Morphology-based Machine-Learning for Predicting Lymph Node Status in Oral Tongue Squamous Cell Carcinoma

Morphology-based Machine-Learning for Predicting Lymph Node Status in Oral Tongue Squamous Cell Carcinoma | 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 Morphology-based Machine-Learning for Predicting Lymph Node Status in Oral Tongue Squamous Cell Carcinoma Yunjing Zhu, Jiliang Ren, Yang Song, Xiaofeng Tao, Ying Yuan This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3909740/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 Purpose To develop machine-learning models based on morphological features extracted from preoperative magnetic resonance imaging (MRI) to predict lymph node status in oral tongue squamous cell carcinoma (OTSCC). Method This study retrospectively enrolled 90 OTSCC patients, of whom 45 and 13 patients, respectively, had confirmed lymph node metastasis (LNM) and extranodal extension (ENE). Fourteen morphological features and two customized metrics were derived from T2-weighted (T2W) images. Tumor maximum diameter and MRI-derived depth of invasion (DOI) were measured on contrast-enhanced T1-weighted (ceT1W) images. Information gain algorithm was applied to select the top five attributes. Models were created using six machine-learning methods, including neural network (NN), random forest (RF), logistic regression (LR), support vector machine (SVM), naïve bayes (NB), and AdaBoost. An internal stratified 10-fold cross-validation was performed to assess their performance. Results For predicting LNM, the NN classifier, which included Situation, Elongation, Top Bottom Area, Least Axis Length, and Minor Axis Length, yielded the best model, with an AUC of 0.746 and accuracy of 72.2%. The performance of the NN model was slightly superior to that of MRI-derived DOI (0.746 vs. 0.655), although the difference was not significant ( P = 0.122). For predicting ENE, the SVM classifier, which included situation, Elongation, Top Bottom Area, Least Axis Length, and Minor Axis Length, performed the best, with an AUC of 0.750 and accuracy of 85.6%. Conclusions Machine-learning models using MRI morphological features have potential in preoperative evaluation of cervical lymph node status in OTSCC. Squamous cell carcinoma of head and neck Lymphatic metastasis Extranodal extension Magnetic resonance imaging Machine learning Figures Figure 1 Figure 2 Figure 3 Figure 4 1. Introduction Oral cancer is a very aggressive malignancy, and oral tongue squamous cell carcinoma (OTSCC) is the most common type [ 1 ]. Cervical lymph node metastasis (LNM) is one of the prognostic factors for OTSCC patients. The 5-year survival rate drops by half with the occurrence of LNM [ 2 ]. An additional drop in survival occurs when extranodal extension (ENE) is present in the metastatic lymph nodes [ 3 ]. The tumor, node, and metastasis (TNM) staging system is the principal criterion to describe and stage tumor extension [ 4 ]. In fact, the T stage index of tumor depth of invasion (DOI) is proved to be strongly associated with the N stage in OTSCC [ 5 , 6 ]. 除了DOI 还有其他形态学指标和淋巴结转移相关。Thus, preoperatively predicting LNM in OTSCC with morphological features and T stage index is theoretically feasible, which is especially crucial for patients with no clinical evidence for LNM preoperatively but was confirmed histopathologically afterwards (cN0). Magnetic resonance imaging (MRI) is recommended for the preoperative evaluation of tongue cancer and can be used as a reference to determine clinical TNM stage owing to its advantage in soft-tissue contrasting and tumor-contour depicting [ 7 ]. Traditionally, morphological appearance of lesions on MRI images are evaluated subjectively and qualitatively [ 8 ]. A more comprehensive and repeatable method for depicting morphology of the original tumor is thus required to potentially predict lymph node status in OTSCC. Radiomics can extract quantitative morphological information from conventional images objectively [ 9 ]. The potential of these quantitative shape features in distinguishing benign from malignant tumors could be emphasized. To avoid sophisticated mathematical computation of radiomics features which are limited in explanation in clinics, we chose to extract explicit morphological features, such as volume, perimeter and contour irregularity of lesions [ 10 ]. Machine learning method has promising application in radiomics given that its algorithms are suited for analysis of high-dimensional data. Previous studies have used machine-learning techniques to predict treatment response and other prognostic outcomes in tongue cancer [ 11 – 13 ]. Therefore, the aim of this study was to construct machine-learning models based on morphological radiomics features from MRI to predict lymph node status of patients with OTSCC. 2. Materials and methods 2.1 Patients This retrospective study was approved by Institutional Review Board of Shanghai Ninth People’s Hospital, and the informed consent was waived. We reviewed the medical records from June 2015 to May 2022 to identify patients with OTSCC. The inclusion criteria were as follows: (1) patients underwent preoperative contrast-enhanced head and neck MRI; (2) patients were treated with primary tumor resection and elective neck dissection; and (3) complete clinical and pathological information was available. The exclusion criteria were as follows: (1) patients previously received radiotherapy, surgery and/or chemotherapy; (2) the presence of other concurrent head and neck malignant tumor; and (3) obvious artifact impacting MRI image analysis. In total, 90 patients were included for further analysis (Fig. 1 ). The presence of pathologically involved LNM and ENE was recorded based on histopathology reports at the time of initial treatment. ENE was considered positive if metastatic carcinoma extended from within a lymph node through the fibrous capsule and into the surrounding connective tissue, regardless of the presence of stromal reaction [ 14 ]. 2.2 MRI acquisition and analysis All MRI scanning was performed with a 3.0 T scanner (Ingenia; Philips Healthcare) by using head and neck array coil. Axial contrast-enhanced T1-weighted images (ceT1WI) was used to measure maximum diameter and MRI-derived DOI, and axial fat-suppressed T2-weighted images (T2WI) was used for feature extraction. The MRI acquisition parameters were as follows: axial T2WI (repetition time [TR]/echo time [TE], 2800 ms/85 ms; matrix, 256 × 192; field of view, 240 × 240 mm; thickness, 3 mm; gap, 1 mm) and axial ceT1WI (TR/TE, 580 ms/15 ms; matrix, 256 × 192; field of view, 240 × 240 mm; thickness, 3 mm; gap, 1 mm). The gadolinium contrast agent (Magnevist, Bayer HealthCare Pharmaceuticals Inc.) was injected intravenously with a dosage of 0.1 mmol/kg at the injection rate of 2 mL/s. Demographic characteristics were derived from medical records. Tumor staging was determined according to the 8th edition of AJCC Staging Manual. Tumor maximum diameter and MRI-derived DOI were measured on axial ceT1WI for clinical T staging. The method of measuring MRI-derived DOI was as follows: two parallel lines were drawn respectively on the mucosa surface and the deepest point of the tumor on the largest section of tumor on axial images, and then the distance between the two parallel lines was measured as the MRI-derived DOI [ 7 ] (Supplementary Fig. 1). 2.3 Image segmentation and shape feature extraction Separate regions of interest (ROIs) encompassing primary tumor and the normal tongue tissue in contact with tumor were separately segmented on consecutive axial T2WI with ITK-SNAP (version 3.6.0; www.itksnap.org ) by a radiologist with 3 years of working experience. Feature extraction was implemented with the “Radiomics” module of 3D Slicer (version 4.10.2; www.slicer.org ). A total of 14 shape- and size-based morphological radiomics features were extracted. Detailed information on these features can be found at https://pyradiomics.readthedocs.io . We also determined two metrics to describe the area of contact between the tumor and surrounding normal tissue. TopBottomArea was defined as the adjacent slice where the top and the bottom plane including the tumor. SideArea was defined as the area where the tumor is laterally in contact with normal tongue tissue. TopBottomArea and SideArea were calculated with Python (version 3.10.8; https://www.python.org ). The segmentation methods are illustrated in Fig. 2 . 2.4 Feature selection and morphology-based radiomics model construction Intraclass correlation coefficients (ICCs) were calculated to evaluate the inter-observer reproducibility of demographic characteristics (tumor maximum diameter and MRI-derived DOI) and radiomics features. Thirty randomly cases of MRI were measured and segmented again by a radiologist skilled in head and neck MRI independently. Demographic characteristics and radiomics features with ICCs > 0.75 were considered having good reproducibility. Information gain algorithm was applied to rank the attributes of the features and select the top 5 attributes with the highest gain so as to reserve the most relevant features and minimize overfitting (refer to the practice of Wu et al. [ 15 ]). Information gain index could be used to measure the significance of features [ 12 ], which was calculated using the following equations: $$H\left(Y\right)=-{\sum }_{{y}_{i}\in Y}p\left({y}_{i}\right)\text{log}p\left({y}_{i}\right)$$ $$H\left(Y|X\right)=-{\sum }_{{{X}_{j}\in X,y}_{i}\in Y}p\left({{x}_{j},y}_{i}\right)\text{log}\frac{p\left({x}_{j},{y}_{i}\right)}{p\left({x}_{j}\right)}$$ $$IG\left(Y|X\right)=H\left(Y\right)- H\left(Y|X\right)$$ where X and Y represent random variables, H(Y) denotes the entropy of dataset Y, H(Y|X) denotes the conditional entropy, and IG(Y|X) represents the IG from X to Y. The feature was effective when the IG value was greater than zero, and features with greater IG are considered to contribute more to classification [ 16 ]. Then, the top 5 features identified by the IG algorithm were applied for the construction of the models. For predicting LNM and ENE, six supervised classification algorithms were applied, including neural network (NN), random forest (RF), logistic regression (LR), support vector machine (SVM), naïve bayes (NB), and AdaBoost, respectively. The details of machine learning algorithms are described in Supplementary Material. An internal stratified 10-fold cross-validation was then performed to alleviate the limitation small-sized dataset. The models were evaluated using the area under the curve (AUC) of the receiver operating characteristic (ROC) curves, accuracy, sensitivity, specificity, F1-score, precision and recall. 2.5 Statistical analysis Machine learning models was conducted using Orange data mining software (version 3.33.0). Statistical analyses were implemented with SPSS (version 19.0, IBM) and MedCalc (version 20.113, MedCalc). Correlations between each tumor characteristic were evaluated with t -tests, the χ 2 test, or the Mann-Whitney U test as appropriate. Morphological radiomics features were also compared with the Mann-Whitney U test between OTSCC with and without LNM. ROC curves were drawn and AUC were calculated to assess the predictive performance of the machine learning models. The DeLong test was performed to compare AUC derived from the MRI-DOI and that from machine learning models. P < 0.05 was deemed to indicate statistical significance. 3. Results 3.1 Clinical and pathologic characteristics In total, 90 patients with OTSCC were included in our study, of which LNM was pathologically detected in 45 cases (50.0%). The mean age of patients was 55.2 ± 13.1 years (range, 29 ~ 86 years). Clinical and pathological characteristics of patients are listed in Table 1 . There was no significant interclass difference in age, gender, tumor site of tongue and maximum diameter between the LNM-positive and LNM-negative groups ( P -value > 0.05). The characteristics with statistically significant differences were T stage, ENE, and MRI-derived DOI between those two groups ( P -value = 0.036, < 0.001, and 0.011, respectively). Table 1 Clinical and pathological characteristics of patients Characteristics LNM-positive (n = 45) LNM-negative (n = 45) P value Age (year) 55.4 ± 12.9 54.9 ± 13.5 0.638 Gender 0.378 Male 31 27 Female 14 18 T stage 0.036* T1 ~ T2 25 35 T3 ~ T4 20 10 ENE 13 0 <0.001* Tumor site of tongue 0.105 Blade 24 32 Abdomen 20 9 Root 1 2 Back 0 1 Tip 0 1 MRI-derived DOI(mm) 11.5(7.9, 16.7) 7.5(5.0, 15.7) 0.011* Maximum diameter(cm) 2.6(1.8, 4.0) 2.1(1.6, 3.4) 0.083 Age is presented as means ± standard deviations. Gender, T stage, ENE and tumor subsite are number (percentage) of patients. MRI-derived DOI and Maximum diameter are expressed as median (interquartile range). LNM lymph node metastasis, ENE extranodal extension, DOI depth of invasion * P < 0.05 3.2 Feature Selection and radiomics models building In reproducibility analysis, tumor maximum diameter, MRI-derived DOI and all morphological radiomics features showed reliable with an ICC higher than 0.75. After IG feature selection, respectively five most optimal features were selected for predicting LNM and for predicting ENE. The outcome attributes of selected features are presented in Fig. 3 . For predicting LNM, the selected five features were as follows: MRI-DOI, Top Bottom Area, Mesh Volume, Voxel Volume, and diameter (IG = 0.115, 0.071, 0.061, 0.061 and 0.018, respectively). For predicting ENE, the top five features were situation, Elongation, Top Bottom Area, Least Axis Length, and Minor Axis Length (IG = 0.052, 0.043, 0.043, 0.041 and 0.037, respectively). Comparisons of the selected top five features between groups and their diagnostic performances are shown in Tables 2 and 3 . Table 2 The selected top five features and their comparison between LNM groups Features LNM-positive (n = 45) LNM-negative (n = 45) P value AUC (95%CI) MRI-DOI 11.5(7.9, 16.7) 7.5(5.0, 15.7) 0.011* 0.655 (0.547,0.752) TopBottomArea 950.8(459.7, 1825.7) 483.3(317.4,940.9) 0.003* 0.684 (0.574,0.793) MeshVolume 5592.8(2298.3, 15797.8) 2689.0(1374.2,10917.3) 0.053 0.618 (0.502,0.735) VoxelVolume 5662.4(2348.0, 15901.8) 2727.3(1406.7, 10993.2) 0.052 0.619 (0.503,0.735) Diameter 2.6(1.8, 4.0) 2.1(1.6, 3.4) 0.083 0.606 (0.489,0.722) Data are expressed as median (interquartile range). LNM lymph node metastasis, DOI depth of invasion * P < 0.05 Table 3 The selected top five features and their comparison between ENE groups Features ENE-positive(n = 13) ENE-negative(n = 77) P value AUC (95%CI) Tumor site of tongue 0.179 0.619(0.459–0.778) Blade 5 51 Abdomen 8 21 Root 0 1 Back 0 3 Tip 0 1 Elongation 0.7(0.5,0.7) 0.7(0.6,0.8) 0.502 0.442(0.249,0.634) TopBottomArea 1349.3(639.4,2033.3) 644.6(363.6,1298.2) 0.050 0.670(0.512,0.828) LeastAxisLength 15.1(10.7,25.8) 11.5(8.5,18.6) 0.056 0.666(0.506,0.827) MinorAxisLength 24.7(13.4,36.7) 19.6(14.0,26.8) 0.326 0.585(0.396,0.775) Tumor site of tongue are described as numbers of patients and radiomics features are expressed as median (interquartile range). ENE extranodal extension Table 4 Performances of the six machine learning classifiers for predicting LNM and ENE Groups Classifier AUC (95%CI) Accuracy (%) Sensitivity (%) Specificity (%) F1-score Precision Recall LNM-positive vs. LNM-negative Neural Network 0.746(0.643–0.832) 72.2 77.8 68.9 0.722 0.722 0.722 SVM 0.654(0.546–0.751) 60.0 84.4 48.9 0.600 0.600 0.600 Naive Bayes 0.613(0.504–0.714) 60.0 62.2 60.0 0.600 0.600 0.600 AdaBoost 0.600(0.491–0.702) 60.0 60.0 60.0 0.600 0.600 0.600 Random Forest 0.594(0.485–0.696) 56.7 48.9 73.3 0.564 0.568 0.567 Logistic Regression 0.565(0.457–0.670) 60.0 44.4 77.8 0.587 0.614 0.600 ENE-positive vs. ENE-negative Neural Network 0.670(0.395–0.609) 84.4 92.3 41.6 0.783 0.731 0.844 SVM 0.750(0.648–0.836) 85.6 76.9 67.5 0.789 0.732 0.856 Naive Bayes 0.635(0.527–0.734) 73.3 84.6 45.5 0.757 0.791 0.733 AdaBoost 0.502(0.395–0.609) 73.3 84.6 16.9 0.741 0.749 0.733 Random Forest 0.526(0.418–0.632) 83.3 46.2 75.3 0.778 0.729 0.833 Logistic Regression 0.532(0.424–0.638) 84.4 30.8 90.9 0.783 0.731 0.844 LNM lymph node metastasis, ENE extranodal extension, AUC area under the receiver operating characteristic curve 3.3 Performance of morphology-based radiomics model As shown in Table 2 and Fig. 4 , MRI-derived DOI alone had an AUC of 0.655 (95% CI, 0.547–0.752) in predicting LNM, with the accuracy, sensitivity, and specificity of 66.7%, 91.1%, and 42.2%, respectively. Comparing the accuracy of different classifiers, the NN achieved the best performance based on selected features, yielded an AUC of 0.746(0.643–0.832), with the accuracy, sensitivity, and specificity of 72.2%, 77.8%, and 68.9%, respectively. Other classification models using SVM, NB, AdaBoost, RF, and LR achieved AUCs of 0.565 ~ 0.654 and accuracies of 56.7%~60.0%. A DeLong test showed that the AUC of the NN model was significantly higher than that of SVM, NB, AdaBoost, RF, and LR model (all P <0.05). The NN model had the slightly higher AUC than the MRI-DOI, although the difference was not significant ( P -value = 0.122). Comparing the accuracy of different classifiers in predicting ENE, the SVM achieved the best performance based on selected features, yielded an AUC of 0.750(0.648–0.836), with the accuracy, sensitivity, and specificity of 85.6%, 76.9%, and 67.5%, respectively. Other classification models using NN, NB, AdaBoost, RF, and LR achieved AUCs of 0.502 ~ 0.670 and accuracies of 73.3%~84.4%. A DeLong test showed that the AUC of the SVM model was significantly higher than that of LR and AdaBoost model ( P <0.05), and slightly higher than that of NN, NB and RF model with no significance ( P -value = 0.493, 0.393, and 0.072, respectively). The ROC curves of various models are shown in Fig. 4 . 4. Discussion To the best of our knowledge, there are only few studies illustrating whether an exclusive morphology-based radiomics analysis would enable the assessment of LNM and ENE risks in OTSCC. The current study constructed machine-learning models with clinical and radiomics morphological features extracted from preoperative MRI to predict lymph node status in OTSCC. For predicting LNM, the NN model showed the best diagnostic performance with an AUC of 0.746 and accuracy of 72.2%. For predicting ENE, the SVM model showed the best diagnostic performance with an AUC of 0.750 and accuracy of 85.6%. Radiomics is a process of high-throughput extraction of quantitative features that result in the conversion of images into mineable data and the subsequent analysis of these data for clinical decision support [ 17 ]. Previous research showed that radiomics could non-invasively help in predicting cervical lymph node status in oral cancers, thus showing tremendous potential for clinical diagnosis and treatment decisions [ 12 , 18 ]. However, as most previous studies have focused on histogram and texture analysis, the literature on shape analysis is relatively sparse. Radiomics-derived shape features capture the geometric properties of ROIs which are not directly based on voxel values, and should therefore not be greatly affected by interpolation effects [ 19 ]. Moreover, shape-based features are conceptually much simpler than other radiomic features. Wang et al. [ 20 ] developed a MRI-based machine learning model to evaluate cervical lymph node status in head and neck cancers, and four morphologic features were included in the model but without detailed description. Kubo et al. [ 11 ] selected Maximum2DDiameter as a shape feature into the model after the least absolute shrinkage and selection of operator logistic regression analysis. However, their contouring was performed at each neck node level instead of the primary tumor. In our study, we extracted the radiomics-based morphological features from the primary tumor and developed morphology-based machine-learning models in patients with OTSCC. Notably, our customized metric—SideArea—was not selected in the models, which may be related to the resolution of the axial plane; therefore, the estimation of its contact area was inaccurate. Another customized metric—TopBottomArea—was included in all models, which might reflect the extent of tumor invasion from a more comprehensive perspective. Further investigation of the predictive performance of TopBottomArea is warranted in future studies. Machine-learning models incorporating MRI-derived features from primary tumors can help to identify LNM in several tumors, with reported AUCs ranging from 0.79 to 0.90 [ 21 , 22 ]. Shan et al. [ 13 ] confirmed that machine learning combined with simple clinical and pathologic features showed a better performance in predicting lymph node metastasis of early-stage OTSCC than conventional prediction methods (AUC 0.786 vs. 0.539). However, their model included both preoperative and intraoperative information, which limited its predictive application before surgery. In the current study, all clinical characteristics and radiomics features were from preoperative data, thereby helping in surgical planning. The presence of ENE places patients in a higher stage, i.e., to either N2a for single small nodal metastases or N3b for multiple or large nodal metastases [ 14 ]. Identifying ENE by clinical and radiological examination is difficult, thereby leading to unnecessary overtreatment of the neck by surgical or chemoradiotherapy interventions. Currently, no definitive predictors are available for ENE [ 23 ]. Frood et al. [ 24 ] found that MRI textural analysis may aid in predicting ENE in oral cancers, with an accuracy of 79%. In the present study, by using the SVM model based on shape features from T2WI and clinical characteristics from ceT1WI, 85.6% of the nodes could be correctly classified preoperatively. This suggests that morphological features combined with machine-learning methods can provide more information and may be a better method for identifying ENE in OTSCC. We used the IG algorithm to select potential indicators and constructed the machine-learning models. This method allowed us to incorporate individual radiomics morphological features into a feature panel to perform multi-feature analyses. The present study results indicate that the combination of extracted radiomics and clinical traditional morphological variables has a complementary and synergistic effect in predicting LNM and ENE in patients with OTSCC. DOI was an important independent prognostic factor for lymph node metastasis and survival in patients with oral cancer. For every 5 mm increase in DOI, the T category increases by a level. Previous studies have indicated that MRI-derived DOI had high agreement with pathological DOI and was an independent prognostic factor for LNM in OTSCC [ 7 , 25 ]. Consistently, the current study identified MRI-derived DOI as statistically significant between LNM-positive and LNM-negative groups. Ren et al. [ 18 ] reported that MRI-derived DOI yielded an AUC of 0.67 for predicting occult cervical LNM in early-stage OTSCC. Similar to their study, our results showed an AUC of 0.655 when using MRI-derived DOI alone. Although the sensitivity of MRI-derived DOI was very high, with being 91.1% at the optimal cut-off, the specificity was only 42.2%. Therefore, MRI-derived DOI alone cannot be a reliable indicator to predict the occurrence of LNM. Our study indicated that radiomics shape features potentially improve the ability of the MRI-derived DOI to predict LNM. There are several limitations in the study. First, this was a retrospective single-institution design without an external validation. Though a stratified 10-fold cross-validation was used to preliminarily verify the feasibility of the study, a prospective multicenter study would be needed for further validation of the models and reduce the risk of overfitting. Secondly, the sample size for analysis was relatively small, especially the proportion of ENE patients. Such imbalance might influence the application of our machine learning models. Future studies should be conducted on a larger cohort, so that we can obtain adequate cases to further train the models and explore the relationship of lymph node status and morphological features. Thirdly, tumors were manually segmented, which may have introduced potential biases, so future research needs a reliable or widely accepted automated segmentation technique to extract the morphological and radiomic parameters of tumors. 5. Conclusion In the current study, we developed predictive models for cervical lymph node status in OTSCC using morphological features and machine learning from preoperative MRI. The models may provide a relatively accurate, convenient, and noninvasive method for distinguishing high risk of LNM and ENE in patients with OTSCC. Abbreviations MRI Magnetic resonance imaging OTSCC Oral tongue squamous cell carcinoma LNM Lymph node metastasis ENE Extranodal extension DOI Depth of invasion T2W T2-weighted ceT1W Contrast-enhanced T1-weighted NN Neural network RF Random forest LR Logistic regression SVM Support vector machine NB Naïve bayes AUC Area under the curve TNM Tumor, node, and metastasis ROIs Regions of interest ICCs Intraclass correlation coefficients AUC Area under the curve ROC Receiver operating characteristic Declarations Author Contribution Yunjing Zhu and Ying Yuan wrote the main manuscript text. Jiliang Ren and Yang Song did the analysis and statistics. All authors reviewed the manuscript. References R.L. Siegel, K.D. Miller, A. Jemal, Cancer statistics, 2018, CA Cancer J. Clin. 68 (2018) 7-30. https://doi.org/10.3322/caac.21442. S.B. Chinn, J.N. Myers, Oral Cavity Carcinoma: Current Management, Controversies, and Future Directions, J. Clin. Oncol. 33 (2015) 3269-76. https://doi.org/10.1200/JCO.2015.61.2929. M. Mermod, G. Tolstonog, C. Simon, Y. Monnier, Extracapsular spread in head and neck squamous cell carcinoma: A systematic review and meta-analysis, Oral Oncol. 62 (2016) 60-71. https://doi.org/10.1016/j.oraloncology.2016.10.003. D. Mattavelli, M. Ferrari, S. Taboni, R. Morello, A. Paderno, V. Rampinelli, F. Del Bon, D. Lombardi, A. Grammatica, P. Bossi, A. Deganello, C. Piazza, P. Nicolai, The 8th TNM classification for oral squamous cell carcinoma: What is gained, what is lost, and what is missing, Oral Oncol. 111 (2020). https://doi.org/ARTN 10493710.1016/j.oraloncology.2020.104937. A. Baba, K. Hashimoto, R. Kayama, H. Yamauchi, K. Ikeda, O. Hiroya, Correction to: Radiological approach for the newly incorporated T staging factor, depth of invasion (DOI), of the oral tongue cancer in the 8th edition of American Joint Committee on Cancer (AJCC) staging manual: assessment of the necessity for elective neck dissection, Jpn J Radiol. 39 (2021) 100. https://doi.org/10.1007/s11604-020-01047-8. A. Almangush, A.A. Makitie, A. Triantafyllou, R. de Bree, P. Strojan, A. Rinaldo, J.C. Hernandez-Prera, C. Suarez, L.P. Kowalski, A. Ferlito, I. Leivo, Staging and grading of oral squamous cell carcinoma: An update, Oral Oncol. 107 (2020) 104799. https://doi.org/10.1016/j.oraloncology.2020.104799. W. Tang, Y. Wang, Y. Yuan, X. Tao, Assessment of tumor depth in oral tongue squamous cell carcinoma with multiparametric MRI: correlation with pathology, Eur. Radiol. 32 (2022) 254-261. https://doi.org/10.1007/s00330-021-08148-6. T. Waech, S. Pazahr, V. Guarda, N.J. Rupp, M.A. Broglie, G.B. Morand, Measurement variations of MRI and CT in the assessment of tumor depth of invasion in oral cancer: A retrospective study, Eur. J. Radiol. 135 (2021) 109480. https://doi.org/10.1016/j.ejrad.2020.109480. Z.Y. Peng, Y.M. Wang, Y.X. Wang, S.J. Jiang, R.H. Fan, H. Zhang, W.H. Jiang, Application of radiomics and machine learning in head and neck cancers, Int. J. Biol. Sci. 17 (2021) 475-486. https://doi.org/10.7150/ijbs.55716. A.A. Ardakani, N.J. Bureau, E.J. Ciaccio, U.R. Acharya, Interpretation of radiomics features-A pictorial review, Comput. Methods Programs Biomed. 215 (2022). https://doi.org/ARTN 10660910.1016/j.cmpb.2021.106609. K. Kubo, D. Kawahara, Y. Murakami, Y. Takeuchi, T. Katsuta, N. Imano, I. Nishibuchi, A. Saito, M. Konishi, N. Kakimoto, Y. Yoshioka, S. Toratani, S. Ono, T. Ueda, S. Takeno, Y. Nagata, 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. 134 (2022) 93-101. https://doi.org/10.1016/j.oooo.2021.12.122. Y. Yuan, J. Ren, X. Tao, Machine learning-based MRI texture analysis to predict occult lymph node metastasis in early-stage oral tongue squamous cell carcinoma, Eur. Radiol. 31 (2021) 6429-6437. https://doi.org/10.1007/s00330-021-07731-1. J. Shan, R. Jiang, X. Chen, Y. Zhong, W. Zhang, L. Xie, J. Cheng, H. Jiang, Machine Learning Predicts Lymph Node Metastasis in Early-Stage Oral Tongue Squamous Cell Carcinoma, J. Oral Maxillofac. Surg. 78 (2020) 2208-2218. https://doi.org/10.1016/j.joms.2020.06.015. W.M. Lydiatt, S.G. Patel, B. O'Sullivan, M.S. Brandwein, J.A. Ridge, J.C. Migliacci, A.M. Loomis, J.P. Shah, Head and Neck cancers-major changes in the American Joint Committee on cancer eighth edition cancer staging manual, CA Cancer J. Clin. 67 (2017) 122-137. https://doi.org/10.3322/caac.21389. W. Wu, C. Parmar, P. Grossmann, J. Quackenbush, P. Lambin, J. Bussink, R. Mak, H.J. Aerts, Exploratory Study to Identify Radiomics Classifiers for Lung Cancer Histology, Front. Oncol. 6 (2016) 71. https://doi.org/10.3389/fonc.2016.00071. R. Chai, Q. Wang, P. Qin, J. Zeng, J. Ren, R. Zhang, L. Chu, X. Zhang, Y. Guan, Differentiating Central Lung Tumors from Atelectasis with Contrast-Enhanced CT-Based Radiomics Features, Biomed Res Int. 2021 (2021) 5522452. https://doi.org/10.1155/2021/5522452. R.J. Gillies, P.E. Kinahan, H. Hricak, Radiomics: Images Are More than Pictures, They Are Data, Radiology. 278 (2016) 563-577. https://doi.org/10.1148/radiol.2015151169. J. Ren, Y. Yuan, X. Tao, 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. (2021). https://doi.org/10.1007/s00330-021-08310-0. P. Bos, M.W.M. van den Brekel, M. Taghavi, Z.A.R. Gouw, A. Al-Mamgani, S. Waktola, H.J.W.L. Aerts, R.G.H. Beets-Tan, J.A. Castelijns, B. Jasperse, Simple delineations cannot substitute full 3d tumor delineations for MR-based radiomics prediction of locoregional control in oropharyngeal cancer, Eur. J. Radiol. 148 (2022). https://doi.org/ARTN 11016710.1016/j.ejrad.2022.110167. F. Wang, R. Tan, K. Feng, J. Hu, Z. Zhuang, C. Wang, J. Hou, X. Liu, Magnetic Resonance Imaging-Based Radiomics Features Associated with Depth of Invasion Predicted Lymph Node Metastasis and Prognosis in Tongue Cancer, J. Magn. Reson. Imaging. 56 (2022) 196-209. https://doi.org/10.1002/jmri.28019. S. Kasai, A. Shiomi, H. Kagawa, H. Hino, S. Manabe, Y. Yamaoka, K. Chen, K. Nanishi, Y. Kinugasa, The Effectiveness of Machine Learning in Predicting Lateral Lymph Node Metastasis From Lower Rectal Cancer: A Single Center Development and Validation Study, Ann Gastroenterol Surg. 6 (2022) 92-100. https://doi.org/10.1002/ags3.12504. A. Jajodia, A. Gupta, H. Prosch, M. Mayerhoefer, S. Mitra, S. Pasricha, A. Mehta, S. Puri, A. Chaturvedi, Combination of Radiomics and Machine Learning with Diffusion-Weighted MR Imaging for Clinical Outcome Prognostication in Cervical Cancer, Tomography. 7 (2021) 344-357. https://doi.org/10.3390/tomography7030031. Y. Tang, C.M. Yang, S. Su, W.J. Wang, L.P. Fan, J. Shu, Machine learning-based Radiomics analysis for differentiation degree and lymphatic node metastasis of extrahepatic cholangiocarcinoma, BMC Cancer. 21 (2021) 1268. https://doi.org/10.1186/s12885-021-08947-6. R. Frood, E. Palkhi, M. Barnfield, R. Prestwich, S. Vaidyanathan, A. Scarsbrook, Can MR textural analysis improve the prediction of extracapsular nodal spread in patients with oral cavity cancer?, Eur. Radiol. 28 (2018) 5010-5018. https://doi.org/10.1007/s00330-018-5524-x. C. Xu, J. Yuan, L. Kang, X. Zhang, L. Wang, X. Chen, Q. Yao, H. Li, Significance of depth of invasion determined by MRI in cT1N0 tongue squamous cell carcinoma, Sci. Rep. 10 (2020) 4695. https://doi.org/10.1038/s41598-020-61474-5. Supplementary Material Supplementary Material is not available with this version Additional Declarations No competing interests reported. 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-3909740","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":280586213,"identity":"eee2e193-2f8d-411a-afc5-04b2d00255f0","order_by":0,"name":"Yunjing Zhu","email":"","orcid":"","institution":"Shanghai Ninth People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yunjing","middleName":"","lastName":"Zhu","suffix":""},{"id":280586214,"identity":"588050de-be8d-43e6-81dd-49286f22a30f","order_by":1,"name":"Jiliang Ren","email":"","orcid":"","institution":"Shanghai Ninth People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jiliang","middleName":"","lastName":"Ren","suffix":""},{"id":280586215,"identity":"24f46160-aaa0-4ed2-8b69-1e42d40c032d","order_by":2,"name":"Yang Song","email":"","orcid":"","institution":"Siemens Healthineers (China)","correspondingAuthor":false,"prefix":"","firstName":"Yang","middleName":"","lastName":"Song","suffix":""},{"id":280586217,"identity":"6b6420ab-b523-433b-88db-18ecca5e8d4d","order_by":3,"name":"Xiaofeng Tao","email":"","orcid":"","institution":"Shanghai Ninth People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xiaofeng","middleName":"","lastName":"Tao","suffix":""},{"id":280586219,"identity":"f89c4645-2c6f-41b6-8a9b-450c2dc34f80","order_by":4,"name":"Ying Yuan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAuUlEQVRIiWNgGAWjYNCCiho5Nvb2A6RoOXPMmI/nTAIJOhjbmBPnSTgYEKfa4Ebys8c8Z9jS2yQYEhh+VGwjRkuauTFPhUxum3TjAcaeM7eJ0ZJgJg20JbdN5kACM2MbUVrSv0nztjGns0kkGBCrJccMpCWBeC2SZ96USc45c8ywDRjIB4nyC9/x9G0Sbypq5OXb2w8++FFBhBaFAwwMTDxQzgHC6oFAvgEYkz+IUjoKRsEoGAUjFgAAPgQ8fdwOSdsAAAAASUVORK5CYII=","orcid":"","institution":"Shanghai Ninth People's Hospital","correspondingAuthor":true,"prefix":"","firstName":"Ying","middleName":"","lastName":"Yuan","suffix":""}],"badges":[],"createdAt":"2024-01-30 06:48:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3909740/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3909740/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":53008671,"identity":"25c57ab3-773c-4bd0-adb1-836236acf25f","added_by":"auto","created_at":"2024-03-19 15:19:42","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":118701,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart showing inclusion and exclusion criteria. \u003cem\u003eLNM\u003c/em\u003e lymph node metastasis, \u003cem\u003eENE\u003c/em\u003eextranodal extension\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-3909740/v1/bf2bfe3eeedeb3adc51ffb9f.png"},{"id":53006752,"identity":"57abd8f0-585e-472b-8325-eb349c8b0bbf","added_by":"auto","created_at":"2024-03-19 15:11:42","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":3182840,"visible":true,"origin":"","legend":"\u003cp\u003eRegion of interest (ROI) segmentation in OTSCC. (\u003cstrong\u003ea\u003c/strong\u003e) ROIs of the primary tumor (red) and normal togue tissue around tumor (green) were segmented on T2-weighted images. By stacking up segmented ROIs slice-by-slice and calculated the area where the tumor is in contact with normal tongue tissue on the side, SideArea was acquired. (\u003cstrong\u003eb-c\u003c/strong\u003e) The area that the adjacent slice where the top and the bottom plane including the tumor (blue) were calculated and defined as TopBottomArea.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-3909740/v1/0863fdc5813246c7c2e0423d.png"},{"id":53006754,"identity":"5923d0f7-dedb-44e1-94c4-b8bc28e28699","added_by":"auto","created_at":"2024-03-19 15:11:42","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":284285,"visible":true,"origin":"","legend":"\u003cp\u003eThe outcome attributes of selected features for building the predictive models of lymph node metastasis (LNM) or extranodal extension (ENE) using the Information Gain algorithm.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-3909740/v1/4a7e271b6ae1bee493b744e2.png"},{"id":53006755,"identity":"91265a6d-b2f4-4f9e-9e57-ce49996c3622","added_by":"auto","created_at":"2024-03-19 15:11:42","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":213797,"visible":true,"origin":"","legend":"\u003cp\u003eReceiver operating characteristic\u003cstrong\u003e \u003c/strong\u003e(ROC) curves of the MRI-derived depth of invasion (DOI) and different models for predicting cervical lymph node status in OTSCC. (\u003cstrong\u003ea\u003c/strong\u003e) ROC curves of MRI-derived DOI for predicting lymph node metastasis (LNM). (\u003cstrong\u003eb\u003c/strong\u003e) Six machine learning classifiers for predicting LNM. The neural network (NN) model performed best (AUC, 0. 746). (\u003cstrong\u003ec\u003c/strong\u003e) Six machine learning classifiers for predicting extranodal extension (ENE). The support vector machine (SVM) model performed best (AUC, 0.750)\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-3909740/v1/9619df3e50d5016aea21e142.png"},{"id":59546680,"identity":"190ee416-182f-45fc-be21-ff31082ac787","added_by":"auto","created_at":"2024-07-03 05:20:30","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6338544,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3909740/v1/957d0598-0e96-4434-a838-9577d25cc4cc.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Morphology-based Machine-Learning for Predicting Lymph Node Status in Oral Tongue Squamous Cell Carcinoma","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eOral cancer is a very aggressive malignancy, and oral tongue squamous cell carcinoma (OTSCC) is the most common type [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Cervical lymph node metastasis (LNM) is one of the prognostic factors for OTSCC patients. The 5-year survival rate drops by half with the occurrence of LNM [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. An additional drop in survival occurs when extranodal extension (ENE) is present in the metastatic lymph nodes [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe tumor, node, and metastasis (TNM) staging system is the principal criterion to describe and stage tumor extension [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. In fact, the T stage index of tumor depth of invasion (DOI) is proved to be strongly associated with the N stage in OTSCC [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. 除了DOI 还有其他形态学指标和淋巴结转移相关。Thus, preoperatively predicting LNM in OTSCC with morphological features and T stage index is theoretically feasible, which is especially crucial for patients with no clinical evidence for LNM preoperatively but was confirmed histopathologically afterwards (cN0).\u003c/p\u003e \u003cp\u003eMagnetic resonance imaging (MRI) is recommended for the preoperative evaluation of tongue cancer and can be used as a reference to determine clinical TNM stage owing to its advantage in soft-tissue contrasting and tumor-contour depicting [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Traditionally, morphological appearance of lesions on MRI images are evaluated subjectively and qualitatively [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. A more comprehensive and repeatable method for depicting morphology of the original tumor is thus required to potentially predict lymph node status in OTSCC.\u003c/p\u003e \u003cp\u003eRadiomics can extract quantitative morphological information from conventional images objectively [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. The potential of these quantitative shape features in distinguishing benign from malignant tumors could be emphasized. To avoid sophisticated mathematical computation of radiomics features which are limited in explanation in clinics, we chose to extract explicit morphological features, such as volume, perimeter and contour irregularity of lesions [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Machine learning method has promising application in radiomics given that its algorithms are suited for analysis of high-dimensional data. Previous studies have used machine-learning techniques to predict treatment response and other prognostic outcomes in tongue cancer [\u003cspan additionalcitationids=\"CR12\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Therefore, the aim of this study was to construct machine-learning models based on morphological radiomics features from MRI to predict lymph node status of patients with OTSCC.\u003c/p\u003e"},{"header":"2. Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003e2.1 Patients\u003c/h2\u003e\n \u003cp\u003eThis retrospective study was approved by Institutional Review Board of Shanghai Ninth People\u0026rsquo;s Hospital, and the informed consent was waived. We reviewed the medical records from June 2015 to May 2022 to identify patients with OTSCC. The inclusion criteria were as follows: (1) patients underwent preoperative contrast-enhanced head and neck MRI; (2) patients were treated with primary tumor resection and elective neck dissection; and (3) complete clinical and pathological information was available. The exclusion criteria were as follows: (1) patients previously received radiotherapy, surgery and/or chemotherapy; (2) the presence of other concurrent head and neck malignant tumor; and (3) obvious artifact impacting MRI image analysis. In total, 90 patients were included for further analysis (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003cp\u003eThe presence of pathologically involved LNM and ENE was recorded based on histopathology reports at the time of initial treatment. ENE was considered positive if metastatic carcinoma extended from within a lymph node through the fibrous capsule and into the surrounding connective tissue, regardless of the presence of stromal reaction [\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e].\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\n \u003ch2\u003e2.2 MRI acquisition and analysis\u003c/h2\u003e\n \u003cp\u003eAll MRI scanning was performed with a 3.0 T scanner (Ingenia; Philips Healthcare) by using head and neck array coil. Axial contrast-enhanced T1-weighted images (ceT1WI) was used to measure maximum diameter and MRI-derived DOI, and axial fat-suppressed T2-weighted images (T2WI) was used for feature extraction. The MRI acquisition parameters were as follows: axial T2WI (repetition time [TR]/echo time [TE], 2800 ms/85 ms; matrix, 256 \u0026times; 192; field of view, 240 \u0026times; 240 mm; thickness, 3 mm; gap, 1 mm) and axial ceT1WI (TR/TE, 580 ms/15 ms; matrix, 256 \u0026times; 192; field of view, 240 \u0026times; 240 mm; thickness, 3 mm; gap, 1 mm). The gadolinium contrast agent (Magnevist, Bayer HealthCare Pharmaceuticals Inc.) was injected intravenously with a dosage of 0.1 mmol/kg at the injection rate of 2 mL/s.\u003c/p\u003e\n \u003cp\u003eDemographic characteristics were derived from medical records. Tumor staging was determined according to the 8th edition of AJCC Staging Manual. Tumor maximum diameter and MRI-derived DOI were measured on axial ceT1WI for clinical T staging. The method of measuring MRI-derived DOI was as follows: two parallel lines were drawn respectively on the mucosa surface and the deepest point of the tumor on the largest section of tumor on axial images, and then the distance between the two parallel lines was measured as the MRI-derived DOI [\u003cspan class=\"CitationRef\"\u003e7\u003c/span\u003e] (Supplementary Fig.\u0026nbsp;1).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\n \u003ch2\u003e2.3 Image segmentation and shape feature extraction\u003c/h2\u003e\n \u003cp\u003eSeparate regions of interest (ROIs) encompassing primary tumor and the normal tongue tissue in contact with tumor were separately segmented on consecutive axial T2WI with ITK-SNAP (version 3.6.0; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ewww.itksnap.org\u003c/span\u003e\u003c/span\u003e) by a radiologist with 3 years of working experience. Feature extraction was implemented with the \u0026ldquo;Radiomics\u0026rdquo; module of 3D Slicer (version 4.10.2; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ewww.slicer.org\u003c/span\u003e\u003c/span\u003e). A total of 14 shape- and size-based morphological radiomics features were extracted. Detailed information on these features can be found at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pyradiomics.readthedocs.io\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003eWe also determined two metrics to describe the area of contact between the tumor and surrounding normal tissue. TopBottomArea was defined as the adjacent slice where the top and the bottom plane including the tumor. SideArea was defined as the area where the tumor is laterally in contact with normal tongue tissue. TopBottomArea and SideArea were calculated with Python (version 3.10.8; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.python.org\u003c/span\u003e\u003c/span\u003e). The segmentation methods are illustrated in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\n \u003ch2\u003e2.4 Feature selection and morphology-based radiomics model construction\u003c/h2\u003e\n \u003cp\u003eIntraclass correlation coefficients (ICCs) were calculated to evaluate the inter-observer reproducibility of demographic characteristics (tumor maximum diameter and MRI-derived DOI) and radiomics features. Thirty randomly cases of MRI were measured and segmented again by a radiologist skilled in head and neck MRI independently. Demographic characteristics and radiomics features with ICCs\u0026thinsp;\u0026gt;\u0026thinsp;0.75 were considered having good reproducibility.\u003c/p\u003e\n \u003cp\u003eInformation gain algorithm was applied to rank the attributes of the features and select the top 5 attributes with the highest gain so as to reserve the most relevant features and minimize overfitting (refer to the practice of Wu et al. [\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e]). Information gain index could be used to measure the significance of features [\u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e], which was calculated using the following equations:\u003c/p\u003e\n \u003cdiv id=\"Equa\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e$$H\\left(Y\\right)=-{\\sum }_{{y}_{i}\\in Y}p\\left({y}_{i}\\right)\\text{log}p\\left({y}_{i}\\right)$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e$$H\\left(Y|X\\right)=-{\\sum }_{{{X}_{j}\\in X,y}_{i}\\in Y}p\\left({{x}_{j},y}_{i}\\right)\\text{log}\\frac{p\\left({x}_{j},{y}_{i}\\right)}{p\\left({x}_{j}\\right)}$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equc\" class=\"Equation\"\u003e\u003cdiv class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e$$IG\\left(Y|X\\right)=H\\left(Y\\right)- H\\left(Y|X\\right)$$\u003c/div\u003e\u003c/div\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\u003cp\u003ewhere X and Y represent random variables, H(Y) denotes the entropy of dataset Y, H(Y|X) denotes the conditional entropy, and IG(Y|X) represents the IG from X to Y. The feature was effective when the IG value was greater than zero, and features with greater IG are considered to contribute more to classification [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e]. Then, the top 5 features identified by the IG algorithm were applied for the construction of the models. For predicting LNM and ENE, six supervised classification algorithms were applied, including neural network (NN), random forest (RF), logistic regression (LR), support vector machine (SVM), na\u0026iuml;ve bayes (NB), and AdaBoost, respectively. The details of machine learning algorithms are described in Supplementary Material. An internal stratified 10-fold cross-validation was then performed to alleviate the limitation small-sized dataset. The models were evaluated using the area under the curve (AUC) of the receiver operating characteristic (ROC) curves, accuracy, sensitivity, specificity, F1-score, precision and recall.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.5 Statistical analysis\u003c/h2\u003e\u003cp\u003eMachine learning models was conducted using Orange data mining software (version 3.33.0). Statistical analyses were implemented with SPSS (version 19.0, IBM) and MedCalc (version 20.113, MedCalc). Correlations between each tumor characteristic were evaluated with \u003cem\u003et\u003c/em\u003e-tests, the \u003cem\u003e\u0026chi;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e test, or the Mann-Whitney \u003cem\u003eU\u003c/em\u003e test as appropriate. Morphological radiomics features were also compared with the Mann-Whitney \u003cem\u003eU\u003c/em\u003e test between OTSCC with and without LNM. ROC curves were drawn and AUC were calculated to assess the predictive performance of the machine learning models. The DeLong test was performed to compare AUC derived from the MRI-DOI and that from machine learning models. \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was deemed to indicate statistical significance.\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Clinical and pathologic characteristics\u003c/h2\u003e \u003cp\u003eIn total, 90 patients with OTSCC were included in our study, of which LNM was pathologically detected in 45 cases (50.0%). The mean age of patients was 55.2\u0026thinsp;\u0026plusmn;\u0026thinsp;13.1 years (range, 29\u0026thinsp;~\u0026thinsp;86 years). Clinical and pathological characteristics of patients are listed in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. There was no significant interclass difference in age, gender, tumor site of tongue and maximum diameter between the LNM-positive and LNM-negative groups (\u003cem\u003eP\u003c/em\u003e-value\u0026thinsp;\u0026gt;\u0026thinsp;0.05). The characteristics with statistically significant differences were T stage, ENE, and MRI-derived DOI between those two groups (\u003cem\u003eP\u003c/em\u003e-value\u0026thinsp;=\u0026thinsp;0.036, \u0026lt;\u0026thinsp;0.001, and 0.011, respectively).\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\u003eClinical and pathological characteristics of patients\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCharacteristics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLNM-positive (n\u0026thinsp;=\u0026thinsp;45)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLNM-negative (n\u0026thinsp;=\u0026thinsp;45)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge (year)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e55.4\u0026thinsp;\u0026plusmn;\u0026thinsp;12.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e54.9\u0026thinsp;\u0026plusmn;\u0026thinsp;13.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.638\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender\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\u003e0.378\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\u003e31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\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\u003e14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\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\u003e0.036*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT1\u0026thinsp;~\u0026thinsp;T2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT3\u0026thinsp;~\u0026thinsp;T4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eENE\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;0.001*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTumor site of tongue\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\u003e0.105\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBlade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAbdomen\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRoot\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBack\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTip\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMRI-derived DOI(mm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11.5(7.9, 16.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.5(5.0, 15.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.011*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMaximum diameter(cm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.6(1.8, 4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.1(1.6, 3.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.083\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003eAge is presented as means\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviations. Gender, T stage, ENE and tumor subsite are number (percentage) of patients. MRI-derived DOI and Maximum diameter are expressed as median (interquartile range). \u003cem\u003eLNM\u003c/em\u003e lymph node metastasis, \u003cem\u003eENE\u003c/em\u003e extranodal extension, \u003cem\u003eDOI\u003c/em\u003e depth of invasion\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003e*\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Feature Selection and radiomics models building\u003c/h2\u003e \u003cp\u003eIn reproducibility analysis, tumor maximum diameter, MRI-derived DOI and all morphological radiomics features showed reliable with an ICC higher than 0.75. After IG feature selection, respectively five most optimal features were selected for predicting LNM and for predicting ENE. The outcome attributes of selected features are presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. For predicting LNM, the selected five features were as follows: MRI-DOI, Top Bottom Area, Mesh Volume, Voxel Volume, and diameter (IG\u0026thinsp;=\u0026thinsp;0.115, 0.071, 0.061, 0.061 and 0.018, respectively). For predicting ENE, the top five features were situation, Elongation, Top Bottom Area, Least Axis Length, and Minor Axis Length (IG\u0026thinsp;=\u0026thinsp;0.052, 0.043, 0.043, 0.041 and 0.037, respectively). Comparisons of the selected top five features between groups and their diagnostic performances are shown in Tables\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e and \u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\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\u003eThe selected top five features and their comparison between LNM groups\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFeatures\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLNM-positive (n\u0026thinsp;=\u0026thinsp;45)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLNM-negative (n\u0026thinsp;=\u0026thinsp;45)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAUC (95%CI)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMRI-DOI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e11.5(7.9, 16.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7.5(5.0, 15.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.011*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.655 (0.547,0.752)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTopBottomArea\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e950.8(459.7, 1825.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e483.3(317.4,940.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.003*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.684 (0.574,0.793)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMeshVolume\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5592.8(2298.3, 15797.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2689.0(1374.2,10917.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.053\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.618 (0.502,0.735)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVoxelVolume\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5662.4(2348.0, 15901.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2727.3(1406.7, 10993.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.052\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.619 (0.503,0.735)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiameter\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.6(1.8, 4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2.1(1.6, 3.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.083\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.606 (0.489,0.722)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003eData are expressed as median (interquartile range). \u003cem\u003eLNM\u003c/em\u003e lymph node metastasis, \u003cem\u003eDOI\u003c/em\u003e depth of invasion\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e*\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe selected top five features and their comparison between ENE groups\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\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=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFeatures\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eENE-positive(n\u0026thinsp;=\u0026thinsp;13)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eENE-negative(n\u0026thinsp;=\u0026thinsp;77)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAUC (95%CI)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTumor site of tongue\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\u003e0.179\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.619(0.459\u0026ndash;0.778)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBlade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAbdomen\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRoot\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBack\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTip\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eElongation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.7(0.5,0.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.7(0.6,0.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.502\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.442(0.249,0.634)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTopBottomArea\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1349.3(639.4,2033.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e644.6(363.6,1298.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.050\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.670(0.512,0.828)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeastAxisLength\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15.1(10.7,25.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11.5(8.5,18.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.056\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.666(0.506,0.827)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMinorAxisLength\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24.7(13.4,36.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e19.6(14.0,26.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.326\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.585(0.396,0.775)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003eTumor site of tongue are described as numbers of patients and radiomics features are expressed as median (interquartile range). \u003cem\u003eENE\u003c/em\u003e extranodal extension\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformances of the six machine learning classifiers for predicting LNM and ENE\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\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=\"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=\"char\" char=\".\" 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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGroups\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClassifier\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC (95%CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAccuracy (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSensitivity (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSpecificity (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eF1-score\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLNM-positive vs. LNM-negative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNeural Network\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.746(0.643\u0026ndash;0.832)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e72.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e77.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e68.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.722\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.722\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.722\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.654(0.546\u0026ndash;0.751)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e60.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e84.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e48.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.600\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.600\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.600\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNaive Bayes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.613(0.504\u0026ndash;0.714)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e60.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e62.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e60.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.600\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.600\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.600\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAdaBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.600(0.491\u0026ndash;0.702)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e60.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e60.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e60.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.600\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.600\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.600\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.594(0.485\u0026ndash;0.696)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e56.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e48.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e73.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.564\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.568\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.567\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLogistic Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.565(0.457\u0026ndash;0.670)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e60.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e44.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e77.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.587\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.614\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.600\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eENE-positive vs. ENE-negative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNeural Network\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.670(0.395\u0026ndash;0.609)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e84.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e92.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e41.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.783\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.731\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.844\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.750(0.648\u0026ndash;0.836)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e85.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e76.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e67.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.789\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.732\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.856\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNaive Bayes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.635(0.527\u0026ndash;0.734)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e73.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e84.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e45.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.757\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.791\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.733\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAdaBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.502(0.395\u0026ndash;0.609)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e73.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e84.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e16.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.741\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.749\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.733\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.526(0.418\u0026ndash;0.632)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e83.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e46.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e75.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.778\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.729\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.833\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLogistic Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.532(0.424\u0026ndash;0.638)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e84.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e30.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e90.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.783\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.731\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.844\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"9\"\u003e\u003cem\u003eLNM\u003c/em\u003e lymph node metastasis, \u003cem\u003eENE\u003c/em\u003e extranodal extension, \u003cem\u003eAUC\u003c/em\u003e area under the receiver operating characteristic curve\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\u003e3.3 Performance of morphology-based radiomics model\u003c/h2\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, MRI-derived DOI alone had an AUC of 0.655 (95% CI, 0.547\u0026ndash;0.752) in predicting LNM, with the accuracy, sensitivity, and specificity of 66.7%, 91.1%, and 42.2%, respectively. Comparing the accuracy of different classifiers, the NN achieved the best performance based on selected features, yielded an AUC of 0.746(0.643\u0026ndash;0.832), with the accuracy, sensitivity, and specificity of 72.2%, 77.8%, and 68.9%, respectively. Other classification models using SVM, NB, AdaBoost, RF, and LR achieved AUCs of 0.565\u0026thinsp;~\u0026thinsp;0.654 and accuracies of 56.7%~60.0%. A DeLong test showed that the AUC of the NN model was significantly higher than that of SVM, NB, AdaBoost, RF, and LR model (all \u003cem\u003eP\u003c/em\u003e\u0026lt;0.05). The NN model had the slightly higher AUC than the MRI-DOI, although the difference was not significant (\u003cem\u003eP\u003c/em\u003e-value\u0026thinsp;=\u0026thinsp;0.122).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eComparing the accuracy of different classifiers in predicting ENE, the SVM achieved the best performance based on selected features, yielded an AUC of 0.750(0.648\u0026ndash;0.836), with the accuracy, sensitivity, and specificity of 85.6%, 76.9%, and 67.5%, respectively. Other classification models using NN, NB, AdaBoost, RF, and LR achieved AUCs of 0.502\u0026thinsp;~\u0026thinsp;0.670 and accuracies of 73.3%~84.4%. A DeLong test showed that the AUC of the SVM model was significantly higher than that of LR and AdaBoost model (\u003cem\u003eP\u003c/em\u003e\u0026lt;0.05), and slightly higher than that of NN, NB and RF model with no significance (\u003cem\u003eP\u003c/em\u003e-value\u0026thinsp;=\u0026thinsp;0.493, 0.393, and 0.072, respectively). The ROC curves of various models are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eTo the best of our knowledge, there are only few studies illustrating whether an exclusive morphology-based radiomics analysis would enable the assessment of LNM and ENE risks in OTSCC.\u003c/p\u003e \u003cp\u003eThe current study constructed machine-learning models with clinical and radiomics morphological features extracted from preoperative MRI to predict lymph node status in OTSCC. For predicting LNM, the NN model showed the best diagnostic performance with an AUC of 0.746 and accuracy of 72.2%. For predicting ENE, the SVM model showed the best diagnostic performance with an AUC of 0.750 and accuracy of 85.6%.\u003c/p\u003e \u003cp\u003eRadiomics is a process of high-throughput extraction of quantitative features that result in the conversion of images into mineable data and the subsequent analysis of these data for clinical decision support [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Previous research showed that radiomics could non-invasively help in predicting cervical lymph node status in oral cancers, thus showing tremendous potential for clinical diagnosis and treatment decisions [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. However, as most previous studies have focused on histogram and texture analysis, the literature on shape analysis is relatively sparse. Radiomics-derived shape features capture the geometric properties of ROIs which are not directly based on voxel values, and should therefore not be greatly affected by interpolation effects [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Moreover, shape-based features are conceptually much simpler than other radiomic features. Wang et al. [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] developed a MRI-based machine learning model to evaluate cervical lymph node status in head and neck cancers, and four morphologic features were included in the model but without detailed description. Kubo et al. [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] selected Maximum2DDiameter as a shape feature into the model after the least absolute shrinkage and selection of operator logistic regression analysis. However, their contouring was performed at each neck node level instead of the primary tumor. In our study, we extracted the radiomics-based morphological features from the primary tumor and developed morphology-based machine-learning models in patients with OTSCC. Notably, our customized metric\u0026mdash;SideArea\u0026mdash;was not selected in the models, which may be related to the resolution of the axial plane; therefore, the estimation of its contact area was inaccurate. Another customized metric\u0026mdash;TopBottomArea\u0026mdash;was included in all models, which might reflect the extent of tumor invasion from a more comprehensive perspective. Further investigation of the predictive performance of TopBottomArea is warranted in future studies.\u003c/p\u003e \u003cp\u003eMachine-learning models incorporating MRI-derived features from primary tumors can help to identify LNM in several tumors, with reported AUCs ranging from 0.79 to 0.90 [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Shan et al. [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e] confirmed that machine learning combined with simple clinical and pathologic features showed a better performance in predicting lymph node metastasis of early-stage OTSCC than conventional prediction methods (AUC 0.786 \u003cem\u003evs.\u003c/em\u003e 0.539). However, their model included both preoperative and intraoperative information, which limited its predictive application before surgery. In the current study, all clinical characteristics and radiomics features were from preoperative data, thereby helping in surgical planning.\u003c/p\u003e \u003cp\u003eThe presence of ENE places patients in a higher stage, i.e., to either N2a for single small nodal metastases or N3b for multiple or large nodal metastases [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Identifying ENE by clinical and radiological examination is difficult, thereby leading to unnecessary overtreatment of the neck by surgical or chemoradiotherapy interventions. Currently, no definitive predictors are available for ENE [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Frood et al. [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] found that MRI textural analysis may aid in predicting ENE in oral cancers, with an accuracy of 79%. In the present study, by using the SVM model based on shape features from T2WI and clinical characteristics from ceT1WI, 85.6% of the nodes could be correctly classified preoperatively. This suggests that morphological features combined with machine-learning methods can provide more information and may be a better method for identifying ENE in OTSCC.\u003c/p\u003e \u003cp\u003eWe used the IG algorithm to select potential indicators and constructed the machine-learning models. This method allowed us to incorporate individual radiomics morphological features into a feature panel to perform multi-feature analyses. The present study results indicate that the combination of extracted radiomics and clinical traditional morphological variables has a complementary and synergistic effect in predicting LNM and ENE in patients with OTSCC.\u003c/p\u003e \u003cp\u003eDOI was an important independent prognostic factor for lymph node metastasis and survival in patients with oral cancer. For every 5 mm increase in DOI, the T category increases by a level. Previous studies have indicated that MRI-derived DOI had high agreement with pathological DOI and was an independent prognostic factor for LNM in OTSCC [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Consistently, the current study identified MRI-derived DOI as statistically significant between LNM-positive and LNM-negative groups. Ren et al. [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] reported that MRI-derived DOI yielded an AUC of 0.67 for predicting occult cervical LNM in early-stage OTSCC. Similar to their study, our results showed an AUC of 0.655 when using MRI-derived DOI alone. Although the sensitivity of MRI-derived DOI was very high, with being 91.1% at the optimal cut-off, the specificity was only 42.2%. Therefore, MRI-derived DOI alone cannot be a reliable indicator to predict the occurrence of LNM. Our study indicated that radiomics shape features potentially improve the ability of the MRI-derived DOI to predict LNM.\u003c/p\u003e \u003cp\u003eThere are several limitations in the study. First, this was a retrospective single-institution design without an external validation. Though a stratified 10-fold cross-validation was used to preliminarily verify the feasibility of the study, a prospective multicenter study would be needed for further validation of the models and reduce the risk of overfitting. Secondly, the sample size for analysis was relatively small, especially the proportion of ENE patients. Such imbalance might influence the application of our machine learning models. Future studies should be conducted on a larger cohort, so that we can obtain adequate cases to further train the models and explore the relationship of lymph node status and morphological features. Thirdly, tumors were manually segmented, which may have introduced potential biases, so future research needs a reliable or widely accepted automated segmentation technique to extract the morphological and radiomic parameters of tumors.\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eIn the current study, we developed predictive models for cervical lymph node status in OTSCC using morphological features and machine learning from preoperative MRI. The models may provide a relatively accurate, convenient, and noninvasive method for distinguishing high risk of LNM and ENE in patients with OTSCC.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003e\u003cstrong\u003eMRI\u003c/strong\u003e Magnetic resonance imaging\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eOTSCC\u0026nbsp;\u003c/strong\u003eOral tongue squamous cell carcinoma\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLNM\u003c/strong\u003e Lymph node metastasis\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eENE\u0026nbsp;\u003c/strong\u003eExtranodal extension\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDOI\u0026nbsp;\u003c/strong\u003eDepth of invasion\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eT2W\u003c/strong\u003e T2-weighted\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eceT1W\u0026nbsp;\u003c/strong\u003eContrast-enhanced T1-weighted\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNN\u003c/strong\u003e Neural network\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRF\u003c/strong\u003e Random forest\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLR\u003c/strong\u003e Logistic regression\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSVM\u003c/strong\u003e Support vector machine\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNB\u003c/strong\u003e Na\u0026iuml;ve bayes\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAUC\u0026nbsp;\u003c/strong\u003eArea under the curve\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTNM\u003c/strong\u003e Tumor, node, and metastasis\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eROIs\u003c/strong\u003e Regions of interest\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eICCs\u003c/strong\u003e Intraclass correlation coefficients\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAUC\u003c/strong\u003e Area under the curve\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eROC\u003c/strong\u003e Receiver operating characteristic\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eYunjing Zhu and Ying Yuan wrote the main manuscript text. Jiliang Ren and Yang Song did the analysis and statistics. All authors reviewed the manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eR.L. Siegel, K.D. Miller, A. Jemal, Cancer statistics, 2018, CA Cancer J. Clin. 68 (2018) 7-30. https://doi.org/10.3322/caac.21442.\u003c/li\u003e\n\u003cli\u003eS.B. Chinn, J.N. Myers, Oral Cavity Carcinoma: Current Management, Controversies, and Future Directions, J. Clin. Oncol. 33 (2015) 3269-76. https://doi.org/10.1200/JCO.2015.61.2929.\u003c/li\u003e\n\u003cli\u003eM. Mermod, G. Tolstonog, C. Simon, Y. Monnier, Extracapsular spread in head and neck squamous cell carcinoma: A systematic review and meta-analysis, Oral Oncol. 62 (2016) 60-71. https://doi.org/10.1016/j.oraloncology.2016.10.003.\u003c/li\u003e\n\u003cli\u003eD. Mattavelli, M. Ferrari, S. Taboni, R. Morello, A. Paderno, V. Rampinelli, F. Del Bon, D. Lombardi, A. Grammatica, P. Bossi, A. Deganello, C. Piazza, P. Nicolai, The 8th TNM classification for oral squamous cell carcinoma: What is gained, what is lost, and what is missing, Oral Oncol. 111 (2020). https://doi.org/ARTN 10493710.1016/j.oraloncology.2020.104937.\u003c/li\u003e\n\u003cli\u003eA. Baba, K. Hashimoto, R. Kayama, H. Yamauchi, K. Ikeda, O. Hiroya, Correction to: Radiological approach for the newly incorporated T staging factor, depth of invasion (DOI), of the oral tongue cancer in the 8th edition of American Joint Committee on Cancer (AJCC) staging manual: assessment of the necessity for elective neck dissection, Jpn J Radiol. 39 (2021) 100. https://doi.org/10.1007/s11604-020-01047-8.\u003c/li\u003e\n\u003cli\u003eA. Almangush, A.A. Makitie, A. Triantafyllou, R. de Bree, P. Strojan, A. Rinaldo, J.C. Hernandez-Prera, C. Suarez, L.P. Kowalski, A. Ferlito, I. Leivo, Staging and grading of oral squamous cell carcinoma: An update, Oral Oncol. 107 (2020) 104799. https://doi.org/10.1016/j.oraloncology.2020.104799.\u003c/li\u003e\n\u003cli\u003eW. Tang, Y. Wang, Y. Yuan, X. Tao, Assessment of tumor depth in oral tongue squamous cell carcinoma with multiparametric MRI: correlation with pathology, Eur. Radiol. 32 (2022) 254-261. https://doi.org/10.1007/s00330-021-08148-6.\u003c/li\u003e\n\u003cli\u003eT. Waech, S. Pazahr, V. Guarda, N.J. Rupp, M.A. Broglie, G.B. Morand, Measurement variations of MRI and CT in the assessment of tumor depth of invasion in oral cancer: A retrospective study, Eur. J. Radiol. 135 (2021) 109480. https://doi.org/10.1016/j.ejrad.2020.109480.\u003c/li\u003e\n\u003cli\u003eZ.Y. Peng, Y.M. Wang, Y.X. Wang, S.J. Jiang, R.H. Fan, H. Zhang, W.H. Jiang, Application of radiomics and machine learning in head and neck cancers, Int. J. Biol. Sci. 17 (2021) 475-486. https://doi.org/10.7150/ijbs.55716.\u003c/li\u003e\n\u003cli\u003eA.A. Ardakani, N.J. Bureau, E.J. Ciaccio, U.R. Acharya, Interpretation of radiomics features-A pictorial review, Comput. Methods Programs Biomed. 215 (2022). https://doi.org/ARTN 10660910.1016/j.cmpb.2021.106609.\u003c/li\u003e\n\u003cli\u003eK. Kubo, D. Kawahara, Y. Murakami, Y. Takeuchi, T. Katsuta, N. Imano, I. Nishibuchi, A. Saito, M. Konishi, N. Kakimoto, Y. Yoshioka, S. Toratani, S. Ono, T. Ueda, S. Takeno, Y. Nagata, 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. 134 (2022) 93-101. https://doi.org/10.1016/j.oooo.2021.12.122.\u003c/li\u003e\n\u003cli\u003eY. Yuan, J. Ren, X. Tao, Machine learning-based MRI texture analysis to predict occult lymph node metastasis in early-stage oral tongue squamous cell carcinoma, Eur. Radiol. 31 (2021) 6429-6437. https://doi.org/10.1007/s00330-021-07731-1.\u003c/li\u003e\n\u003cli\u003eJ. Shan, R. Jiang, X. Chen, Y. Zhong, W. Zhang, L. Xie, J. Cheng, H. Jiang, Machine Learning Predicts Lymph Node Metastasis in Early-Stage Oral Tongue Squamous Cell Carcinoma, J. Oral Maxillofac. Surg. 78 (2020) 2208-2218. https://doi.org/10.1016/j.joms.2020.06.015.\u003c/li\u003e\n\u003cli\u003eW.M. Lydiatt, S.G. Patel, B. O\u0026apos;Sullivan, M.S. Brandwein, J.A. Ridge, J.C. Migliacci, A.M. Loomis, J.P. Shah, Head and Neck cancers-major changes in the American Joint Committee on cancer eighth edition cancer staging manual, CA Cancer J. Clin. 67 (2017) 122-137. https://doi.org/10.3322/caac.21389.\u003c/li\u003e\n\u003cli\u003eW. Wu, C. Parmar, P. Grossmann, J. Quackenbush, P. Lambin, J. Bussink, R. Mak, H.J. Aerts, Exploratory Study to Identify Radiomics Classifiers for Lung Cancer Histology, Front. Oncol. 6 (2016) 71. https://doi.org/10.3389/fonc.2016.00071.\u003c/li\u003e\n\u003cli\u003eR. Chai, Q. Wang, P. Qin, J. Zeng, J. Ren, R. Zhang, L. Chu, X. Zhang, Y. Guan, Differentiating Central Lung Tumors from Atelectasis with Contrast-Enhanced CT-Based Radiomics Features, Biomed Res Int. 2021 (2021) 5522452. https://doi.org/10.1155/2021/5522452.\u003c/li\u003e\n\u003cli\u003eR.J. Gillies, P.E. Kinahan, H. Hricak, Radiomics: Images Are More than Pictures, They Are Data, Radiology. 278 (2016) 563-577. https://doi.org/10.1148/radiol.2015151169.\u003c/li\u003e\n\u003cli\u003eJ. Ren, Y. Yuan, X. Tao, 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. (2021). https://doi.org/10.1007/s00330-021-08310-0.\u003c/li\u003e\n\u003cli\u003eP. Bos, M.W.M. van den Brekel, M. Taghavi, Z.A.R. Gouw, A. Al-Mamgani, S. Waktola, H.J.W.L. Aerts, R.G.H. Beets-Tan, J.A. Castelijns, B. Jasperse, Simple delineations cannot substitute full 3d tumor delineations for MR-based radiomics prediction of locoregional control in oropharyngeal cancer, Eur. J. Radiol. 148 (2022). https://doi.org/ARTN 11016710.1016/j.ejrad.2022.110167.\u003c/li\u003e\n\u003cli\u003eF. Wang, R. Tan, K. Feng, J. Hu, Z. Zhuang, C. Wang, J. Hou, X. Liu, Magnetic Resonance Imaging-Based Radiomics Features Associated with Depth of Invasion Predicted Lymph Node Metastasis and Prognosis in Tongue Cancer, J. Magn. Reson. Imaging. 56 (2022) 196-209. https://doi.org/10.1002/jmri.28019.\u003c/li\u003e\n\u003cli\u003eS. Kasai, A. Shiomi, H. Kagawa, H. Hino, S. Manabe, Y. Yamaoka, K. Chen, K. Nanishi, Y. Kinugasa, The Effectiveness of Machine Learning in Predicting Lateral Lymph Node Metastasis From Lower Rectal Cancer: A Single Center Development and Validation Study, Ann Gastroenterol Surg. 6 (2022) 92-100. https://doi.org/10.1002/ags3.12504.\u003c/li\u003e\n\u003cli\u003eA. Jajodia, A. Gupta, H. Prosch, M. Mayerhoefer, S. Mitra, S. Pasricha, A. Mehta, S. Puri, A. Chaturvedi, Combination of Radiomics and Machine Learning with Diffusion-Weighted MR Imaging for Clinical Outcome Prognostication in Cervical Cancer, Tomography. 7 (2021) 344-357. https://doi.org/10.3390/tomography7030031.\u003c/li\u003e\n\u003cli\u003eY. Tang, C.M. Yang, S. Su, W.J. Wang, L.P. Fan, J. Shu, Machine learning-based Radiomics analysis for differentiation degree and lymphatic node metastasis of extrahepatic cholangiocarcinoma, BMC Cancer. 21 (2021) 1268. https://doi.org/10.1186/s12885-021-08947-6.\u003c/li\u003e\n\u003cli\u003eR. Frood, E. Palkhi, M. Barnfield, R. Prestwich, S. Vaidyanathan, A. Scarsbrook, Can MR textural analysis improve the prediction of extracapsular nodal spread in patients with oral cavity cancer?, Eur. Radiol. 28 (2018) 5010-5018. https://doi.org/10.1007/s00330-018-5524-x.\u003c/li\u003e\n\u003cli\u003eC. Xu, J. Yuan, L. Kang, X. Zhang, L. Wang, X. Chen, Q. Yao, H. Li, Significance of depth of invasion determined by MRI in cT1N0 tongue squamous cell carcinoma, Sci. Rep. 10 (2020) 4695. https://doi.org/10.1038/s41598-020-61474-5.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Supplementary Material","content":"\u003cp\u003eSupplementary Material is not available with this version\u003c/p\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":false,"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":"Squamous cell carcinoma of head and neck, Lymphatic metastasis, Extranodal extension, Magnetic resonance imaging, Machine learning","lastPublishedDoi":"10.21203/rs.3.rs-3909740/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3909740/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003ePurpose\u003c/h2\u003e \u003cp\u003eTo develop machine-learning models based on morphological features extracted from preoperative magnetic resonance imaging (MRI) to predict lymph node status in oral tongue squamous cell carcinoma (OTSCC).\u003c/p\u003e\u003ch2\u003eMethod\u003c/h2\u003e \u003cp\u003eThis study retrospectively enrolled 90 OTSCC patients, of whom 45 and 13 patients, respectively, had confirmed lymph node metastasis (LNM) and extranodal extension (ENE). Fourteen morphological features and two customized metrics were derived from T2-weighted (T2W) images. Tumor maximum diameter and MRI-derived depth of invasion (DOI) were measured on contrast-enhanced T1-weighted (ceT1W) images. Information gain algorithm was applied to select the top five attributes. Models were created using six machine-learning methods, including neural network (NN), random forest (RF), logistic regression (LR), support vector machine (SVM), na\u0026iuml;ve bayes (NB), and AdaBoost. An internal stratified 10-fold cross-validation was performed to assess their performance.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eFor predicting LNM, the NN classifier, which included Situation, Elongation, Top Bottom Area, Least Axis Length, and Minor Axis Length, yielded the best model, with an AUC of 0.746 and accuracy of 72.2%. The performance of the NN model was slightly superior to that of MRI-derived DOI (0.746 vs. 0.655), although the difference was not significant (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.122). For predicting ENE, the SVM classifier, which included situation, Elongation, Top Bottom Area, Least Axis Length, and Minor Axis Length, performed the best, with an AUC of 0.750 and accuracy of 85.6%.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eMachine-learning models using MRI morphological features have potential in preoperative evaluation of cervical lymph node status in OTSCC.\u003c/p\u003e","manuscriptTitle":"Morphology-based Machine-Learning for Predicting Lymph Node Status in Oral Tongue Squamous Cell Carcinoma","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-03-19 15:11:37","doi":"10.21203/rs.3.rs-3909740/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":"8c17a723-efc4-4924-be43-638aab6fafda","owner":[],"postedDate":"March 19th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-07-03T05:04:19+00:00","versionOfRecord":[],"versionCreatedAt":"2024-03-19 15:11:37","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3909740","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3909740","identity":"rs-3909740","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}