallbladder malignancy\u003csup\u003e[11\u0026ndash;13]\u003c/sup\u003e. Elevated levels of these markers have been correlated with malignant transformation\u003csup\u003e[14]\u003c/sup\u003e, yet their independent predictive value is insufficient for clinical decision-making. Despite the complementary nature of ultrasound and serological markers, limited research has integrated these modalities to enhance the prediction of GBP malignancy.\u003c/p\u003e \u003cp\u003eAlthough both ultrasound features and serological markers offer valuable insights, their independent application has limitations. Notably, there is a scarcity of studies comprehensively integrating these two modalities for GBP malignancy prediction, representing a significant gap in the current diagnostic approach. The challenge lies in effectively combining these disparate data types to improve diagnostic accuracy.\u003c/p\u003e \u003cp\u003eMachine learning (ML) provides a robust solution to this challenge\u003csup\u003e[15]\u003c/sup\u003e. ML algorithms are specifically designed to manage complex interactions and non-linear relationships among multiple variables, rendering them ideally suited for integrating diverse data sources\u003csup\u003e[16]\u003c/sup\u003e. In oncology, ML has demonstrated considerable success in enhancing diagnostic accuracy, predicting treatment response, and identifying prognostic factors across various cancer types. For instance, ML models incorporating radiomic features from CT scans have been utilized to predict lymph node metastasis in lung cancer\u003csup\u003e[17]\u003c/sup\u003e. In breast cancer, ML algorithms integrating genomic data and clinical parameters have improved the prediction of recurrence risk\u003csup\u003e[18]\u003c/sup\u003e. Furthermore, ML has been applied to differentiate benign from malignant liver lesions using a combination of imaging and serological data, achieving high diagnostic performance\u003csup\u003e[19, 20]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eInspired by these advancements, we hypothesized that an ML model integrating both ultrasound features and serological biomarkers could significantly improve the non-invasive prediction of malignancy in GBPs. By leveraging the capacity of ML to analyze complex interactions between these data types, we aimed to develop a more accurate and reliable diagnostic tool. This study, therefore, sought to develop and validate such a model, potentially transforming the clinical management of GBPs by enabling more informed decision-making, reducing unnecessary cholecystectomies, and facilitating earlier detection of malignant lesions.\u003c/p\u003e"},{"header":"Patients and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy Design and Population\u003c/h2\u003e \u003cp\u003eThis study employed a retrospective cohort design, reviewing data from patients who underwent cholecystectomy at Fujian Provincial People\u0026rsquo;s Hospital between January 1, 2018, and January 1, 2024. The study protocol was approved by the Institutional Review Board of Fujian Provincial People\u0026rsquo;s Hospital and conducted in accordance with the Declaration of Helsinki. Patients were included if they had undergone cholecystectomy and had a confirmed histopathological diagnosis of GBPs. Exclusion criteria were: (1) patients with a pre-operative diagnosis of gallbladder carcinoma; (2) patients who underwent cholecystectomy for reasons other than GBPs (e.g., cholecystitis, biliary dyskinesia without polyps); (3) patients with incomplete clinical data, defined as missing ultrasound reports, serological marker results, or histopathology reports; and (4) patients with a history of prior biliary tract surgery or malignancy. The patient inclusion flow was showed in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eData Collection\u003c/h3\u003e\n\u003cp\u003eData were retrospectively extracted from electronic medical records, radiology information systems (RIS) for ultrasound reports, and laboratory information systems (LIS) for serological marker results. The following variables were collected for each patient:\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eUltrasound Features\u003c/strong\u003e \u003cp\u003eAll examinations were performed using high-resolution ultrasound systems (Mindray Pesona 7S/8S, Philips EPIQ5; C5-2 transducer) under standardized protocols. Two board-certified sonographers independently interpreted the findings through consensus-based evaluation. In cases of diagnostic discrepancy (observed in 3.8% of examinations), a third senior sonographer (with \u0026gt;\u0026thinsp;15 years\u0026rsquo; experience) conducted blinded reassessment to achieve definitive interpretation. Finalized parameters were prospectively documented in radiology reports for subsequent analysis.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003ePolyp Diameter (mm): Maximum diameter of the largest polyp, measured in millimeters.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePolyp Number: Total number of polyps identified in the gallbladder, categorized as \u0026lsquo;single\u0026rsquo; or \u0026lsquo;multiple\u0026rsquo;.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePolyp Stalk: Presence or absence of a stalk, categorized as \u0026lsquo;pedunculated\u0026rsquo; (presence of a stalk) or \u0026lsquo;sessile\u0026rsquo; (absence of a stalk). In cases of pedunculated polyps, stalk thickness was not routinely measured and thus not included.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePolyp Location: Anatomical location of the polyp within the gallbladder, categorized as \u0026lsquo;bottom\u0026rsquo;, \u0026lsquo;body\u0026rsquo;, or \u0026lsquo;neck\u0026rsquo;.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePolyp Surface Characteristics: Described as \u0026lsquo;smooth\u0026rsquo; or \u0026lsquo;coarse\u0026rsquo; based on the ultrasound report.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eEcho Homogeneity: Polyp echogenicity was categorized as \u0026lsquo;homogeneous\u0026rsquo; or \u0026lsquo;heterogeneous\u0026rsquo;.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eHyperechoic Foci: Presence or absence of hyperechoic foci within the polyp, documented as \u0026lsquo;yes\u0026rsquo; or \u0026lsquo;no\u0026rsquo;.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eGallbladder Stones: Presence or absence of gallstones within the gallbladder, documented as \u0026lsquo;yes\u0026rsquo; or \u0026lsquo;no\u0026rsquo;.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eGallbladder Wall Roughness: Presence or absence of gallbladder wall roughness, described in reports as irregular or thickened gallbladder wall contour, documented as \u0026lsquo;yes\u0026rsquo; or \u0026lsquo;no\u0026rsquo;. This was distinct from gallbladder wall thickness, which was not consistently reported and therefore not included.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eSerological Markers\u003c/strong\u003e \u003cp\u003ePre-operative serum samples were collected within one week prior to cholecystectomy. The following hematological markers were recorded, with units of measurement specified\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eAlanine Aminotransferase (ALT, U/L)\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eAspartate Aminotransferase (AST, U/L)\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eTotal Bilirubin (TBIL, \u0026micro;mol/L)\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eNeutrophil-to-Lymphocyte Ratio (NLR): Calculated as absolute neutrophil count divided by absolute lymphocyte count.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eCarbohydrate Antigen 19\u0026thinsp;\u0026minus;\u0026thinsp;9 (CA19-9, U/mL)\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eCarcinoembryonic Antigen (CEA, ng/mL)\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eAlkaline Phosphatase (ALP, U/L)\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eAdenosine Deaminase (ADA, IU/L)\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eDemographic and Clinical Data\u003c/strong\u003e \u003cp\u003eThe following demographic and clinical variables were collected\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eAge (years): Age at the time of cholecystectomy.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eSex: Categorized as \u0026lsquo;male\u0026rsquo; or \u0026lsquo;female\u0026rsquo;.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eBody Mass Index (BMI, kg/m\u0026sup2;): Calculated as weight in kilograms divided by height in meters squared.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eWaist-to-Hip Ratio: Measured at the time of admission, calculated as waist circumference divided by hip circumference, and categorized as \u0026lsquo;normal\u0026rsquo; or \u0026lsquo;fatty\u0026rsquo;.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eComorbidities: Presence or absence of hypertension, diabetes, hyperlipidemia, and viral hepatitis, documented as \u0026lsquo;yes\u0026rsquo; or \u0026lsquo;no\u0026rsquo;.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eAlcohol History: History of alcohol consumption, documented as \u0026lsquo;yes\u0026rsquo; or \u0026lsquo;no\u0026rsquo;.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e\n\u003ch3\u003eData Preprocessing\u003c/h3\u003e\n\u003cp\u003ePrior to model development, data preprocessing steps were undertaken. Missing values, present in less than 5% of the dataset and primarily in CEA and ALP measurements, were imputed using median imputation, as these markers were not expected to be highly skewed. Outliers were assessed using boxplots and defined as values exceeding 1.5 times the interquartile range above the 75th percentile or below the 25th percentile. No outliers were removed as clinically implausible, but extreme values were winsorized to the 99th percentile to mitigate undue influence on model training. Continuous variables were standardized using z-score normalization to ensure features were on a comparable scale, improving the performance and convergence of certain ML algorithms.\u003c/p\u003e\n\u003ch3\u003eMachine Learning Model Development\u003c/h3\u003e\n\u003cp\u003eThe dataset was randomly split into a training set (70%) and a validation set (30%) using a fixed random seed (seed\u0026thinsp;=\u0026thinsp;123456) to ensure reproducibility. Ten ML algorithms were employed for model development using the scikit-learn library in Python (3.6.1). The algorithms included: Logistic Regression, Random Forest, Gradient Boosting, Support Vector Classifier (SVC) with a radial basis function kernel, Decision Tree, K-Nearest Neighbors, Naive Bayes (Gaussian Naive Bayes), LightGBM, Extra Trees, and AdaBoost.\u003c/p\u003e \u003cp\u003eHyperparameter tuning for each algorithm was performed using a grid search approach with 10-fold cross-validation on the training set to optimize model performance and prevent overfitting. Ten-fold cross-validation was implemented to robustly estimate the performance of each model on unseen data within the training set. In each fold, the training data was further divided into 9 folds for training and 1 fold for validation. The average Area Under the Receiver Operating Characteristic Curve (AUC) across the 10 folds was used to evaluate each algorithm\u0026rsquo;s performance.\u003c/p\u003e \u003cp\u003eModel selection was based on the highest average AUC achieved during the 10-fold cross-validation in the training set. The Extra Trees algorithm demonstrated the highest AUC and was selected for further analysis and evaluation on the validation set.\u003c/p\u003e \u003cp\u003eFeature importance analysis was conducted using SHapley Additive exPlanations (SHAP) values, SHAP values provide a unified measure of feature importance by quantifying the contribution of each feature to the prediction of individual instances. The top ten most important variables, ranked by mean absolute SHAP value, were visualized to understand their relative influence on the model\u0026rsquo;s predictions.\u003c/p\u003e \u003cp\u003eTo explore model parsimony, we iteratively built models with varying numbers of top-ranked features, starting with the single most important feature and incrementally adding features based on their SHAP ranking. The AUC was calculated for each model configuration using 10-fold cross-validation on the training set to identify the optimal number of variables that maximized performance while maintaining model simplicity.\u003c/p\u003e \u003cp\u003eBased on the variables included, three model categories were defined:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eClinical Model: Included polyp diameter, polyp stalk (sessile/pedunculated), and patient age \u0026ndash; variables readily available from routine clinical assessment and basic ultrasound.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eHematological Model: Included neutrophil-to-lymphocyte ratio (NLR) and carbohydrate antigen 19\u0026thinsp;\u0026minus;\u0026thinsp;9 (CA19-9) \u0026ndash; serological markers with potential relevance to inflammation and malignancy.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eCombined Model: Included all collected variables, representing the integration of clinical, ultrasound, and hematological data.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e\n\u003ch3\u003eModel Evaluation\u003c/h3\u003e\n\u003cp\u003eThe performance of the selected Extra Trees model and the three defined model categories (Clinical, Hematological, and Combined) were evaluated on the independent validation set. Performance metrics included: AUC, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy. Receiver Operating Characteristic (ROC) curves and Precision-Recall (PR) curves were plotted to visualize the trade-off between sensitivity and specificity, and precision and recall, respectively. Calibration curves were generated to assess the reliability of the predicted probabilities, using isotonic regression for calibration. Decision Curve Analysis (DCA) was performed to evaluate the clinical utility of the models by quantifying the net benefit across a range of clinically relevant risk thresholds.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eWeb-Based Calculator Development\u003c/h2\u003e \u003cp\u003eTo facilitate clinical translation, a user-friendly web-based calculator was developed using specify web framework. The calculator is hosted at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://gallbladder-kwljafh4ile9dlb9qu9y46.streamlit.app/\u003c/span\u003e\u003cspan address=\"https://gallbladder-kwljafh4ile9dlb9qu9y46.streamlit.app/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u0026rdquo;. The calculator allows clinicians to input patient clinical characteristics, ultrasound features, and serological marker values, and obtain a predicted probability of gallbladder polyp malignancy based on the Combined Model. The calculator interface is designed for ease of use and rapid risk assessment in a clinical setting.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eStatistical analyses were performed using Python (3.6.1) with the libraries scikit-learn, pandas, numpy, matplotlib, seaborn, and statsmodels. Continuous variables are presented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation (SD) or median (interquartile range [IQR]) depending on normality, assessed using the Shapiro-Wilk validation. Categorical variables are presented as frequencies and percentages. Differences between the training and validation sets for continuous variables were assessed using the independent samples t-validation or Mann-Whitney U validation as appropriate, and for categorical variables using the chi-square validation. A two-sided p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eBasic Information of All Patients\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study included a total of 202 patients, of whom 50 were diagnosed with gallbladder cancer. The cohort was divided into a training set (n = 142, including 35 with gallbladder cancer) and a validation set (n = 60, including 15 with gallbladder cancer). Of the entire cohort, 110 patients (54.5%) were male, and the mean age was 59.9 \u0026plusmn; 15.89 years. Additional baseline characteristics are presented in Table 1. Crucially, the training and validation sets were well-balanced, with no statistically significant differences observed between the groups for any of the measured variables (all p \u0026gt; 0.05). The information in the training set is shown in Supplementary Table 1.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eComparison of Predictive Performance of Different Machine Learning Models\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTable 2 presents a comprehensive comparison of the predictive performance of various ML models using multiple evaluation metrics, including mean AUC, 95% confidence intervals for AUC, mean accuracy, sensitivity, specificity, PPV, and NPV. Among the models evaluated, the Extra Trees classifier demonstrated the highest predictive performance, with a mean AUC of 0.97 (95% CI: 0.94\u0026ndash;0.99), significantly outperforming other models. Additionally, the Extra Trees model achieved the highest mean accuracy (0.95), sensitivity (0.95), specificity (0.94), PPV (0.92), and NPV (0.92), indicating its robust ability to discriminate between positive and negative cases. Given its superior performance across all metrics, the Extra Trees model was selected for further analysis in this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSelection of the Optimal Number of Variables\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAfter selecting the Extra Trees model as the predictive algorithm, we ranked the variables based on their importance and visualized the results using SHAP, as illustrated in Figure 2. The top ten variables identified were: diameter, stalk, NLR, age, CA19-9, AST, BMI, CEA, ALP, and TBIL. To adhere to the principle of model parsimony, we calculated the corresponding AUC values for models incorporating different numbers of variables, as shown inFigure 3. The results demonstrated that the AUC reached a high value when the model included five variables, and further inclusion of additional variables did not significantly improve the AUC. Therefore, for subsequent analyses, we constructed a fusion model incorporating the following five variables: diameter, stalk, NLR, age, and CA19-9.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel Classification and Comparative Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBased on the selected variables, we categorized diameter, stalk, and age as the \u0026lsquo;Clinical Model\u0026rsquo;. CA19-9 and NLR were classified as the \u0026lsquo;Hematological Model\u0026rsquo;. We then compared the predictive performance of these two models, along with a \u0026lsquo;Combined Model\u0026rsquo; incorporating all five variables. ROC curves for each model are presented in Figure 4, and the PR curve is presented in Figure 5. Both the ROC and PR curves demonstrated the superior performance of the Combined model. AUC values and performance metrics are summarized in Table 3. The Combined Model consistently outperformed the other two models, achieving the highest AUC values (0.93 for both training and validation sets) and demonstrating balanced sensitivity, specificity, and accuracy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCalibration Curve\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 6 displays the calibration curves for the training and validation cohorts. In both datasets, the Combined Model exhibits the best calibration, with its curve tracking closest to the ideal 45-degree diagonal, indicating good agreement between predicted probabilities and observed proportions. The Clinical Model shows a slight tendency towards over-calibration in the higher probability ranges, particularly in the validation set. The Hematological Model demonstrates the least satisfactory calibration, deviating more noticeably from the ideal line in both datasets. These calibration plots visually confirm the superior calibration of the Combined Model compared to the other two models.Table 4 presents the calibration metrics for all models. The Combined model demonstrated good calibration in both the training and validation sets. In the training set, the Combined model had a Brier score of 0.09, an HL p-value of 0.25, a calibration slope of 1.03, and a calibration intercept of \u0026lt;0.01. In the validation set, the Combined model maintained good calibration, with a Brier score of 0.08, an HL p-value of 0.32, a calibration slope of 0.95, and a calibration intercept of 0.11. These values indicate that the Combined model\u0026rsquo;s predicted probabilities are well-calibrated with the observed outcomes. The Clinical and Hematological models showed less optimal calibration, as detailed in Table 4.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDCA\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNet Reclassification Improvement (NRI), and Integrated Discrimination Improvement (IDI) values comparing the Clinical and Hematological models to the Combined model are also presented inTable 4. Combined model demonstrated superior performance across all other evaluation metrics. DCA; Figure 7 further evaluated the clinical utility of the models. The Combined model exhibited the highest net benefit across a wide range of threshold probabilities in both the training and validation sets, indicating its superior clinical value compared to the Clinical and Hematological models, as well as the strategies of treating all or treating no patients.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIndividualized Prediction Explanations using Waterfall Plot\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo provide deeper insight into the decision-making process of the Extra Trees model, we visualized individual predictions using SHAP force plots (Figure 8). These plots reveal how specific feature values contribute to each prediction, shifting it from the average model output towards either malignancy or benignity. For instance, a true negative case (Figure 8A) was correctly classified primarily due to a small polyp diameter, a well-established indicator of low risk. Conversely, a false positive prediction (Figure 8B) was driven by a moderately elevated NLR, suggesting that systemic inflammation can, in some cases, outweigh the influence of favorable ultrasound characteristics. A false negative case (Figure 8C) illustrates the complexity of the interplay; despite an elevated CA19-9, other features, such as a smaller diameter, led to an incorrect benign prediction. Finally, a true positive case (Figure 8D) was correctly identified, largely due to a significantly larger polyp diameter, reinforcing the critical role of size in malignancy risk. These individual case analyses underscore the model\u0026rsquo;s ability to integrate diverse data types and highlight the nuanced interplay between clinical, ultrasound, and serological features in predicting gallbladder polyp malignancy. They also reveal potential areas for future refinement, such as incorporating additional biomarkers or imaging modalities to improve accuracy in cases where conflicting indicators are present.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eWeb-based Calculator for Gallbladder Polyp Malignancy Prediction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo facilitate the clinical translation of our prediction model, we developed a user-friendly, web-based calculator (Figure 9): https://gallbladder-kwljafh4ile9dlb9qu9y46.streamlit.app/. This tool allows clinicians to easily input the five key variables \u0026ndash; polyp diameter, stalk presence, age, NLR, and CA19-9 \u0026ndash; and obtain an immediate prediction of malignancy risk. The calculator, built upon the Combined Model, provides a readily accessible and practical means of implementing our research findings in a clinical setting. By providing a quantitative risk assessment, the calculator has the potential to aid in clinical decision-making, promoting more personalized management strategies for patients with GBPs. This represents a crucial step towards moving beyond subjective assessments and towards a more data-driven approach to GBP management.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis retrospective study successfully developed and validated a non-invasive ML model for predicting gallbladder polyp malignancy, demonstrating excellent performance in our validation cohort. Our key finding is the robust predictive capability of a Combined Model integrating readily available clinical data, ultrasound features, and serological markers, outperforming models relying on single data modalities. The Extra Trees algorithm emerged as the optimal ML approach, achieving a high AUC of 0.97 in training and robust performance in validation. Furthermore, SHAP analysis provided valuable insights into the key predictors driving model performance, identifying polyp diameter, stalk morphology, NLR, age, and CA19-9 as the most influential variables. Finally, the development of a user-friendly web-based calculator represents a significant step towards clinical translation of our model.\u003c/p\u003e \u003cp\u003eConsistent with established literature, polyp diameter emerged as a dominant predictor of malignancy in our model. Larger polyp size has been repeatedly shown to correlate with increased risk of gallbladder cancer\u003csup\u003e[21, 22]\u003c/sup\u003e. This is biologically plausible as larger polyps are more likely to harbor dysplastic or malignant transformation due to prolonged growth and increased cellular turnover. Our model effectively leveraged this well-established clinical risk factor, further validating its clinical relevance\u003csup\u003e[9]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eSessile polyp morphology, indicated by the absence of a stalk (i.e., a wide-based or sessile polyp), was also identified as a significant predictor. Sessile polyps are known to have a higher malignant potential compared to pedunculated polyps, Song\u0026rsquo;s research also found that when the base of the polyp widens, it is more likely to be malignant \u003csup\u003e[23]\u003c/sup\u003e. Wang\u0026rsquo;s study also found that pedunculated polyps had a higher rate of malignancy compared to sessile polyps\u003csup\u003e[24]\u003c/sup\u003e. This may be attributed to their broader attachment to the gallbladder wall, potentially facilitating deeper invasion and lymphatic spread in case of malignancy. The model\u0026rsquo;s ability to incorporate this morphological feature underscores its capacity to capture nuanced ultrasound characteristics beyond simple size.\u003c/p\u003e \u003cp\u003eAdvanced age was another important predictor identified by SHAP analysis. Increasing age is a well-recognized risk factor for various cancers\u003csup\u003e[25, 26]\u003c/sup\u003e, including gallbladder cancer\u003csup\u003e[27, 28]\u003c/sup\u003e. This reflects the cumulative effect of genetic mutations, environmental exposures, and declining immune surveillance over time, increasing the likelihood of malignant transformation in GBPs as well\u003csup\u003e[29, 30]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eElevated levels of CA19-9, a serological marker commonly associated with pancreatobiliary malignancies, also contributed significantly to the model\u0026rsquo;s predictive power\u003csup\u003e[31]\u003c/sup\u003e. While CA19-9 is not recommended for routine screening of gallbladder cancer, its elevation can reflect underlying malignant processes and has been shown to correlate with advanced stage and poorer prognosis in gallbladder cancer\u003csup\u003e[32]\u003c/sup\u003e. Our findings suggest that even within the context of GBPs, pre-operative CA19-9 levels can provide valuable information regarding malignancy risk, enhancing the discriminatory ability of our model.\u003c/p\u003e \u003cp\u003eFinally, a higher NLR was identified as a significant predictor. NLR, a readily available marker of systemic inflammation, has gained increasing attention as a prognostic indicator in various cancers\u003csup\u003e[12]\u003c/sup\u003e. Chronic inflammation is recognized as a key driver of carcinogenesis, and an elevated NLR, reflecting a pro-tumorigenic inflammatory microenvironment, may indicate a higher likelihood of malignancy within a GBP. The inclusion of NLR in our model highlights the potential of readily available systemic inflammatory markers to improve non-invasive risk stratification\u003csup\u003e[33, 34]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eCompared to previous studies focusing primarily on ultrasound features or clinical risk factors alone, our study\u0026rsquo;s novelty lies in the integrated approach, combining ultrasound morphology, serological markers, and clinical data within an ML framework. While some studies have explored radiomics or advanced imaging techniques for gallbladder lesion characterization\u003csup\u003e[35]\u003c/sup\u003e, our model leverages routinely collected, clinically accessible data, making it readily translatable to real-world practice. Furthermore, the use of Extra Trees, a robust ensemble learning algorithm, and the comprehensive model evaluation using ROC curves, PR curves, calibration curves, and DCA, ensures the rigor and reliability of our findings.\u003c/p\u003e \u003cp\u003eA key strength of our study is the visualization of feature importance using SHAP analysis. This not only enhances the interpretability of our \u0026ldquo;black box\u0026rdquo; ML model but also provides clinically relevant insights into the relative contribution of each predictor\u003csup\u003e[36, 37]\u003c/sup\u003e. Understanding which factors are most influential in driving the model\u0026rsquo;s predictions can improve clinician confidence in adopting and applying the model in practice.\u003c/p\u003e \u003cp\u003eMoreover, the development of a user-friendly web-based calculator is a significant step towards facilitating clinical implementation\u003csup\u003e[38]\u003c/sup\u003e. By providing a readily accessible tool for risk prediction, we aim to bridge the gap between research and clinical practice, enabling clinicians to easily utilize our model to aid in decision-making for patients with GBPs. This calculator has the potential to streamline risk assessment, reduce reliance on subjective interpretation of ultrasound features, and ultimately contribute to more personalized and evidence-based management strategies.\u003c/p\u003e \u003cp\u003eDespite these strengths, our study has limitations. The retrospective, single-center design may limit the generalizability of our findings to other populations and healthcare settings. External validation in independent, multi-center cohorts is crucial to confirm the robustness and broad applicability of our model. While our sample size of 202 patients is reasonable for a single-center study, a larger sample size in future studies could further enhance model performance and stability. Furthermore, we acknowledge that our model is based on pre-operative ultrasound and serological markers. Future research could explore the incorporation of more advanced imaging modalities, such as contrast-enhanced ultrasound or endoscopic ultrasound, and novel biomarkers to potentially further improve predictive accuracy. Finally, while SHAP analysis provides valuable insights into feature importance, further investigation into the underlying biological mechanisms linking these predictors to GBP malignancy is warranted.\u003c/p\u003e \u003cp\u003eFuture research directions should focus on external validation of our model in diverse patient populations and clinical settings. Prospective studies are needed to evaluate the clinical impact of using our model to guide management decisions for patients with GBPs, specifically assessing its ability to reduce unnecessary cholecystectomies and improve patient outcomes. Furthermore, exploring the cost-effectiveness of implementing our model in routine clinical practice is essential for its widespread adoption. Investigating the potential of combining our model with other diagnostic modalities, such as advanced imaging or molecular markers, could further refine risk stratification strategies.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eOur study demonstrates the successful development and validation of a non-invasive ML model for predicting gallbladder polyp malignancy, integrating ultrasound features and serological markers. The Combined Model, leveraging readily available clinical data and facilitated by a user-friendly web calculator, offers a promising tool to improve risk stratification and potentially optimize the management of patients with GBPs. Further validation and prospective clinical implementation studies are warranted to fully realize the clinical potential of this novel approach.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eGBPs: Gallbladder polyps; CA19-9: Carbohydrate Antigen 19-9; CEA: Carcinoembryonic Antigen; NLR: Neutrophil-to-Lymphocyte Ratio; ML: Machine Learning; RIS: Radiology Information Systems; LIS: Laboratory Information Systems; ALT: Alanine Aminotransferase; AST: Aspartate Aminotransferase; TBIL: Total Bilirubin; ALP: Alkaline Phosphatase; ADA: Adenosine Deaminase; BMI: Body Mass Index; SHAP: SHapley Additive exPlanations; SVC: Support Vector Classifier; AUC: Area Under the Receiver Operating Characteristic Curve; ROC: Receiver Operating Characteristic; PR: Precision-Recall; PPV: Positive Predictive Value; NPV: Negative Predictive Value; DCA: Decision Curve Analysis; SD: Standard Deviation; IQR: Interquartile Range; NRI: Net Reclassification Improvement; IDI: Integrated Discrimination Improvement.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki. The study protocol was approved by the Institutional Review Board of Fujian Provincial People’s Hospital. Informed consent was waived due to the retrospective nature of the study and the use of anonymized patient data. All data were handled in compliance with hospital policies and data protection regulations to ensure patient confidentiality.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets used and analyzed during the current study are available from the corresponding author upon reasonable request. The data are not publicly available due to privacy and ethical restrictions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no conflicts of interest related to this study. No financial or non-financial benefits have been received or will be received from any party related directly or indirectly to the subject of this manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo funding.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors' contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eYang Yan: Conceptualization, Methodology, Formal analysis, Data curation, Software, Writing – Original Draft, Writing – Review \u0026amp; Editing, Project administration, Funding acquisition. Tu Haibin: Methodology, Data curation, Investigation, Resources, Validation, Visualization, Writing – Review \u0026amp; Editing. \u0026nbsp;Lin Youguo: Data curation, Investigation, Resources, Validation, Writing – Review \u0026amp; Editing.Wei Jianting: Data curation, Investigation, Resources, Validation, Writing – Review \u0026amp; Editing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe sincerely thank the patients who participated in this study and whose data made this research possible. We are grateful to the medical records staff at Fujian Provincial Hospital for their assistance in data collection. We also appreciate the contributions of the sonographers and pathologists who performed the ultrasound examinations and histopathological analyses, respectively.\u003c/p\u003e"},{"header":" References","content":"\u003col\u003e\n\u003cli\u003ePineros M, Vignat J, Colombet M, Laversanne M, Ferreccio C, Heise K, Mhatre S, Koshiol J, Bray F. Global variations in gallbladder cancer incidence: What do recorded data and national estimates tell us? Int J Cancer 2025; 156(7): 1358-1368 [PMID: 39580808 DOI: 10.1002/ijc.35232]\u003c/li\u003e\n\u003cli\u003eZhang X, Xu C, Zhang H, Du X, Zhang Q, Lu M, Ma Y, Ma W. Gallbladder cancer incidence and mortality rate trends in China: analysis of data from the population-based cancer registry. BMC Public Health 2024; 24(1): 3122-3130 [PMID: 39529002 PMCID: PMC11555955 DOI: 10.1186/s12889-024-20584-9]\u003c/li\u003e\n\u003cli\u003eBhalla S, Shabbir N, Yadav K, Kumar M, Gupta N, Chaudhary S, Mithilesh, Sharma A, Agarwal P. Evaluating the Incidence of Incidental Gallbladder Carcinoma in a Tertiary Care Centre: A Retrospective Analysis in North India. Cureus 2024; 16(12): e76217-e76230 [PMID: 39867094 PMCID: PMC11757650 DOI: 10.7759/cureus.76217]\u003c/li\u003e\n\u003cli\u003eMin JH, Choi SY, Kim SH, Kim YK, Hwang JA, Cha DI, Lee JH, Baek SY, Lee JE. Should we suspect gallbladder cancer if which CT finding is observed in patients with localized gallbladder wall thickening? Eur J Radiol 2024; 176: 111505-111518 [PMID: 38796886 DOI: 10.1016/j.ejrad.2024.111505]\u003c/li\u003e\n\u003cli\u003eSon JH. Recent Updates on Management and Follow-up of Gallbladder Polyps. Korean J Gastroenterol 2023; 81(5): 197-202 [PMID: 37226819 DOI: 10.4166/kjg.2023.038]\u003c/li\u003e\n\u003cli\u003eLiu JS, Wang XL, Fang J, Wang A, Yang XM, He B, Zhu WH. Current situation of surgical treatment of gallbladder polyps and some problems that should be paid attention to. Zhonghua Yi Xue Za Zhi 2024; 104(34): 3171-3174 [PMID: 39193604 DOI: 10.3760/cma.j.cn112137-20240415-00879]\u003c/li\u003e\n\u003cli\u003eLiu H, Lu Y, Shen K, Zhou M, Mao X, Li R. Advances in the management of gallbladder polyps: establishment of predictive models and the rise of gallbladder-preserving polypectomy procedures. BMC Gastroenterol 2024; 24(1): 7-19 [PMID: 38166603 PMCID: PMC10759486 DOI: 10.1186/s12876-023-03094-7]\u003c/li\u003e\n\u003cli\u003eWang K, Xu Q, Xia L, Sun J, Shen K, Liu H, Xu L, Li R. Gallbladder polypoid lesions: Current practices and future prospects. Chin Med J (Engl) 2024; 137(14): 1674-1683 [PMID: 38420780 PMCID: PMC11268823 DOI: 10.1097/CM9.0000000000003019]\u003c/li\u003e\n\u003cli\u003eJiang D, Qian Y, Gu Y, Wang R, Yu H, Wang Z, Dong H, Chen D, Chen Y, Jiang H, Li Y. Establishing a radiomics model using contrast-enhanced ultrasound for preoperative prediction of neoplastic gallbladder polyps exceeding 10 mm. Eur J Med Res 2025; 30(1): 66-72 [PMID: 39901203 PMCID: PMC11789348 DOI: 10.1186/s40001-025-02292-1]\u003c/li\u003e\n\u003cli\u003eZhu L, Li N, Zhu Y, Han P, Jiang B, Li M, Luo Y, Clevert DA, Fei X. Value of high frame rate contrast enhanced ultrasound in gallbladder wall thickening in non-acute setting. Cancer Imaging 2024; 24(1): 7-19 [PMID: 38191513 PMCID: PMC10775603 DOI: 10.1186/s40644-023-00651-x]\u003c/li\u003e\n\u003cli\u003eDhingra S, Raman P, Ramsaroop T, Harrison I, Bergsten T, Nusbaum E, Feldman LE. Elevated serum CA 19-9 level mimicking pancreaticobiliary carcinoma from a hepatic abscess: case report and literature review. Front Med (Lausanne) 2024; 11: 1470046-1470058 [PMID: 39876872 PMCID: PMC11772410 DOI: 10.3389/fmed.2024.1470046]\u003c/li\u003e\n\u003cli\u003eLiu F, Yin P, Jiao B, Shi Z, Qiao F, Xu J. Detecting the preoperative peripheral blood systemic immune-inflammation index (SII) as a tool for early diagnosis and prognosis of gallbladder cancer. BMC Immunol 2025; 26(1): 7-18 [PMID: 39966731 PMCID: PMC11834489 DOI: 10.1186/s12865-025-00683-x]\u003c/li\u003e\n\u003cli\u003eHuang L, Deng X, Fan RZ, Hao TT, Zhang S, Sun B, Xu YH, Li SB, Feng YF. Coagulation and fibrinolytic markers offer utility when distinguishing between benign and malignant gallbladder tumors: A cross-sectional study. Clin Chim Acta 2024; 560: 119751-119762 [PMID: 38830523 DOI: 10.1016/j.cca.2024.119751]\u003c/li\u003e\n\u003cli\u003eHuang EY, Reeves JJ, Broderick RC, Serra JL, Goldhaber NH, An JY, Fowler KJ, Hosseini M, Sandler BJ, Jacobsen GR, Horgan S, Clary BM. Distinguishing characteristics of xanthogranulomatous cholecystitis and gallbladder adenocarcinoma: a persistent diagnostic dilemma. Surg Endosc 2024; 38(1): 348-355 [PMID: 37783778 DOI: 10.1007/s00464-023-10461-8]\u003c/li\u003e\n\u003cli\u003eRuffle JK, Gray RJ, Mohinta S, Pombo G, Kaul C, Hyare H, Rees G, Nachev P. Computational limits to the legibility of the imaged human brain. Neuroimage 2024; 291(6): 120600-120614 [PMID: 38569979 DOI: 10.1016/j.neuroimage.2024.120600]\u003c/li\u003e\n\u003cli\u003eWang LF, Wang Q, Mao F, Xu SH, Sun LP, Wu TF, Zhou BY, Yin HH, Shi H, Zhang YQ, Li XL, Sun YK, Lu D, Tang CY, Yuan HX, Zhao CK, Xu HX. Risk stratification of gallbladder masses by machine learning-based ultrasound radiomics models: a prospective and multi-institutional study. Eur Radiol 2023; 33(12): 8899-8911 [PMID: 37470825 DOI: 10.1007/s00330-023-09891-8]\u003c/li\u003e\n\u003cli\u003eYuan K, Zhang X, Yang Q, Deng X, Deng Z, Liao X, Si W. Risk prediction and analysis of gallbladder polyps with deep neural network. Comput Assist Surg (Abingdon) 2024; 29(1): 2331774-2331785 [PMID: 38520294 DOI: 10.1080/24699322.2024.2331774]\u003c/li\u003e\n\u003cli\u003eKovacs KA, Kerepesi C, Rapcsak D, Madaras L, Nagy A, Takacs A, Dank M, Szentmartoni G, Szasz AM, Kulka J, Tokes AM. Machine learning prediction of breast cancer local recurrence localization, and distant metastasis after local recurrences. Sci Rep 2025; 15(1): 4868-4875 [PMID: 39929942 PMCID: PMC11811162 DOI: 10.1038/s41598-025-89339-9]\u003c/li\u003e\n\u003cli\u003eDeng X, Liao Z. A machine-learning model based on dynamic contrast-enhanced MRI for preoperative differentiation between hepatocellular carcinoma and combined hepatocellular-cholangiocarcinoma. Clin Radiol 2024; 79(6): e817-e825 [PMID: 38413354 DOI: 10.1016/j.crad.2024.02.001]\u003c/li\u003e\n\u003cli\u003eLong X, Zeng H, Zhang Y, Lu Q, Cao Z, Shu H. Development of a Reliable GADSAH Model for Differentiating AFP-negative Hepatic Benign and Malignant Occupying Lesions. J Hepatocell Carcinoma 2024; 11(6): 607-618 [PMID: 38549786 PMCID: PMC10973090 DOI: 10.2147/JHC.S452628]\u003c/li\u003e\n\u003cli\u003eFujiwara K, Abe A, Masatsugu T, Hirano T, Sada M. Effect of gallbladder polyp size on the prediction and detection of gallbladder cancer. Surg Endosc 2021; 35(9): 5179-5185 [PMID: 32974780 DOI: 10.1007/s00464-020-08010-8]\u003c/li\u003e\n\u003cli\u003eSarici IS, Duzgun O. Gallbladder polypoid lesions \u0026gt;15mm as indicators of T1b gallbladder cancer risk. Arab J Gastroenterol 2017; 18(3): 156-158 [PMID: 28958638 DOI: 10.1016/j.ajg.2017.09.003]\u003c/li\u003e\n\u003cli\u003eLiu XS, Chen T, Gu LH, Guo YF, Li CY, Li FH, Wang J. Ultrasound-based scoring system for differential diagnosis of polypoid lesions of the gallbladder. J Gastroenterol Hepatol 2018; 33(6): 1295-1299 [PMID: 29280187 DOI: 10.1111/jgh.14080]\u003c/li\u003e\n\u003cli\u003eWang Y, Peng J, Liu K, Sun P, Ma Y, Zeng J, Jiang Y, Tan B, Cao J, Hu W. Preoperative prediction model for non-neoplastic and benign neoplastic polyps of the gallbladder. Eur J Surg Oncol 2024; 50(2): 107930 [PMID: 38159390 DOI: 10.1016/j.ejso.2023.107930]\u003c/li\u003e\n\u003cli\u003eSatoh H, Okuma Y, Shinno Y, Masuda K, Matsumoto Y, Yoshida T, Goto Y, Horinouchi H, Yamamoto N, Ohe Y. Evolving treatments and prognosis in Stage IV non-small cell lung cancer: 20 years of progress of novel therapies. Lung Cancer 2025; 202: 108453-108462 [PMID: 40020466 DOI: 10.1016/j.lungcan.2025.108453]\u003c/li\u003e\n\u003cli\u003eLi J, Zeng J, Yang Y, Huang B. Trend of skin cancer mortality and years of life lost in China from 2013 to 2021. Front Public Health 2025; 13(2): 1522790-1522801 [PMID: 40013033 PMCID: PMC11861555 DOI: 10.3389/fpubh.2025.1522790]\u003c/li\u003e\n\u003cli\u003eBozer A, Durgun N. Radiological Findings for Distinguishing Between Xanthogranulomatous Cholecystitis and Gallbladder Cancer. Arch Iran Med 2024; 27(12): 674-682 [PMID: 39891455 PMCID: PMC11786208 DOI: 10.34172/aim.31710]\u003c/li\u003e\n\u003cli\u003ePavlidis ET, Galanis IN, Pavlidis TE. Current considerations for the surgical management of gallbladder adenomas. World J Gastrointest Surg 2024; 16(6): 1507-1512 [PMID: 38983335 PMCID: PMC11229988 DOI: 10.4240/wjgs.v16.i6.1507]\u003c/li\u003e\n\u003cli\u003eYokoyama A, Kakiuchi N, Yoshizato T, Nannya Y, Suzuki H, Takeuchi Y, Shiozawa Y, Sato Y, Aoki K, Kim SK, Fujii Y, Yoshida K, Kataoka K, Nakagawa MM, Inoue Y, Hirano T, Shiraishi Y, Chiba K, Tanaka H, Sanada M, Nishikawa Y, Amanuma Y, Ohashi S, Aoyama I, Horimatsu T, Miyamoto S, Tsunoda S, Sakai Y, Narahara M, Brown JB, Sato Y, Sawada G, Mimori K, Minamiguchi S, Haga H, Seno H, Miyano S, Makishima H, Muto M, Ogawa S. Age-related remodelling of oesophageal epithelia by mutated cancer drivers. Nature 2019; 565(7739): 312-317 [PMID: 30602793 DOI: 10.1038/s41586-018-0811-x]\u003c/li\u003e\n\u003cli\u003eGuan X, Meng X, Zhong G, Zhang Z, Wang C, Xiao Y, Fu M, Zhao H, Zhou Y, Hong S, Xu X, Bai Y, Kan H, Chen R, Wu T, Guo H. Particulate matter pollution, polygenic risk score and mosaic loss of chromosome Y in middle-aged and older men from the Dongfeng-Tongji cohort study. J Hazard Mater 2024; 471: 134315 [PMID: 38678703 DOI: 10.1016/j.jhazmat.2024.134315]\u003c/li\u003e\n\u003cli\u003eKim SH, Lee MJ, Hwang HK, Lee SH, Kim H, Paik YK, Kang CM. Prognostic potential of the preoperative plasma complement factor B in resected pancreatic cancer: A pilot study. Cancer Biomark 2019; 24(3): 335-342 [PMID: 30829612 DOI: 10.3233/CBM-181847]\u003c/li\u003e\n\u003cli\u003eDou C, Han Y, Lin L, Wen J, Zhao W, Yang Y, Guan S, Li X, Gao M, Lu J. Development and Validation of a Comprehensive Risk Prediction Model for Polypoid Lesions of the Gallbladder. Clin Exp Pharmacol Physiol 2025; 52(4): e70028-e70038 [PMID: 39929712 DOI: 10.1111/1440-1681.70028]\u003c/li\u003e\n\u003cli\u003eSikora-Skrabaka M, Walkiewicz KW, Waniczek D, Strzelczyk JK, Nowakowska-Zajdel E. Relationship Between Systemic Inflammatory Response Exponents, Levels of ADAM10, ADAM17 Proteins and Selected Clinical Parameters in Patients with Colorectal Cancer: Original Research Study. Int J Mol Sci 2025; 26(3): 68-79 [PMID: 39940871 PMCID: PMC11817235 DOI: 10.3390/ijms26031104]\u003c/li\u003e\n\u003cli\u003eMaeda T, Shirakami Y, Taguchi D, Miwa T, Kubota M, Sakai H, Ibuka T, Mori K, Tomita H, Shimizu M. Glyburide Suppresses Inflammation-Related Colorectal Tumorigenesis Through Inhibition of NLRP3 Inflammasome. Int J Mol Sci 2024; 25(21): 68-79 [PMID: 39519191 PMCID: PMC11546087 DOI: 10.3390/ijms252111640]\u003c/li\u003e\n\u003cli\u003eWang Y, Qu C, Zeng J, Jiang Y, Sun R, Li C, Li J, Xing C, Tan B, Liu K, Liu Q, Zhao D, Cao J, Hu W. Establishing a preoperative predictive model for gallbladder adenoma and cholesterol polyps based on machine learning: a multicentre retrospective study. World J Surg Oncol 2025; 23(1): 27-38 [PMID: 39875897 PMCID: PMC11773841 DOI: 10.1186/s12957-025-03671-y]\u003c/li\u003e\n\u003cli\u003eFan Z, Jiang J, Xiao C, Chen Y, Xia Q, Wang J, Fang M, Wu Z, Chen F. Construction and validation of prognostic models in critically Ill patients with sepsis-associated acute kidney injury: interpretable machine learning approach. J Transl Med 2023; 21(1): 406-422 [PMID: 37349774 PMCID: PMC10286378 DOI: 10.1186/s12967-023-04205-4]\u003c/li\u003e\n\u003cli\u003eJiang A, Li J, Wang L, Zha W, Lin Y, Zhao J, Fang Z, Shen G. Multi-feature, Chinese-Western medicine-integrated prediction model for diabetic peripheral neuropathy based on machine learning and SHAP. Diabetes Metab Res Rev 2024; 40(4): e3801-e3815 [PMID: 38616511 DOI: 10.1002/dmrr.3801]\u003c/li\u003e\n\u003cli\u003eBrenner T, Kuo A, Sperna Weiland CJ, Kamal A, Elmunzer BJ, Luo H, Buxbaum J, Gardner TB, Mok SS, Fogel ES, Phillip V, Choi JH, Lua GW, Lin CC, Reddy DN, Lakhtakia S, Goenka MK, Kochhar R, Khashab MA, van Geenen EJM, Singh VK, Tomasetti C, Akshintala VS. Development and validation of a machine learning-based, point-of-care risk calculator for post-ERCP pancreatitis and prophylaxis selection. Gastrointest Endosc 2025; 101(1): 129-138 [PMID: 39147103 DOI: 10.1016/j.gie.2024.08.009]\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1 Basic information of all patients\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"984\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eVariable\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003eTotal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003eTraining\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003eStatistic\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003eP_Value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eGallbladder polyps\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e152 (75.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e107 (75.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e45 (75%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eGallbladder cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e50 (24.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e35 (24.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e15 (25%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003efemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e92 (45.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e67 (47.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e25 (41.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.57\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003emale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e110 (54.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e75 (52.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e35 (58.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003estalk(sessile)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e112 (55.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e85 (59.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e27 (45%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e3.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.07\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003estalk(pedunculated)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e90 (44.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e57 (40.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e33 (55%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eWaist_Hip_Ratio(Normal)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e95 (47%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e67 (47.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e28 (46.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eWaist_Hip_Ratio(fatty)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e107 (53%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e75 (52.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e32 (53.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eHypertension(no)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e152 (75.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e104 (73.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e48 (80%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eHypertension(yes)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e50 (24.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e38 (26.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e12 (20%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eDiabetes(no)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e173 (85.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e124 (87.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e49 (81.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.41\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eDiabetes(yes)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e29 (14.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e18 (12.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e11 (18.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eHyperlipidemia(no)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e157 (77.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e115 (81%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e42 (70%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e2.34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.13\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eHyperlipidemia(yes)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e45 (22.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e27 (19%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e18 (30%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eAlcohol history(no)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e192 (95%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e134 (94.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e58 (96.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.74\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eAlcohol history(yes)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e10 (5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e8 (5.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e2 (3.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003ePolyp_Location(bottom)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e79 (39.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e54 (38%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e25 (41.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003ePolyp_Location(body)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e74 (36.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e54 (38%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e20 (33.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003ePolyp_Location(neck)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e49 (24.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e34 (23.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e15 (25%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003ePolyp_Number(single)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e100 (49.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e68 (47.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e32 (53.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003ePolyp_Number(multiple)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e102 (50.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e74 (52.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e28 (46.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eEcho_Homogeneity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e75 (37.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e52 (36.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e23 (38.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eEcho_Heterogeneous\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e127 (62.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e90 (63.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e37 (61.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003ePolyp Surface Smoothness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e170 (84.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e122 (85.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e48 (80%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003ePolyp Surface Coarse\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e32 (15.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e20 (14.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e12 (20%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eHyperechoic_Foci(no)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e101 (50%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e77 (54.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e24 (40%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e2.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.09\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eHyperechoic_Foci(yes)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e101 (50%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e65 (45.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e36 (60%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eGallstones(no)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e61 (30.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e43 (30.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e18 (30%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eGallstones(yes)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e141 (69.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e99 (69.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e42 (70%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eGallbladder_Smooth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e155 (76.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e108 (76.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e47 (78.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eGallbladder_Roughness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e47 (23.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e34 (23.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e13 (21.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eViral_Hepatitis(no)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e81 (40.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e59 (41.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e22 (36.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eViral_Hepatitis(yes)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e121 (59.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e83 (58.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e38 (63.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eage(years)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e59.9 \u0026plusmn; 15.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e59.45 \u0026plusmn; 15.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e60.97 \u0026plusmn; 15.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003ediameter(mm)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e13.99 \u0026plusmn; 3.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e13.75 \u0026plusmn; 3.43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e14.55 \u0026plusmn; 2.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e2.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.11\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eNLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e2.97 \u0026plusmn; 1.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e3.03 \u0026plusmn; 1.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e2.82 \u0026plusmn; 1.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.42\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eBMI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e22.1 \u0026plusmn; 4.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e22.1 \u0026plusmn; 3.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e22.09 \u0026plusmn; 4.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.99\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eTBIL（mg/dL）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e16.9 \u0026plusmn; 8.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e16.72 \u0026plusmn; 8.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e17.33 \u0026plusmn; 8.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.59\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eADA（IU/L）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e12.1 \u0026plusmn; 8.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e12.58 \u0026plusmn; 8.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e10.98 \u0026plusmn; 7.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e2.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.14\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eCEA（ng/mL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e4.07 \u0026plusmn; 3.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e4.21 \u0026plusmn; 3.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e3.74 \u0026plusmn; 3.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.38\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eCA199（U/mL）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e24.24 \u0026plusmn; 20.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e23.54 \u0026plusmn; 19.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e25.89 \u0026plusmn; 20.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.33\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eALP（IU/L）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e104.67 \u0026plusmn; 50.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e106.18 \u0026plusmn; 52.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e101.1 \u0026plusmn; 43.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.96\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eAST（IU/L）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e29.74 \u0026plusmn; 16.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e28.88 \u0026plusmn; 15.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e31.78 \u0026plusmn; 17.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e1.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.21\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 249px;\"\u003e\n \u003cp\u003eALT（IU/L）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 273px;\"\u003e\n \u003cp\u003e34.82 \u0026plusmn; 21.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 178px;\"\u003e\n \u003cp\u003e33.48 \u0026plusmn; 21.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 129px;\"\u003e\n \u003cp\u003e37.98 \u0026plusmn; 21.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 52px;\"\u003e\n \u003cp\u003e2.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eNLR: Neutrophil-to-Lymphocyte Ratio; BMI: Body Mass Index; TBIL: Total Bilirubin; ADA: Adenosine Deaminase; CEA: Carcinoembryonic Antigen; CA199: Carbohydrate Antigen 19-9; ALP: Alkaline Phosphatase; AST: Aspartate Aminotransferase; ALT: Alanine Aminotransferase.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"993\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"9\" style=\"width: 993px;\"\u003e\n \u003cp\u003eTable 2 Comparison of Predictive Performance of Different Machine learning\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMean AUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAUC 95% CI Lower\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAUC 95% CI Upper\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMean Accuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eNPV\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eLogistic Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eGradient Boosting\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSVC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eDecision Tree\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eK-Nearest Neighbors\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eNaive Bayes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eLightGBM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.77\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eExtra Trees\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eAdaBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"973\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"10\" style=\"width: 973px;\"\u003e\n \u003cp\u003eTable 3 Comparison of Predictive Performance of Different Models\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eDataset\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAUC 95% CI Lower\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAUC 95% CI Upper\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eNPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\"\u003e\n \u003cp\u003eTrain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eCombined Model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eClinical Model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHematological Model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eCombined Model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eClinical Model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHematological Model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.74\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eAUC: Area Under the Curve; CI: Confidence Interval; PPV: Positive Predictive Value; NPV: Negative Predictive Value.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"987\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"8\" valign=\"bottom\" style=\"width: 987px;\"\u003e\n \u003cp\u003eTable 4 Calibration Metrics and NRI/IDI of Different Models in Different Cohorts\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eDataset\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eBrier Score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eHL p-value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eCalibration Slope\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eCalibration Intercept\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eNRI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eIDI\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\"\u003e\n \u003cp\u003eTrain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eCombined\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eClinical\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHematological\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e-0.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.41\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eCombined\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eClinical\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.11\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHematological\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.29\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eHL: Hosmer-Lemeshow; NRI: Net Reclassification Improvement; IDI: Integrated Discrimination Improvement.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"gallbladder cancer, ultrasound, hematological markers, prediction, machine learning","lastPublishedDoi":"10.21203/rs.3.rs-6744318/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6744318/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground: \u003c/strong\u003eDifferentiating benign from malignant gallbladder polyps (GBPs) is critical for clinical decisions. Pathological biopsy, the gold standard, requires cholecystectomy, underscoring the need for non-invasive alternatives.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods: \u003c/strong\u003eThis retrospective study included 202 patients (50 malignant, 152 benign) who underwent cholecystectomy (2018–2024) at Fujian Provincial Hospital. Ultrasound features (polyp diameter, stalk presence), serological markers (neutrophil-to-lymphocyte ratio [NLR], CA19-9), and demographics were analyzed. Patients were split into training (70%) and validation (30%) sets. Ten machine learning (ML) algorithms were trained; the model with the highest area under the receiver operating characteristic curve (AUC) was selected. SHapley Additive exPlanations (SHAP) identified key predictors. Models were categorized as Clinical (ultrasound + age), Hematological (NLR + CA19-9), and Combined (all five variables). ROC, Precision-Recall (PR), calibration, and Decision Curve Analysis (DCA) curves were generated. A web-based calculator was developed.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults: \u003c/strong\u003eThe Extra Trees model achieved the highest AUC (0.97 in training, 0.93 in validation). SHAP analysis highlighted polyp diameter, sessile morphology, NLR, age, and CA19-9 as top predictors. The Combined Model outperformed Clinical (AUC 0.89) and Hematological (AUC 0.68) models, with balanced sensitivity (66–54%), specificity (94–93%), and accuracy (87–83%).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion: \u003c/strong\u003eThis ML model integrating ultrasound and serological markers accurately predicts GBP malignancy. The web-based calculator facilitates clinical adoption, potentially reducing unnecessary surgeries.\u003c/p\u003e","manuscriptTitle":"A Multimodal Machine Learning Model Integrating Ultrasound and Serological Biomarkers for Non-Invasive Prediction of Gallbladder Polyp Malignancy: Development, Validation, and Clinical Translation","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-25 02:30:53","doi":"10.21203/rs.3.rs-6744318/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-02-20T15:24:21+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-12T04:58:08+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-06T10:02:04+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"152293470186179555026782732328916777197","date":"2025-07-02T20:49:58+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"6024278456177793882522105716515816212","date":"2025-06-30T02:02:06+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-06-20T05:03:58+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-05-28T14:52:56+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-05-27T11:17:01+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-05-27T11:13:35+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Imaging","date":"2025-05-25T14:23:40+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"b8cd3ea8-fb60-484b-985f-6af2dabf1b7c","owner":[],"postedDate":"June 25th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-22T11:38:23+00:00","versionOfRecord":[],"versionCreatedAt":"2025-06-25 02:30:53","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6744318","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6744318","identity":"rs-6744318","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}