Preoperative Prediction of Neck Lymph Node Metastasis in Oral Squamous Cell Carcinoma Using 18 F-FDG PET/CT-based Radiomics Analysis

Preoperative Prediction of Neck Lymph Node Metastasis in Oral Squamous Cell Carcinoma Using 18 F-FDG PET/CT-based Radiomics Analysis | 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 Preoperative Prediction of Neck Lymph Node Metastasis in Oral Squamous Cell Carcinoma Using 18 F-FDG PET/CT-based Radiomics Analysis Chao Huang, Zhenying Chen, Shaoming Chen, Zenan Wu, Xiao Zhao, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7389512/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Preoperative knowledge of neck lymph node metastasis (LNM) of oral squamous cell carcinoma (OSCC) can provide valuable information for determining the necessity of adjuvant treatment and the adequacy of surgical resection, thereby facilitating pretreatment decision-making. Therefore, this study aimed to create and evaluate an 18 F-FDG PET/CT radiomics model for the preoperative prediction of neck LNM in patients with OSCC. Results This retrospective study enrolled 174 OSCC patients who underwent 18F-FDG PET/CT scans before surgery and were randomly allocated to training and test sets. The research process involved lesion segmentation, feature extraction, model construction and evaluation. The radiomics signature, comprising five selected features, significantly correlated with LNM (p < 0.001 for both training and test sets). The radiomics model outperformed the clinical model in distinguishing LNM, with AUCs (area under the receiver operating characteristic (ROC) curves) of 0.928 and 0.890 in the training and test sets, respectively, compared to 0.863 and 0.853 for the clinical model. The integrated model based on clinical factors and the radiomics signature improved AUCs to 0.937 (95%CI: 0.902, 0.966) in the training set and 0.903 (95%CI: 0.825, 0.966) in the test set, showing superior LNM prediction. The nomogram exhibited satisfactory discrimination and good calibration, and decision curve analysis confirmed its clinical value. Conclusion 18 F-FDG PET/CT-based radiomics demonstrated significant preoperative predictive power of neck LNM in OSCC patients, providing valuable insights for pretreatment management. Radiomics 18F-FDG PET/CT Lymph node metastasis Oral squamous cell carcinoma Preoperative prediction Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Background Oral squamous cell carcinoma (OSCC) ranks as the eighth most common cancer globally [ 1 ]. Neck lymph node metastasis (LNM) is a crucial independent predictor of prognosis in OSCC patients, with those having neck LNM experiencing a significant decrease in their 5-year survival rate [ 2 – 4 ]. For early-stage OSCC patients with clinically negative necks (cN0), the decision to perform elective neck lymph node dissection remains controversy [ 5 ]. Traditional treatment methods typically involve performing neck lymph node dissection on cN0 patients to prevent occult neck LNM. However, for patients who ultimately do not have LNM, this preventive treatment may lead to unnecessary surgical trauma and potential complications. Therefore, accurately diagnosing the presence of neck LNM in OSCC patients preoperatively can reduce the negative impact of unnecessary surgical resections, thereby aiding clinical decision-making. Various imaging techniques, such as computed tomography (CT), magnetic resonance imaging (MRI), and ultrasonography (US), are routinely used for preoperative assessment of OSCC patients and are critical for clinical staging. In fact, many articles have introduced these conventional imaging methods to early predict neck LN status in OSCC. Van den Brekel et al . [ 6 ] compared the performance of US, CT, and MRI in 88 cN0 OSCC patients. The sensitivity, specificity, and accuracy for US were 58%, 75%, and 68%; for CT were 49%, 78%, and 66%; and for MRI were 55%, 88%, and 75%. However, the diagnostic efficacy of these imaging modalities for detecting neck LNM in OSCC patients needs improvement. In recent years, radiomics has transformed medical imaging into high-dimensional extractable data through the high-throughput extraction of quantitative features, which are subsequently analyzed to support clinical decision-making [ 7 , 8 ]. Several studies have demonstrated that CT or MRI-based radiomics can serve as a non-invasive preoperative diagnostic tool for predicting neck LNM in OSCC patients [ 9 – 17 ]. Wang et al . [ 11 ] reported the highest AUC value, which was 0.995, whereas Kudoh [ 15 ] reported the lowest value of 0.79. The sensitivity and specificity are over 0.65 and 0.70, respectively, and some of these values are even above 0.90. To the best of our knowledge, research on diagnosing LNM in OSCC patients using functional imaging-based radiomics remains limited. This study aims to develop and validate a radiomics model based on 18 F-FDG PET/CT for predicting neck LNM in OSCC patients. Methods Patients This study retrospectively included OSCC patients who underwent pretreatment 18 F-FDG PET/CT scans at our institution between August 2018 and March 2023. The inclusion criteria were as follows: (1) pathologically confirmed OSCC; (2) underwent surgery and lymph node dissection with complete clinical data; and (3) underwent 18 F-FDG PET/CT within two weeks prior to surgery. The exclusion criteria were as follows: (1) receipt of other treatments before surgery; (2) target lesion identified as a recurrent lesion; and (3) target lesion identified as a metastasis secondary to other malignant tumors. Clinical data, conventional imaging findings, 18 F-FDG PET/CT data, and pathological information of the patients were collected for subsequent analysis. The LNM status (positive or negative) was defined by postoperative pathology after lymph node dissection. Therefore, a total of 174 OSCC patients were enrolled in this study (125 males and 49 females; mean age, 60.84 ± 11.50 years, ranging from 25–84 years), including 86 LNM (+) and 88 LNM (-) patients. These patients were randomly divided into a training set and an independent test set at a 7:3 ratio, with balanced proportions of LNM- and LNM + considered in the random allocation principle. The patient selection procedure is depicted in Fig. 1 . PET/CT imaging A PET/CT scan was performed via PET/CT scanners (Biograph mCT64 PET/CT, Siemens Healthineers, Germany; or GE Discovery MI Gen 2 PET/CT, GE HealthCare, USA). Patients fasted for at least 4 hours, and their blood glucose levels were managed to remain below 11.1 mmol/L prior to the administration of 18 F-FDG. Approximately 60 ± 5 minutes after an intravenous injection of 3.70–4.44 MBq/kg 18F-FDG, a PET/CT scan was performed using a digital detector scanner. The PET/CT scanning protocol and data collection adhered to the European Association of Nuclear Medicine (EANM) guidelines (version 2.0) to ensure the comparability of PET/CT data across different scanners [ 18 ]. All patients underwent a whole-body (WB) PET/CT scan from the head to the upper thighs, followed by a local PET/CT scan (because of inaccurate matching of PET and CT due to head movement in some patients during WB scan) from the skull base to the neck base. CT scanning was initially performed using a low-dose protocol (120 kV; 70–120 mAs for Siemens Healthineers scanner; 30–180 mAs for GE HealthCare scanner with automated dose modulation; slice thickness: 3.75 mm; matrix: 512 × 512). PET scans were performed immediately after the CT scan using a 3D acquisition mode (matrix: 200×200), with 6–8 bed positions, each lasting 2 minutes. The PET data were reconstructed using Ordered Subsets Expectation Maximization (2 iterations, 21 subsets) in Siemens Healthineers PET/CT or Q.Clear (β = 450) in GE HealthCare PET/CT, with CT used as an attenuation correction reference. PET/CT images were analyzed by two experienced nuclear medicine physicians. Regions exhibiting abnormal 18 F-FDG uptake on PET and/or abnormal density on CT were identified as target lesions. Volume of interest segmentation and Image preprocessing Two experienced nuclear medicine physicians analyzed the local 18 F-FDG PET/CT scan images and identified target primary tumors based on abnormal 18 F-FDG uptake and/or abnormal density on CT. Image preprocessing 18 F-FDG PET/CT was performed using 3D Slicer (version 5.2.2, www.slicer.com ). Initially, CT images were fixed and PET images were aligned using the rigid transformation method available in the Elastix package within 3D Slicer. Second, a semiautomatic PET lesion volume segmentation method was employed to segment OSCC primary tumors by automatically specifying the center position of the lesions for volume of interest (VOI) segmentation [ 19 ]. To ensure reproducibility, the delineation of the VOI was carried out based on a consensus reached between two nuclear medicine physicians, and a final decision was made. The VOIs segmented from PET images were transferred to CT images to obtain the CT-based VOIs of the lesions. Manual adjustments were made to VOIs of lesions with segmentation errors. For lesions with low or no 18 F-FDG uptake, the VOIs were manually delineated on CT images and then mapped onto PET images to obtain the PET-based VOIs. 112 PET and 112 CT radiomic features were automatically extracted from the VOIs using the PyRadiomics package (version 3.1.0). Feature extraction, selection and radiomics signature construction Texture features of OSCC lesions were extracted using the PyRadiomics package (version 3.1.0) in Python (version 3.9.10). A total of 112 radiomic features were extracted from both PET and CT images using the same VOI. The normalization procedure was performed to change the values of extracted features to a range of 0 to 1. The missing data was handled by using median filling method. The least absolute shrinkage and selection operator (LASSO) algorithm was employed to identify the most relevant features from the PET and CT images within the training set [ 20 ]. Subsequently, the remaining radiomics features were integrated into Rad-score (radiomics score, the radiomics signature) using the formula: Rad-score = α + β1X1 + β2X2 + β3X3 + ... + βnXn. Here, α represents the constant term, Xn denotes the radiomics feature value, and βn is the regression coefficient for each feature. Model development and validation Univariate analysis identified significant differences of the following clinical candidate predictors: age, gender, primary tumor diameter (PTD), depth of invasion (DOI), T stage, histologic grade, and maximum standard uptake value (SUVmax) between patients who were LNM (+) and LNM (-) in the training dataset. Variables found substantial in this analysis were included in multivariate logistic regression to identify potential LNM risk factors. Prediction models were constructed using clinicopathological risk factors (clinical model), selected radiomics features (radiomics model), and combinations of these factors with the Rad-score (integrated model). The training and test datasets evaluated the models' classification performance through sensitivity, specificity, classification accuracy, receiver operating characteristic (ROC) curves, and AUC. Subsequently, paired ROC comparisons were conducted using the Delong test. A nomogram was created to visualize the clinicopathological and radiomics features, enhancing interpretability. Calibration curves assessed the model’s fit, and the Hosmer-Lemeshow test evaluated the nomogram's calibration ability. Decision curve analysis (DCA) was performed to determine the clinical use of the nomogram by measuring net benefits across various threshold probabilities for the entire cohort. Statistical analysis Continuous variables are presented as means ± SD or medians with interquartile ranges, while categorical variables are reported as numbers (percentages). Comparisons of continuous variables were made using the unpaired Student’s t-test or Mann-Whitney U test, as appropriate. Categorical variables were compared using the χ2 test or Fisher’s exact test. Model calibration was assessed using the Hosmer-Lemeshow goodness of fit test, and ROC curves among different prediction models were compared using the Delong test. Statistical analyses were conducted using SPSS (version 27.0) and R software (version 3.5.1). A two-tailed P value < 0.05 was considered statistically significant. Results Clinical characteristics According to the above criteria, 174 OSCC patients were enrolled, including 86 and 88 LNM and non-LNM patients, who were randomly allocated into 2 groups (7:3) for a training set of 121 and a independent test set of 53 (Fig. 1 ). Table 1 presents the baseline clinical characteristics of the patients in the training and test datasets. LNM positivity in the training set (49.6%, 60/121) was similar to that in the test set (49.1%, 26/53) ( p = 0.946). No statistically significant differences were observed in age ( p = 0.469), gender ( p = 0.284), PTD ( p = 0.357), DOI ( p = 0.770), T stage ( p = 0.308), histologic grade ( p = 0.142) or SUVmax ( p = 0.986) between the training and test sets. These findings suggest that the patients in the training and test sets exhibited a well-balanced distribution of baseline clinical characteristics, confirming the appropriateness of their selection for the training and test sets. Table 1 Characteristics of Patients in the Training and Test Sets Characteristic Training Set P Test Set P LNM (+) LNM (-) LNM (+) LNM (-) Age, mean ± SD, years 60.15 ± 12.85 61.36 ± 10.36 0.299 60.42 ± 12.41 59.37 ± 10.03 0.735 Gender, No. (%) 0.694 0.941 Male 43 (71.7) 41 (67.2) 20 (76.9) 21 (77.8) Female 17 (28.3) 20 (32.8) 6 (23.1) 6 (22.2) PTD, median (IQR), mm 40.00 (27.25, 55.00) 22.00 (15.00, 31.50) < 0.001 * 40.00 (29.25, 50.50) 19.00 (15.00, 29.00) < 0.001 * DOI, median (IQR), mm 10.50 (9.00, 17.00) 7.00 (4.00, 9.50) < 0.001 * 10.00 (7.75, 18.50) 8.00 (7.00, 10.00) < 0.023 * T stage < 0.001 * < 0.001 * T1 1 (1.7) 15 (24.6) 1 (3.8) 4 (14.8) T2 10 (16.7) 27 (44.3) 5 (19.2) 19 (70.4) T3 22 (36.7) 14 (23.0) 10 (38.5) 2 (7.4) T4 27 (45.0) 5 (8.2) 10 (38.5) 2 (7.4) Histologic grade, No. (%) 0.040 * 0.391 Poorly differentiated 10 (16.7) 7 (11.5) 2 (7.7) 1 (3.7) Moderately differentiated 33 (55.0) 23 (37.7) 13 (50.0) 9 (33.3) Well differentiated 17 (28.3) 31 (50.8) 11 (42.3) 17 (63.0) SUVmax, median (IQR) 14.85 (10.63, 17.25) 9.40 (6.20, 12.85) < 0.001 * 14.10 (11.58, 17.88) 9.5 (6.70, 13.60) 0.002 * Rad-score, median (IQR) 1.99 (0.80, 2.65) -2.93 (-4.55, -0.28) < 0.001 * 1.33 (-0.17, 2.64) -1.61 (-3.40, -0.51) < 0.001 * NOTE. SD, standard deviation; IQR, interquartile range; PTD, primary tumor diameter; DOI, depth of invasion; LNM, lymph node metastasis; SUVmax, maximum standard uptake value Feature selection and radiomics signature diagnostic validation A total of 224 radiomics texture features were extracted via PyRadiomics and reduced to 5 potential predictive features via the LASSO algorithm based on the data from 121 patients in the training set, which included 3 PET texture features (glrlm_LongRunLowGrayLevelEmphasis, Image-original_Mean, glszm_SmallAreaLowGrayLevelEmphasis) and 2 CT texture features (shape_SurfaceVolumeRatio, ngtdm_Strength). The formula for calculating the Rad score, which relies on these five radiomics features, is as follows: Rad-score = -0.48285132–0.9061895×shape_SurfaceVolumeRatio − 8.29990138×glrlm_LongRunLowGrayLevelEmphasis − 0.82567987×Image-original_Mean + 7.59501679×glszm_SmallAreaLowGrayLevelEmphasis − 0.97372374× ngtdm_Strength. A significant difference in the Rad-score was observed between the LNM-positive and LNM-negative groups in the training set ( p < 0.001), and this finding was subsequently confirmed in the test set ( p < 0.001) (Table 1 ). Specifically, patients with LNM-positive OSCC presented a higher Rad-score than did those with LNM-negative OSCC in both the training (Rad-score, 1.99 versus − 2.93) and test (Rad-score, 1.33 versus − 1.61) sets. The individual Rad-scores for each patient in both sets are presented as bar charts in Fig. 2 a and 2 b. Individualized prediction model establishment Univariate analyses were employed to investigate the associations between clinical features and the status of LNM (Table 1 ). On the basis of the univariate analysis results, PTD, DOI, SUVmax, and T stage were significantly different between the LNM-positive and LNM-negative groups in both the training and test sets ( p < 0.001), whereas age and gender were not significantly different. LNM was more frequently observed in patients with larger PTD and DOI, higher SUVmax, and advanced T stage. Histologic grade was slightly significantly different in the training set, but the opposite was observed in the test set. Binary logistic regression analyses revealed that PTD (OR, 1.040; 95% CI: 1.007, 1.073; p = 0.016), DOI (OR, 1.173; 95% CI: 1.039, 1.326; p = 0.010), and T stage (OR, 1.803; 95% CI: 0.940, 3.457; p = 0.036) were predictors of LNM and were used to construct a predictive model (clinical model) for LNM. The radiomics signature Rad-score was selected to establish a predictive model (radiomics model) for predicting LNM via logistic regression. The Rad-score, along with clinical information variables (PTD, DOI and T stage), was incorporated into a multivariate logistic regression model to create a comprehensive analysis framework (integrated model). Assessment and validation of the prediction models To assess the effectiveness of radiomics features in LNM prediction, models that were based on the radiomics signature, clinical risk factors (PTD, DOI, and T stage), and their combinations were compared (Fig. 3 ). The AUCs of the clinical, radiomics and integrated models were 0.863 (95% CI: 0.808, 0.915), 0.928 (95% CI: 0.889, 0.962), and 0.937 (95% CI: 0.902, 0.966), respectively, in the training set and 0.853 (95% CI: 0.753, 0.935), 0.890 (95% CI: 0.813, 0.952, 0.903 (95% CI: 0.825, 0.966), respectively, in the test set (Table 2 ). The radiomics and integrated models demonstrated excellent predictive performance in distinguishing LNM, significantly outperforming the clinical model ( p 0.05). The predictive capacities of the aforementioned three models, encompassing measures such as sensitivity, specificity, and accuracy, are presented in Table 2 . Table 2 Predictive performance of three models in the training and test sets Model Training Set Test Set AUC [95% CI] Sensitivity Specificity Accuracy AUC [95% CI] Sensitivity Specificity Accuracy Clinical model 0.863 [0.808, 0.915] 0.717 0.902 0.810 0.853 [0.753, 0.935] 0.654 0.889 0.774 Radiomics model 0.928 [0.889, 0.962] 0.867 0.852 0.860 0.890 [0.813, 0.952] 0.654 0.889 0.774 Integrated model 0.937 [0.902, 0.966] 0.900 0.803 0.851 0.903 [0.825, 0.966] 0.808 0.889 0.849 Performance and validation of the nomogram Considering that the integrated model, which merges the Rad-score with clinical risk factors (PTD, DOI, and T stage), demonstrated superior predictive capability for LNM, we developed a nomogram to provide personalized predictions based on the multivariate logistic analysis of the training set (Fig. 4 a). The calibration curve of the nomogram for estimating the likelihood of LNM exhibited a high level of agreement between the predicted and observed outcomes in both the training and test sets (Fig. 4 b and 4 c). The Hosmer–Lemeshow test yielded nonsignificant results for both the training set (χ2, 10.076; p = 0.260) and the test set (χ2, 5.002; p = 0.757), suggesting that there was no appreciable departure from a perfect match. Clinical use Figure 5 presents the DCA results for the clinical, radiomics, and integrated models in the training and test sets. The DCA revealed that the integrated and radiomics models achieved greater overall net benefits than did the clinical model over most of the risk threshold ranges (Fig. 5 a). Although the curves of the DCA in the test set were slightly less satisfactory, they maintained a similar trend as that observed in the training set (Fig. 5 b). In addition, the net benefits were comparable, with some overlap, between the radiomics and integrated models. Discussion This study developed a radiomics model using PET/CT images to predict LNM pre-surgery. In this investigation, the radiomics analysis of 18 F-FDG PET/CT images of the primary lesion, combined with clinical features, demonstrated high predictive accuracy (AUC, 0.937) for neck LNM in OSCC patients. Previous studies have established the usefulness of radiomics as a noninvasive tool for neck LNM prediction in patients with OSCC [ 9 – 17 ]. Wang Y et al . [ 9 ] found that a radiomics model based on MRI can precisely identify neck LNM in OSCC patients, with an AUC of 0.87. Tomita [ 10 ] claims that radiomics methods outperform traditional CT in distinguishing benign and metastatic neck lymph nodes. Moreover, PET/CT radiomics has been utilized to predict LNM in OSCC. Kudoh et al . [ 15 ] showed that the 18 F-FDG PET model was superior to the clinicopathological model in diagnosing neck LNM and predicting advanced LNM in OSCC patients, with an AUC of 0.79. While our findings were partially in line with those of Kudoh et al .[ 15 ], their research model did not incorporate clinical characteristic variables. In contrast, our study integrated clinicopathological factors to create a model with a higher AUC. In this study, the radiomics data for predicting LNM were obtained from the primary tumor lesion. The reasons are as follows: First, the primary lesion may be related to tumor biological heterogeneity and invasiveness, which may provide more information. As shown in our study, PTD and DOI were predictors of LNM in OSCC patients, which was consistent with the findings of previous studies [ 21 , 22 ]. Second, the analysis of the primary lesion is simpler and more feasible because of the clear presentation in the image and corresponding pathological results. In contrast, the lymph nodes found on the image cannot be confirmed by pathology one by one. In fact, most of the previous studies extracted radiomics features from primary tumors to predict LNM in OSCC [ 11 , 13 – 16 ]. We also observed that the radiomics model significantly outperformed the clinical model in predicting LNM. Upon attempting to create an integrated model by adding clinicopathologic risk factors into the radiomics model, we discovered improved predictive accuracy (the AUC increased to 0.937), indicating the complementarity of the clinical and radiomics signatures. This discovery could support the notion that incorporating markers representing diverse aspects is the most promising strategy to transform clinical management [ 23 ]. Therefore, we developed a nomogram that integrates both clinical risk factors (PTD, DOI, and T stage) and radiomics signature, which offers a visual representation of the prediction outcomes and serves as a user-friendly tool that facilitates personalized assessments of LNM, with satisfactory discrimination achieved (Fig. 4 ). This user-friendly scoring system enables both doctors and patients to perform preoperative, personalized assessments of the risk for LNM, conforming to the current trend toward individualized medicine [ 24 ]. The primary and ultimate argument for employing the nomogram lies in addressing the necessity of tailoring additional individual treatments or care. Nonetheless, the performance of risk prediction, discrimination, and calibration alone may not fully capture the clinical implications of a specific level of discrimination or degree of miscalibration [ 25 – 27 ]. To substantiate its clinical value, we evaluated whether the inclusion of the radiomics nomogram in the decision-making process would lead to improved patient outcomes. To this end, decision curve analysis was utilized in our research. The DCA revealed that the radiomics nomogram outperformed the clinical nomogram over a wide array of plausible threshold probabilities (Fig. 5 ), suggesting that the radiomics signature provides additional significance to clinicopathological risk factors for individual prediction of LNM. Additionally, we noted that the predictive ability of the radiomics model was nearly on par with that of the integrated model (AUC, 0.928 versus 0.937). This revelation implies that when certain clinical data are absent or difficult to obtain in actual clinical practice, relying solely on the radiomics model can still be effective for accurately forecasting LNM. Our study acknowledges several limitations. First, this study explored only the predictive performance of the model for OSCC; however, whether this model is applicable to other types of head and neck carcinoma requires further investigation. Second, most 18 F-FDG PET/CT radiomics research is based on single-center, small-sample retrospective designs. Therefore, large-sample, multicenter, randomized controlled prospective studies are needed to validate the models’ robustness and reproducibility, including potential bootstrapping and external validation, to confirm our preliminary findings. Third, the lack of a validation set to prevent model overfitting during training, coupled with the sole reliance on an independent test set for evaluation, makes the evaluation results less convincing. Conclusion A radiomics model based on 18 F-FDG PET/CT was developed to predict neck LNM in OSCC patients. Integrated model using radiomic features and clinical information improved the predictive accuracy for neck LNM, providing valuable information for pre-treatment management. Abbreviations OSCC Oral squamous cell carcinoma LNM Lymph node metastasis LN Lymph node CT Computed tomography MRI Magnetic resonance imaging 18 F-FDG 18 F-fluoro2-deoxy-d-glucose PET/CT Positron emission tomography/computed tomography VOI Volume of interest EANM European Association of Nuclear Medicine LASSO Least absolute shrinkage and selection operator PTD Primary tumor diameter DOI Depth of invasion SUVmax Maximum standard uptake value SD Standard deviation IQR Interquartile range ROC Receiver operating characteristic AUC Area under the receiver operating characteristic curve DCA Decision curve analysis. Declarations Ethics approval and consent to participate The study was approved by the Medical Research Ethics Committee of the First Affiliated Hospital of Fujian Medical University (No. ECFAH of FMU [2025]580). Informed consent was waived due to the retrospective nature of the study. Consent for publication Not applicable. Availability of data and material The datasets generated and analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request. Competing interests The authors declare that they have no competing interests. Funding This work was supported in part by Natural Science Foundation of Fujian Province (No.2023J01585), Scientific Research Project from the Education Department of Fujian Province (No.JAT210097), Joint Funds for the Innovation of Science and Technology, Fujian Province (No.2021Y9134) and Fujian Provincial Clinical Key Specialty Construction Project (NO.2023SZDZK-HYXK). Authors' contributions CH and WM contributed to the study conception and design. Data collection was performed by CH, ZC and SC. CH and ZC analyzed the data. Radiopharmaceutical preparation and PET/CT scan were conducted respectively by XZ and ZW. The first draft of the manuscript was written by CH and revised by WM. All authors read and approved the final manuscript. Acknowledgments The authors thank Dr. Zhoushe Zhao for his training and technical support in the process of radiomics analysis, thank Dr. Xing Wang and Jie Gao for their strongly support in data statistics and analysis, and also thank Dr. Zihao Liu and Jianyuan Zhang for their patient instruction and revision in writing and submitting of this paper. In addition, Chao Huang wants to thank Qin Xu for her constant care, support all the way and the English polishing for this paper. References Elaiwy O, El Ansari W, AlKhalil M, Ammar A. Epidemiology and pathology of oral squamous cell carcinoma in a multi-ethnic population: Retrospective study of 154 cases over 7 years in Qatar. Ann Med Surg (Lond). 2020;60:195-200. Zanoni DK, Montero PH, Migliacci JC, Shah JP, Wong RJ, Ganly I, et al. Survival outcomes after treatment of cancer of the oral cavity (1985-2015). Oral Oncol. 2019;90:115-21. Mamic M, Luksic I. Lymph node characteristics and their prognostic significance in oral squamous cell carcinoma. Head Neck. 2021;43(8):2554-5. Matos LL, Guimarães Y, Leite AK, Cernea CR. Management of Stage III Oral Cavity Squamous Cell Carcinoma in Light of the New Staging System: a Critical Review. Curr Oncol Rep. 2023;25(2):107-13. de Bree R, Takes RP, Shah JP, Hamoir M, Kowalski LP, Robbins KT, et al. Elective neck dissection in oral squamous cell carcinoma: Past, present and future. Oral Oncol. 2019;90:87-93. van den Brekel MW, Castelijns JA, Stel HV, Golding RP, Meyer CJ, Snow GB. Modern imaging techniques and ultrasound-guided aspiration cytology for the assessment of neck node metastases: a prospective comparative study. Eur Arch Otorhinolaryngol. 1993;250(1):11-7. Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun. 2014;5:4006. Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology. 2016;278(2):563-77. Wang Y, Yu T, Yang Z, Zhou Y, Kang Z, Wang Y, et al. Radiomics based on magnetic resonance imaging for preoperative prediction of lymph node metastasis in head and neck cancer: Machine learning study. Head Neck. 2022;44(12):2786-95. Tomita H, Yamashiro T, Heianna J, Nakasone T, Kimura Y, Mimura H, et al. Nodal-based radiomics analysis for identifying cervical lymph node metastasis at levels I and II in patients with oral squamous cell carcinoma using contrast-enhanced computed tomography. Eur Radiol. 2021;31(10):7440-9. Wang F, Tan R, Feng K, Hu J, Zhuang Z, Wang C, et al. Magnetic Resonance Imaging-Based Radiomics Features Associated with Depth of Invasion Predicted Lymph Node Metastasis and Prognosis in Tongue Cancer. J Magn Reson Imaging. 2022;56(1):196-209. Kubo K, Kawahara D, Murakami Y, Takeuchi Y, Katsuta T, Imano N, et al. Development of a radiomics and machine learning model for predicting occult cervical lymph node metastasis in patients with tongue cancer. Oral Surg Oral Med Oral Pathol Oral Radiol. 2022;134(1):93-101. Zhong YW, Jiang Y, Dong S, Wu WJ, Wang LX, Zhang J, et al. Tumor radiomics signature for artificial neural network-assisted detection of neck metastasis in patient with tongue cancer. J Neuroradiol. 2022;49(2):213-8. Committeri U, Fusco R, Di Bernardo E, Abbate V, Salzano G, Maglitto F, et al. Radiomics Metrics Combined with Clinical Data in the Surgical Management of Early-Stage (cT1-T2 N0) Tongue Squamous Cell Carcinomas: A Preliminary Study. Biology (Basel). 2022;11(3). Kudoh T, Haga A, Kudoh K, Takahashi A, Sasaki M, Kudo Y, et al. Radiomics analysis of [(18)F]-fluoro-2-deoxyglucose positron emission tomography for the prediction of cervical lymph node metastasis in tongue squamous cell carcinoma. Oral Radiol. 2023;39(1):41-50. Ren J, Yuan Y, Tao X. Histogram analysis of diffusion-weighted imaging and dynamic contrast-enhanced MRI for predicting occult lymph node metastasis in early-stage oral tongue squamous cell carcinoma. Eur Radiol. 2022;32(4):2739-47. Ariji Y, Fukuda M, Kise Y, Nozawa M, Yanashita Y, Fujita H, et al. Contrast-enhanced computed tomography image assessment of cervical lymph node metastasis in patients with oral cancer by using a deep learning system of artificial intelligence. Oral Surg Oral Med Oral Pathol Oral Radiol. 2019;127(5):458-63. Boellaard R, Delgado-Bolton R, Oyen WJ, Giammarile F, Tatsch K, Eschner W, et al. FDG PET/CT: EANM procedure guidelines for tumour imaging: version 2.0. Eur J Nucl Med Mol Imaging. 2015;42(2):328-54. Beichel RR, Van Tol M, Ulrich EJ, Bauer C, Chang T, Plichta KA, et al. Semiautomated segmentation of head and neck cancers in 18F-FDG PET scans: A just-enough-interaction approach. Med Phys. 2016;43(6):2948-64. Zhang J, Zhao X, Zhao Y, Zhang J, Zhang Z, Wang J, et al. Value of pre-therapy (18)F-FDG PET/CT radiomics in predicting EGFR mutation status in patients with non-small cell lung cancer. Eur J Nucl Med Mol Imaging. 2020;47(5):1137-46. Tam S, Amit M, Zafereo M, Bell D, Weber RS. Depth of invasion as a predictor of nodal disease and survival in patients with oral tongue squamous cell carcinoma. Head Neck. 2019;41(1):177-84. Tarsitano A, Del Corso G, Tardio ML, Marchetti C. Tumor Infiltration Depth as Predictor of Nodal Metastasis in Early Tongue Squamous Cell Carcinoma. J Oral Maxillofac Surg. 2016;74(3):523-7. Birkhahn M, Mitra AP, Cote RJ. Molecular markers for bladder cancer: the road to a multimarker approach. Expert Rev Anticancer Ther. 2007;7(12):1717-27. Balachandran VP, Gonen M, Smith JJ, DeMatteo RP. Nomograms in oncology: more than meets the eye. Lancet Oncol. 2015;16(4):e173-80. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350:g7594. Localio AR, Goodman S. Beyond the usual prediction accuracy metrics: reporting results for clinical decision making. Ann Intern Med. 2012;157(4):294-5. Van Calster B, Vickers AJ. Calibration of risk prediction models: impact on decision-analytic performance. Med Decis Making. 2015;35(2):162-9. Cite Share Download PDF Status: Posted Version 1 posted 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-7389512","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":504136174,"identity":"eafffb3c-2b07-4f8e-8a03-9f6c4a38602d","order_by":0,"name":"Chao Huang","email":"","orcid":"","institution":"First Affiliated Hospital of Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Chao","middleName":"","lastName":"Huang","suffix":""},{"id":504136175,"identity":"b2271cde-442e-4ffb-9ae7-5f4eb485b7c6","order_by":1,"name":"Zhenying Chen","email":"","orcid":"","institution":"First Affiliated Hospital of Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Zhenying","middleName":"","lastName":"Chen","suffix":""},{"id":504136176,"identity":"66646d44-f5aa-40e9-bc8d-347bffb511f2","order_by":2,"name":"Shaoming Chen","email":"","orcid":"","institution":"First Affiliated Hospital of Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shaoming","middleName":"","lastName":"Chen","suffix":""},{"id":504136177,"identity":"e921c712-b032-4118-9a6b-6e7a83926c3a","order_by":3,"name":"Zenan Wu","email":"","orcid":"","institution":"First Affiliated Hospital of Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Zenan","middleName":"","lastName":"Wu","suffix":""},{"id":504136178,"identity":"41acf0c9-916e-4004-a082-a432101ce6db","order_by":4,"name":"Xiao Zhao","email":"","orcid":"","institution":"First Affiliated Hospital of Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Xiao","middleName":"","lastName":"Zhao","suffix":""},{"id":504136179,"identity":"466cd9c5-fdb8-44a1-9f08-1a23a5b8cb5e","order_by":5,"name":"Weibing Miao","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA0klEQVRIiWNgGAWjYDACZjApkcDGwMD4IKHChjQtzAYPzqQRb1kCELNJPmw7RFipbjvzw8c8FRZ5fNLt1yoS2A4w8Ld3J+DVYnaYzdiY54xEMZvMmbIbCTx3GCTOnN1AQAuDmXRum0Rim0RO2o0EiWcMBhK5hLSwf5PO/QfRUpBgcJgYLTxAWxpAWtKPMSQkEKel2PjPMaBfJHKYJRIOpPEQ9sv54xsfzqipy5Ofkf7w489/NnL87b34tSABHgMwSaxyEGB/QIrqUTAKRsEoGEEAAAD7Rso8q9tQAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0002-0058-4953","institution":"First Affiliated Hospital of Fujian Medical University","correspondingAuthor":true,"prefix":"","firstName":"Weibing","middleName":"","lastName":"Miao","suffix":""}],"badges":[],"createdAt":"2025-08-16 21:31:02","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7389512/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7389512/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":90364974,"identity":"b7dada73-070b-450a-a832-64e71b23350f","added_by":"auto","created_at":"2025-09-02 02:23:22","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":298341,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of patient selection\u003c/p\u003e","description":"","filename":"Fig1.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7389512/v1/e938e48cff4a94a9af44921c.jpg"},{"id":90364344,"identity":"188dc451-228f-4428-9dbb-3676499f25ac","added_by":"auto","created_at":"2025-09-02 02:15:21","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":150559,"visible":true,"origin":"","legend":"\u003cp\u003eRad-scores for each patient in the training and test sets\u003c/p\u003e","description":"","filename":"Fig2.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7389512/v1/0d1802fbd9c206c728aff6bc.jpg"},{"id":90365821,"identity":"9c2c9e6f-629e-4279-bd70-bf52702c9a60","added_by":"auto","created_at":"2025-09-02 02:31:22","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1158235,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves for different models in differentiating LNM (a, training set; b, test set)\u003c/p\u003e","description":"","filename":"Fig3.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7389512/v1/32ec89c69ddab043757309b9.jpg"},{"id":90364360,"identity":"3da95ef8-e24b-4985-9844-2765be7b2676","added_by":"auto","created_at":"2025-09-02 02:15:22","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":2445579,"visible":true,"origin":"","legend":"\u003cp\u003eDevelopment and performance of a nomogram. a, Nomogram based on the rad-score and clinical factors. Calibration curves of the nomogram in the training (b) and test sets (c)\u003c/p\u003e","description":"","filename":"Fig4.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7389512/v1/c71a286a12ec96614e8c006e.jpg"},{"id":90364346,"identity":"5f183675-30f5-419d-b2fb-7049ecb20e1e","added_by":"auto","created_at":"2025-09-02 02:15:22","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":743645,"visible":true,"origin":"","legend":"\u003cp\u003eDecision curve analysis for the clinical, radiomics and integrated models in the training set (a) and test set (b)\u003c/p\u003e","description":"","filename":"Fig5.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7389512/v1/1a9d6ba2532c9332b69783b5.jpg"},{"id":98429234,"identity":"13623493-3683-42da-ad5c-0ee131fafe0c","added_by":"auto","created_at":"2025-12-17 16:43:01","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5648358,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7389512/v1/193b612f-cda7-4e13-b523-93287c63f30c.pdf"}],"financialInterests":"","formattedTitle":"Preoperative Prediction of Neck Lymph Node Metastasis in Oral Squamous Cell Carcinoma Using 18 F-FDG PET/CT-based Radiomics Analysis","fulltext":[{"header":"Background","content":"\u003cp\u003eOral squamous cell carcinoma (OSCC) ranks as the eighth most common cancer globally [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Neck lymph node metastasis (LNM) is a crucial independent predictor of prognosis in OSCC patients, with those having neck LNM experiencing a significant decrease in their 5-year survival rate [\u003cspan additionalcitationids=\"CR3\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eFor early-stage OSCC patients with clinically negative necks (cN0), the decision to perform elective neck lymph node dissection remains controversy [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Traditional treatment methods typically involve performing neck lymph node dissection on cN0 patients to prevent occult neck LNM. However, for patients who ultimately do not have LNM, this preventive treatment may lead to unnecessary surgical trauma and potential complications. Therefore, accurately diagnosing the presence of neck LNM in OSCC patients preoperatively can reduce the negative impact of unnecessary surgical resections, thereby aiding clinical decision-making.\u003c/p\u003e\u003cp\u003eVarious imaging techniques, such as computed tomography (CT), magnetic resonance imaging (MRI), and ultrasonography (US), are routinely used for preoperative assessment of OSCC patients and are critical for clinical staging. In fact, many articles have introduced these conventional imaging methods to early predict neck LN status in OSCC. Van den Brekel \u003cem\u003eet al\u003c/em\u003e. [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] compared the performance of US, CT, and MRI in 88 cN0 OSCC patients. The sensitivity, specificity, and accuracy for US were 58%, 75%, and 68%; for CT were 49%, 78%, and 66%; and for MRI were 55%, 88%, and 75%. However, the diagnostic efficacy of these imaging modalities for detecting neck LNM in OSCC patients needs improvement.\u003c/p\u003e\u003cp\u003eIn recent years, radiomics has transformed medical imaging into high-dimensional extractable data through the high-throughput extraction of quantitative features, which are subsequently analyzed to support clinical decision-making [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Several studies have demonstrated that CT or MRI-based radiomics can serve as a non-invasive preoperative diagnostic tool for predicting neck LNM in OSCC patients [\u003cspan additionalcitationids=\"CR10 CR11 CR12 CR13 CR14 CR15 CR16\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Wang \u003cem\u003eet al\u003c/em\u003e. [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] reported the highest AUC value, which was 0.995, whereas Kudoh [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] reported the lowest value of 0.79. The sensitivity and specificity are over 0.65 and 0.70, respectively, and some of these values are even above 0.90. To the best of our knowledge, research on diagnosing LNM in OSCC patients using functional imaging-based radiomics remains limited. This study aims to develop and validate a radiomics model based on \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT for predicting neck LNM in OSCC patients.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003ePatients\u003c/h2\u003e\u003cp\u003eThis study retrospectively included OSCC patients who underwent pretreatment \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT scans at our institution between August 2018 and March 2023. The inclusion criteria were as follows: (1) pathologically confirmed OSCC; (2) underwent surgery and lymph node dissection with complete clinical data; and (3) underwent \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT within two weeks prior to surgery. The exclusion criteria were as follows: (1) receipt of other treatments before surgery; (2) target lesion identified as a recurrent lesion; and (3) target lesion identified as a metastasis secondary to other malignant tumors. Clinical data, conventional imaging findings, \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT data, and pathological information of the patients were collected for subsequent analysis.\u003c/p\u003e\u003cp\u003eThe LNM status (positive or negative) was defined by postoperative pathology after lymph node dissection. Therefore, a total of 174 OSCC patients were enrolled in this study (125 males and 49 females; mean age, 60.84\u0026thinsp;\u0026plusmn;\u0026thinsp;11.50 years, ranging from 25\u0026ndash;84 years), including 86 LNM (+) and 88 LNM (-) patients. These patients were randomly divided into a training set and an independent test set at a 7:3 ratio, with balanced proportions of LNM- and LNM\u0026thinsp;+\u0026thinsp;considered in the random allocation principle. The patient selection procedure is depicted 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\u003ePET/CT imaging\u003c/h3\u003e\n\u003cp\u003eA PET/CT scan was performed via PET/CT scanners (Biograph mCT64 PET/CT, Siemens Healthineers, Germany; or GE Discovery MI Gen 2 PET/CT, GE HealthCare, USA). Patients fasted for at least 4 hours, and their blood glucose levels were managed to remain below 11.1 mmol/L prior to the administration of \u003csup\u003e18\u003c/sup\u003eF-FDG. Approximately 60\u0026thinsp;\u0026plusmn;\u0026thinsp;5 minutes after an intravenous injection of 3.70\u0026ndash;4.44 MBq/kg 18F-FDG, a PET/CT scan was performed using a digital detector scanner. The PET/CT scanning protocol and data collection adhered to the European Association of Nuclear Medicine (EANM) guidelines (version 2.0) to ensure the comparability of PET/CT data across different scanners [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. All patients underwent a whole-body (WB) PET/CT scan from the head to the upper thighs, followed by a local PET/CT scan (because of inaccurate matching of PET and CT due to head movement in some patients during WB scan) from the skull base to the neck base. CT scanning was initially performed using a low-dose protocol (120 kV; 70\u0026ndash;120 mAs for Siemens Healthineers scanner; 30\u0026ndash;180 mAs for GE HealthCare scanner with automated dose modulation; slice thickness: 3.75 mm; matrix: 512 \u0026times; 512). PET scans were performed immediately after the CT scan using a 3D acquisition mode (matrix: 200\u0026times;200), with 6\u0026ndash;8 bed positions, each lasting 2 minutes. The PET data were reconstructed using Ordered Subsets Expectation Maximization (2 iterations, 21 subsets) in Siemens Healthineers PET/CT or Q.Clear (β\u0026thinsp;=\u0026thinsp;450) in GE HealthCare PET/CT, with CT used as an attenuation correction reference. PET/CT images were analyzed by two experienced nuclear medicine physicians. Regions exhibiting abnormal \u003csup\u003e18\u003c/sup\u003eF-FDG uptake on PET and/or abnormal density on CT were identified as target lesions.\u003c/p\u003e\n\u003ch3\u003eVolume of interest segmentation and Image preprocessing\u003c/h3\u003e\n\u003cp\u003eTwo experienced nuclear medicine physicians analyzed the local \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT scan images and identified target primary tumors based on abnormal \u003csup\u003e18\u003c/sup\u003eF-FDG uptake and/or abnormal density on CT. Image preprocessing \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT was performed using 3D Slicer (version 5.2.2, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ewww.slicer.com\u003c/span\u003e\u003cspan address=\"http://www.slicer.com\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Initially, CT images were fixed and PET images were aligned using the rigid transformation method available in the Elastix package within 3D Slicer. Second, a semiautomatic PET lesion volume segmentation method was employed to segment OSCC primary tumors by automatically specifying the center position of the lesions for volume of interest (VOI) segmentation [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. To ensure reproducibility, the delineation of the VOI was carried out based on a consensus reached between two nuclear medicine physicians, and a final decision was made. The VOIs segmented from PET images were transferred to CT images to obtain the CT-based VOIs of the lesions. Manual adjustments were made to VOIs of lesions with segmentation errors. For lesions with low or no \u003csup\u003e18\u003c/sup\u003eF-FDG uptake, the VOIs were manually delineated on CT images and then mapped onto PET images to obtain the PET-based VOIs. 112 PET and 112 CT radiomic features were automatically extracted from the VOIs using the PyRadiomics package (version 3.1.0).\u003c/p\u003e\n\u003ch3\u003eFeature extraction, selection and radiomics signature construction\u003c/h3\u003e\n\u003cp\u003eTexture features of OSCC lesions were extracted using the PyRadiomics package (version 3.1.0) in Python (version 3.9.10). A total of 112 radiomic features were extracted from both PET and CT images using the same VOI. The normalization procedure was performed to change the values of extracted features to a range of 0 to 1. The missing data was handled by using median filling method. The least absolute shrinkage and selection operator (LASSO) algorithm was employed to identify the most relevant features from the PET and CT images within the training set [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Subsequently, the remaining radiomics features were integrated into Rad-score (radiomics score, the radiomics signature) using the formula: Rad-score\u0026thinsp;=\u0026thinsp;α\u0026thinsp;+\u0026thinsp;β1X1\u0026thinsp;+\u0026thinsp;β2X2\u0026thinsp;+\u0026thinsp;β3X3 + ...\u0026thinsp;+\u0026thinsp;βnXn. Here, α represents the constant term, Xn denotes the radiomics feature value, and βn is the regression coefficient for each feature.\u003c/p\u003e\n\u003ch3\u003eModel development and validation\u003c/h3\u003e\n\u003cp\u003eUnivariate analysis identified significant differences of the following clinical candidate predictors: age, gender, primary tumor diameter (PTD), depth of invasion (DOI), T stage, histologic grade, and maximum standard uptake value (SUVmax) between patients who were LNM (+) and LNM (-) in the training dataset. Variables found substantial in this analysis were included in multivariate logistic regression to identify potential LNM risk factors. Prediction models were constructed using clinicopathological risk factors (clinical model), selected radiomics features (radiomics model), and combinations of these factors with the Rad-score (integrated model). The training and test datasets evaluated the models' classification performance through sensitivity, specificity, classification accuracy, receiver operating characteristic (ROC) curves, and AUC. Subsequently, paired ROC comparisons were conducted using the Delong test. A nomogram was created to visualize the clinicopathological and radiomics features, enhancing interpretability. Calibration curves assessed the model\u0026rsquo;s fit, and the Hosmer-Lemeshow test evaluated the nomogram's calibration ability. Decision curve analysis (DCA) was performed to determine the clinical use of the nomogram by measuring net benefits across various threshold probabilities for the entire cohort.\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eStatistical analysis\u003c/h2\u003e\u003cp\u003eContinuous variables are presented as means\u0026thinsp;\u0026plusmn;\u0026thinsp;SD or medians with interquartile ranges, while categorical variables are reported as numbers (percentages). Comparisons of continuous variables were made using the unpaired Student\u0026rsquo;s t-test or Mann-Whitney U test, as appropriate. Categorical variables were compared using the χ2 test or Fisher\u0026rsquo;s exact test. Model calibration was assessed using the Hosmer-Lemeshow goodness of fit test, and ROC curves among different prediction models were compared using the Delong test. Statistical analyses were conducted using SPSS (version 27.0) and R software (version 3.5.1). A two-tailed P value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003eClinical characteristics\u003c/h2\u003e\u003cp\u003eAccording to the above criteria, 174 OSCC patients were enrolled, including 86 and 88 LNM and non-LNM patients, who were randomly allocated into 2 groups (7:3) for a training set of 121 and a independent test set of 53 (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents the baseline clinical characteristics of the patients in the training and test datasets. LNM positivity in the training set (49.6%, 60/121) was similar to that in the test set (49.1%, 26/53) (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.946). No statistically significant differences were observed in age (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.469), gender (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.284), PTD (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.357), DOI (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.770), T stage (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.308), histologic grade (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.142) or SUVmax (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.986) between the training and test sets. These findings suggest that the patients in the training and test sets exhibited a well-balanced distribution of baseline clinical characteristics, confirming the appropriateness of their selection for the training and test sets.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eCharacteristics of Patients in the Training and Test Sets\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eCharacteristic\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u003cp\u003eTraining Set\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e\u003cp\u003eTest Set\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLNM (+)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLNM (-)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eLNM (+)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eLNM (-)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD, years\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e60.15\u0026thinsp;\u0026plusmn;\u0026thinsp;12.85\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e61.36\u0026thinsp;\u0026plusmn;\u0026thinsp;10.36\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.299\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e60.42\u0026thinsp;\u0026plusmn;\u0026thinsp;12.41\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e59.37\u0026thinsp;\u0026plusmn;\u0026thinsp;10.03\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.735\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGender, No. (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.694\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.941\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e43 (71.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e41 (67.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e20 (76.9)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e21 (77.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFemale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e17 (28.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e20 (32.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e6 (23.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e6 (22.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePTD, median (IQR), mm\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e40.00 (27.25, 55.00)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e22.00 (15.00, 31.50)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003csup\u003e\u003cb\u003e*\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e40.00 (29.25, 50.50)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e19.00 (15.00, 29.00)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003csup\u003e\u003cb\u003e*\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDOI, median (IQR), mm\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e10.50 (9.00, 17.00)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e7.00 (4.00, 9.50)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003csup\u003e\u003cb\u003e*\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e10.00 (7.75, 18.50)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e8.00 (7.00, 10.00)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.023\u003csup\u003e\u003cb\u003e*\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eT stage\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003csup\u003e\u003cb\u003e*\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003csup\u003e\u003cb\u003e*\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eT1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1 (1.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e15 (24.6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1 (3.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e4 (14.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eT2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e10 (16.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e27 (44.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e5 (19.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e19 (70.4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eT3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e22 (36.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e14 (23.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e10 (38.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e2 (7.4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eT4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e27 (45.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5 (8.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e10 (38.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e2 (7.4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHistologic grade, No. (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.040\u003csup\u003e\u003cb\u003e*\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.391\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePoorly differentiated\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e10 (16.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e7 (11.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e2 (7.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e1 (3.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModerately differentiated\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e33 (55.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e23 (37.7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e13 (50.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e9 (33.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWell differentiated\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e17 (28.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e31 (50.8)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e11 (42.3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e17 (63.0)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSUVmax, median (IQR)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e14.85 (10.63, 17.25)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e9.40 (6.20, 12.85)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003csup\u003e\u003cb\u003e*\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e14.10 (11.58, 17.88)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e9.5 (6.70, 13.60)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.002\u003csup\u003e\u003cb\u003e*\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRad-score, median (IQR)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.99 (0.80, 2.65)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-2.93 (-4.55, -0.28)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003csup\u003e\u003cb\u003e*\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.33 (-0.17, 2.64)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-1.61 (-3.40, -0.51)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003csup\u003e\u003cb\u003e*\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"7\"\u003eNOTE. SD, standard deviation; IQR, interquartile range; PTD, primary tumor diameter; DOI, depth of invasion; LNM, lymph node metastasis; SUVmax, maximum standard uptake value\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eFeature selection and radiomics signature diagnostic validation\u003c/h2\u003e\u003cp\u003eA total of 224 radiomics texture features were extracted via PyRadiomics and reduced to 5 potential predictive features via the LASSO algorithm based on the data from 121 patients in the training set, which included 3 PET texture features (glrlm_LongRunLowGrayLevelEmphasis, Image-original_Mean, glszm_SmallAreaLowGrayLevelEmphasis) and 2 CT texture features (shape_SurfaceVolumeRatio, ngtdm_Strength). The formula for calculating the Rad score, which relies on these five radiomics features, is as follows: Rad-score = -0.48285132\u0026ndash;0.9061895\u0026times;shape_SurfaceVolumeRatio \u0026minus;\u0026thinsp;8.29990138\u0026times;glrlm_LongRunLowGrayLevelEmphasis \u0026minus;\u0026thinsp;0.82567987\u0026times;Image-original_Mean\u0026thinsp;+\u0026thinsp;7.59501679\u0026times;glszm_SmallAreaLowGrayLevelEmphasis \u0026minus;\u0026thinsp;0.97372374\u0026times; ngtdm_Strength.\u003c/p\u003e\u003cp\u003eA significant difference in the Rad-score was observed between the LNM-positive and LNM-negative groups in the training set (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and this finding was subsequently confirmed in the test set (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Specifically, patients with LNM-positive OSCC presented a higher Rad-score than did those with LNM-negative OSCC in both the training (Rad-score, 1.99 versus \u0026minus;\u0026thinsp;2.93) and test (Rad-score, 1.33 versus \u0026minus;\u0026thinsp;1.61) sets. The individual Rad-scores for each patient in both sets are presented as bar charts in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eIndividualized prediction model establishment\u003c/h2\u003e\u003cp\u003eUnivariate analyses were employed to investigate the associations between clinical features and the status of LNM (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). On the basis of the univariate analysis results, PTD, DOI, SUVmax, and T stage were significantly different between the LNM-positive and LNM-negative groups in both the training and test sets (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), whereas age and gender were not significantly different. LNM was more frequently observed in patients with larger PTD and DOI, higher SUVmax, and advanced T stage. Histologic grade was slightly significantly different in the training set, but the opposite was observed in the test set. Binary logistic regression analyses revealed that PTD (OR, 1.040; 95% CI: 1.007, 1.073; \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.016), DOI (OR, 1.173; 95% CI: 1.039, 1.326; \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.010), and T stage (OR, 1.803; 95% CI: 0.940, 3.457; \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.036) were predictors of LNM and were used to construct a predictive model (clinical model) for LNM. The radiomics signature Rad-score was selected to establish a predictive model (radiomics model) for predicting LNM via logistic regression. The Rad-score, along with clinical information variables (PTD, DOI and T stage), was incorporated into a multivariate logistic regression model to create a comprehensive analysis framework (integrated model).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003eAssessment and validation of the prediction models\u003c/h2\u003e\u003cp\u003eTo assess the effectiveness of radiomics features in LNM prediction, models that were based on the radiomics signature, clinical risk factors (PTD, DOI, and T stage), and their combinations were compared (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The AUCs of the clinical, radiomics and integrated models were 0.863 (95% CI: 0.808, 0.915), 0.928 (95% CI: 0.889, 0.962), and 0.937 (95% CI: 0.902, 0.966), respectively, in the training set and 0.853 (95% CI: 0.753, 0.935), 0.890 (95% CI: 0.813, 0.952, 0.903 (95% CI: 0.825, 0.966), respectively, in the test set (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The radiomics and integrated models demonstrated excellent predictive performance in distinguishing LNM, significantly outperforming the clinical model (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Although the AUC of the integrated model marginally exceeded that of the radiomics model, the difference in AUC between them was not statistically significant (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05). The predictive capacities of the aforementioned three models, encompassing measures such as sensitivity, specificity, and accuracy, are presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePredictive performance of three models in the training and test sets\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"10\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"4\" nameend=\"c5\" namest=\"c2\"\u003e\u003cp\u003eTraining Set\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\" morerows=\"1\" rowspan=\"2\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colspan=\"4\" nameend=\"c10\" namest=\"c7\"\u003e\u003cp\u003eTest Set\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAUC [95% CI]\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSensitivity\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSpecificity\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eAUC [95% CI]\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c8\"\u003e\u003cp\u003eSensitivity\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c9\"\u003e\u003cp\u003eSpecificity\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c10\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eClinical model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.863 [0.808, 0.915]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.717\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.902\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.810\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.853 [0.753, 0.935]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.654\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.889\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.774\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRadiomics model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.928 [0.889, 0.962]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.867\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.852\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.860\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.890 [0.813, 0.952]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.654\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.889\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.774\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eIntegrated model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e0.937 [0.902, 0.966]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.900\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.803\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.851\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.903 [0.825, 0.966]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.808\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.889\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e\u003cp\u003e0.849\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003ePerformance and validation of the nomogram\u003c/h2\u003e\u003cp\u003eConsidering that the integrated model, which merges the Rad-score with clinical risk factors (PTD, DOI, and T stage), demonstrated superior predictive capability for LNM, we developed a nomogram to provide personalized predictions based on the multivariate logistic analysis of the training set (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea). The calibration curve of the nomogram for estimating the likelihood of LNM exhibited a high level of agreement between the predicted and observed outcomes in both the training and test sets (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb and \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ec). The Hosmer\u0026ndash;Lemeshow test yielded nonsignificant results for both the training set (χ2, 10.076; \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.260) and the test set (χ2, 5.002; \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.757), suggesting that there was no appreciable departure from a perfect match.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003eClinical use\u003c/h2\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e presents the DCA results for the clinical, radiomics, and integrated models in the training and test sets. The DCA revealed that the integrated and radiomics models achieved greater overall net benefits than did the clinical model over most of the risk threshold ranges (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea). Although the curves of the DCA in the test set were slightly less satisfactory, they maintained a similar trend as that observed in the training set (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb). In addition, the net benefits were comparable, with some overlap, between the radiomics and integrated models.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study developed a radiomics model using PET/CT images to predict LNM pre-surgery. In this investigation, the radiomics analysis of \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT images of the primary lesion, combined with clinical features, demonstrated high predictive accuracy (AUC, 0.937) for neck LNM in OSCC patients.\u003c/p\u003e\u003cp\u003ePrevious studies have established the usefulness of radiomics as a noninvasive tool for neck LNM prediction in patients with OSCC [\u003cspan additionalcitationids=\"CR10 CR11 CR12 CR13 CR14 CR15 CR16\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Wang Y \u003cem\u003eet al\u003c/em\u003e. [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] found that a radiomics model based on MRI can precisely identify neck LNM in OSCC patients, with an AUC of 0.87. Tomita [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e] claims that radiomics methods outperform traditional CT in distinguishing benign and metastatic neck lymph nodes. Moreover, PET/CT radiomics has been utilized to predict LNM in OSCC. Kudoh \u003cem\u003eet al\u003c/em\u003e. [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] showed that the \u003csup\u003e18\u003c/sup\u003eF-FDG PET model was superior to the clinicopathological model in diagnosing neck LNM and predicting advanced LNM in OSCC patients, with an AUC of 0.79. While our findings were partially in line with those of Kudoh \u003cem\u003eet al\u003c/em\u003e.[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], their research model did not incorporate clinical characteristic variables. In contrast, our study integrated clinicopathological factors to create a model with a higher AUC.\u003c/p\u003e\u003cp\u003eIn this study, the radiomics data for predicting LNM were obtained from the primary tumor lesion. The reasons are as follows: First, the primary lesion may be related to tumor biological heterogeneity and invasiveness, which may provide more information. As shown in our study, PTD and DOI were predictors of LNM in OSCC patients, which was consistent with the findings of previous studies [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Second, the analysis of the primary lesion is simpler and more feasible because of the clear presentation in the image and corresponding pathological results. In contrast, the lymph nodes found on the image cannot be confirmed by pathology one by one. In fact, most of the previous studies extracted radiomics features from primary tumors to predict LNM in OSCC [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan additionalcitationids=\"CR14 CR15\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eWe also observed that the radiomics model significantly outperformed the clinical model in predicting LNM. Upon attempting to create an integrated model by adding clinicopathologic risk factors into the radiomics model, we discovered improved predictive accuracy (the AUC increased to 0.937), indicating the complementarity of the clinical and radiomics signatures. This discovery could support the notion that incorporating markers representing diverse aspects is the most promising strategy to transform clinical management [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Therefore, we developed a nomogram that integrates both clinical risk factors (PTD, DOI, and T stage) and radiomics signature, which offers a visual representation of the prediction outcomes and serves as a user-friendly tool that facilitates personalized assessments of LNM, with satisfactory discrimination achieved (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). This user-friendly scoring system enables both doctors and patients to perform preoperative, personalized assessments of the risk for LNM, conforming to the current trend toward individualized medicine [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. The primary and ultimate argument for employing the nomogram lies in addressing the necessity of tailoring additional individual treatments or care. Nonetheless, the performance of risk prediction, discrimination, and calibration alone may not fully capture the clinical implications of a specific level of discrimination or degree of miscalibration [\u003cspan additionalcitationids=\"CR26\" citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. To substantiate its clinical value, we evaluated whether the inclusion of the radiomics nomogram in the decision-making process would lead to improved patient outcomes. To this end, decision curve analysis was utilized in our research. The DCA revealed that the radiomics nomogram outperformed the clinical nomogram over a wide array of plausible threshold probabilities (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e), suggesting that the radiomics signature provides additional significance to clinicopathological risk factors for individual prediction of LNM. Additionally, we noted that the predictive ability of the radiomics model was nearly on par with that of the integrated model (AUC, 0.928 versus 0.937). This revelation implies that when certain clinical data are absent or difficult to obtain in actual clinical practice, relying solely on the radiomics model can still be effective for accurately forecasting LNM.\u003c/p\u003e\u003cp\u003eOur study acknowledges several limitations. First, this study explored only the predictive performance of the model for OSCC; however, whether this model is applicable to other types of head and neck carcinoma requires further investigation. Second, most \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT radiomics research is based on single-center, small-sample retrospective designs. Therefore, large-sample, multicenter, randomized controlled prospective studies are needed to validate the models\u0026rsquo; robustness and reproducibility, including potential bootstrapping and external validation, to confirm our preliminary findings. Third, the lack of a validation set to prevent model overfitting during training, coupled with the sole reliance on an independent test set for evaluation, makes the evaluation results less convincing.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eA radiomics model based on \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT was developed to predict neck LNM in OSCC patients. Integrated model using radiomic features and clinical information improved the predictive accuracy for neck LNM, providing valuable information for pre-treatment management.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eOSCC \u0026nbsp; \u0026nbsp;Oral squamous cell carcinoma\u003c/p\u003e\n\u003cp\u003eLNM \u0026nbsp; \u0026nbsp; Lymph node metastasis\u003c/p\u003e\n\u003cp\u003eLN \u0026nbsp; \u0026nbsp; \u0026nbsp; Lymph node\u003c/p\u003e\n\u003cp\u003eCT \u0026nbsp; \u0026nbsp; \u0026nbsp; Computed tomography\u003c/p\u003e\n\u003cp\u003eMRI \u0026nbsp; \u0026nbsp; \u0026nbsp;Magnetic resonance imaging\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e18\u003c/sup\u003eF-FDG \u0026nbsp;\u0026nbsp;\u003csup\u003e18\u003c/sup\u003eF-fluoro2-deoxy-d-glucose\u003c/p\u003e\n\u003cp\u003ePET/CT \u0026nbsp; Positron emission tomography/computed tomography\u003c/p\u003e\n\u003cp\u003eVOI \u0026nbsp; \u0026nbsp; \u0026nbsp;Volume of interest\u003c/p\u003e\n\u003cp\u003eEANM \u0026nbsp; \u0026nbsp;European Association of Nuclear Medicine\u003c/p\u003e\n\u003cp\u003eLASSO \u0026nbsp; Least absolute shrinkage and selection operator\u003c/p\u003e\n\u003cp\u003ePTD \u0026nbsp; \u0026nbsp; \u0026nbsp;Primary tumor diameter\u003c/p\u003e\n\u003cp\u003eDOI \u0026nbsp; \u0026nbsp; \u0026nbsp;Depth of invasion\u003c/p\u003e\n\u003cp\u003eSUVmax \u0026nbsp;Maximum standard uptake value\u003c/p\u003e\n\u003cp\u003eSD \u0026nbsp; \u0026nbsp; \u0026nbsp; Standard deviation\u003c/p\u003e\n\u003cp\u003eIQR \u0026nbsp; \u0026nbsp; \u0026nbsp;Interquartile range\u003c/p\u003e\n\u003cp\u003eROC \u0026nbsp; \u0026nbsp; Receiver operating characteristic\u003c/p\u003e\n\u003cp\u003eAUC \u0026nbsp; \u0026nbsp; Area under the receiver operating characteristic curve\u003c/p\u003e\n\u003cp\u003eDCA \u0026nbsp; \u0026nbsp; Decision curve analysis.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was approved by the Medical Research Ethics Committee of the First Affiliated Hospital of Fujian Medical University (No. ECFAH of FMU [2025]580). Informed consent was waived due to the retrospective nature of the study.\u0026nbsp;\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 material\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported in part by Natural Science Foundation of Fujian Province (No.2023J01585), Scientific Research Project from the Education Department of Fujian Province (No.JAT210097), Joint Funds for the Innovation of Science and Technology, Fujian Province (No.2021Y9134) and Fujian Provincial Clinical Key Specialty Construction Project (NO.2023SZDZK-HYXK).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCH and WM contributed to the study conception and design. Data collection was performed by CH, ZC and SC. CH and ZC analyzed the data. Radiopharmaceutical preparation and PET/CT scan were conducted respectively by XZ and ZW. The first draft of the manuscript was written by CH and revised by WM. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors thank Dr. Zhoushe Zhao for his training and technical support in the process of radiomics analysis, thank Dr. Xing Wang and Jie Gao for their strongly support in data statistics and analysis, and also thank Dr. Zihao Liu and Jianyuan Zhang for their patient instruction and revision in writing and submitting of this paper. In addition, Chao Huang wants to thank Qin Xu for her constant care, support all the way and the English polishing for this paper.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eElaiwy O, El Ansari W, AlKhalil M, Ammar A. Epidemiology and pathology of oral squamous cell carcinoma in a multi-ethnic population: Retrospective study of 154 cases over 7 years in Qatar. Ann Med Surg (Lond). 2020;60:195-200.\u003c/li\u003e\n\u003cli\u003eZanoni DK, Montero PH, Migliacci JC, Shah JP, Wong RJ, Ganly I, et al. Survival outcomes after treatment of cancer of the oral cavity (1985-2015). Oral Oncol. 2019;90:115-21.\u003c/li\u003e\n\u003cli\u003eMamic M, Luksic I. Lymph node characteristics and their prognostic significance in oral squamous cell carcinoma. Head Neck. 2021;43(8):2554-5.\u003c/li\u003e\n\u003cli\u003eMatos LL, Guimar\u0026atilde;es Y, Leite AK, Cernea CR. Management of Stage III Oral Cavity Squamous Cell Carcinoma in Light of the New Staging System: a Critical Review. Curr Oncol Rep. 2023;25(2):107-13.\u003c/li\u003e\n\u003cli\u003ede Bree R, Takes RP, Shah JP, Hamoir M, Kowalski LP, Robbins KT, et al. Elective neck dissection in oral squamous cell carcinoma: Past, present and future. Oral Oncol. 2019;90:87-93.\u003c/li\u003e\n\u003cli\u003evan den Brekel MW, Castelijns JA, Stel HV, Golding RP, Meyer CJ, Snow GB. Modern imaging techniques and ultrasound-guided aspiration cytology for the assessment of neck node metastases: a prospective comparative study. Eur Arch Otorhinolaryngol. 1993;250(1):11-7.\u003c/li\u003e\n\u003cli\u003eAerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun. 2014;5:4006.\u003c/li\u003e\n\u003cli\u003eGillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology. 2016;278(2):563-77.\u003c/li\u003e\n\u003cli\u003eWang Y, Yu T, Yang Z, Zhou Y, Kang Z, Wang Y, et al. Radiomics based on magnetic resonance imaging for preoperative prediction of lymph node metastasis in head and neck cancer: Machine learning study. Head Neck. 2022;44(12):2786-95.\u003c/li\u003e\n\u003cli\u003eTomita H, Yamashiro T, Heianna J, Nakasone T, Kimura Y, Mimura H, et al. Nodal-based radiomics analysis for identifying cervical lymph node metastasis at levels I and II in patients with oral squamous cell carcinoma using contrast-enhanced computed tomography. Eur Radiol. 2021;31(10):7440-9.\u003c/li\u003e\n\u003cli\u003eWang F, Tan R, Feng K, Hu J, Zhuang Z, Wang C, et al. Magnetic Resonance Imaging-Based Radiomics Features Associated with Depth of Invasion Predicted Lymph Node Metastasis and Prognosis in Tongue Cancer. J Magn Reson Imaging. 2022;56(1):196-209.\u003c/li\u003e\n\u003cli\u003eKubo K, Kawahara D, Murakami Y, Takeuchi Y, Katsuta T, Imano N, et al. Development of a radiomics and machine learning model for predicting occult cervical lymph node metastasis in patients with tongue cancer. Oral Surg Oral Med Oral Pathol Oral Radiol. 2022;134(1):93-101.\u003c/li\u003e\n\u003cli\u003eZhong YW, Jiang Y, Dong S, Wu WJ, Wang LX, Zhang J, et al. Tumor radiomics signature for artificial neural network-assisted detection of neck metastasis in patient with tongue cancer. J Neuroradiol. 2022;49(2):213-8.\u003c/li\u003e\n\u003cli\u003eCommitteri U, Fusco R, Di Bernardo E, Abbate V, Salzano G, Maglitto F, et al. Radiomics Metrics Combined with Clinical Data in the Surgical Management of Early-Stage (cT1-T2 N0) Tongue Squamous Cell Carcinomas: A Preliminary Study. Biology (Basel). 2022;11(3).\u003c/li\u003e\n\u003cli\u003eKudoh T, Haga A, Kudoh K, Takahashi A, Sasaki M, Kudo Y, et al. Radiomics analysis of [(18)F]-fluoro-2-deoxyglucose positron emission tomography for the prediction of cervical lymph node metastasis in tongue squamous cell carcinoma. Oral Radiol. 2023;39(1):41-50.\u003c/li\u003e\n\u003cli\u003eRen J, Yuan Y, Tao X. Histogram analysis of diffusion-weighted imaging and dynamic contrast-enhanced MRI for predicting occult lymph node metastasis in early-stage oral tongue squamous cell carcinoma. Eur Radiol. 2022;32(4):2739-47.\u003c/li\u003e\n\u003cli\u003eAriji Y, Fukuda M, Kise Y, Nozawa M, Yanashita Y, Fujita H, et al. Contrast-enhanced computed tomography image assessment of cervical lymph node metastasis in patients with oral cancer by using a deep learning system of artificial intelligence. Oral Surg Oral Med Oral Pathol Oral Radiol. 2019;127(5):458-63.\u003c/li\u003e\n\u003cli\u003eBoellaard R, Delgado-Bolton R, Oyen WJ, Giammarile F, Tatsch K, Eschner W, et al. FDG PET/CT: EANM procedure guidelines for tumour imaging: version 2.0. Eur J Nucl Med Mol Imaging. 2015;42(2):328-54.\u003c/li\u003e\n\u003cli\u003eBeichel RR, Van Tol M, Ulrich EJ, Bauer C, Chang T, Plichta KA, et al. Semiautomated segmentation of head and neck cancers in 18F-FDG PET scans: A just-enough-interaction approach. Med Phys. 2016;43(6):2948-64.\u003c/li\u003e\n\u003cli\u003eZhang J, Zhao X, Zhao Y, Zhang J, Zhang Z, Wang J, et al. Value of pre-therapy (18)F-FDG PET/CT radiomics in predicting EGFR mutation status in patients with non-small cell lung cancer. Eur J Nucl Med Mol Imaging. 2020;47(5):1137-46.\u003c/li\u003e\n\u003cli\u003eTam S, Amit M, Zafereo M, Bell D, Weber RS. Depth of invasion as a predictor of nodal disease and survival in patients with oral tongue squamous cell carcinoma. Head Neck. 2019;41(1):177-84.\u003c/li\u003e\n\u003cli\u003eTarsitano A, Del Corso G, Tardio ML, Marchetti C. Tumor Infiltration Depth as Predictor of Nodal Metastasis in Early Tongue Squamous Cell Carcinoma. J Oral Maxillofac Surg. 2016;74(3):523-7.\u003c/li\u003e\n\u003cli\u003eBirkhahn M, Mitra AP, Cote RJ. Molecular markers for bladder cancer: the road to a multimarker approach. Expert Rev Anticancer Ther. 2007;7(12):1717-27.\u003c/li\u003e\n\u003cli\u003eBalachandran VP, Gonen M, Smith JJ, DeMatteo RP. Nomograms in oncology: more than meets the eye. Lancet Oncol. 2015;16(4):e173-80.\u003c/li\u003e\n\u003cli\u003eCollins GS, Reitsma JB, Altman DG, Moons KG. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350:g7594.\u003c/li\u003e\n\u003cli\u003eLocalio AR, Goodman S. Beyond the usual prediction accuracy metrics: reporting results for clinical decision making. Ann Intern Med. 2012;157(4):294-5.\u003c/li\u003e\n\u003cli\u003eVan Calster B, Vickers AJ. Calibration of risk prediction models: impact on decision-analytic performance. Med Decis Making. 2015;35(2):162-9.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Radiomics, 18F-FDG PET/CT, Lymph node metastasis, Oral squamous cell carcinoma, Preoperative prediction","lastPublishedDoi":"10.21203/rs.3.rs-7389512/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7389512/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003ePreoperative knowledge of neck lymph node metastasis (LNM) of oral squamous cell carcinoma (OSCC) can provide valuable information for determining the necessity of adjuvant treatment and the adequacy of surgical resection, thereby facilitating pretreatment decision-making. Therefore, this study aimed to create and evaluate an \u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT radiomics model for the preoperative prediction of neck LNM in patients with OSCC.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eThis retrospective study enrolled 174 OSCC patients who underwent 18F-FDG PET/CT scans before surgery and were randomly allocated to training and test sets. The research process involved lesion segmentation, feature extraction, model construction and evaluation. The radiomics signature, comprising five selected features, significantly correlated with LNM (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 for both training and test sets). The radiomics model outperformed the clinical model in distinguishing LNM, with AUCs (area under the receiver operating characteristic (ROC) curves) of 0.928 and 0.890 in the training and test sets, respectively, compared to 0.863 and 0.853 for the clinical model. The integrated model based on clinical factors and the radiomics signature improved AUCs to 0.937 (95%CI: 0.902, 0.966) in the training set and 0.903 (95%CI: 0.825, 0.966) in the test set, showing superior LNM prediction. The nomogram exhibited satisfactory discrimination and good calibration, and decision curve analysis confirmed its clinical value.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003e\u003csup\u003e18\u003c/sup\u003eF-FDG PET/CT-based radiomics demonstrated significant preoperative predictive power of neck LNM in OSCC patients, providing valuable insights for pretreatment management.\u003c/p\u003e","manuscriptTitle":"Preoperative Prediction of Neck Lymph Node Metastasis in Oral Squamous Cell Carcinoma Using 18 F-FDG PET/CT-based Radiomics Analysis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-02 02:15:17","doi":"10.21203/rs.3.rs-7389512/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"211a2145-d5f3-4996-baa5-c7d2f0c0d205","owner":[],"postedDate":"September 2nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-12-12T13:49:35+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-02 02:15:17","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7389512","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7389512","identity":"rs-7389512","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}