Abstract
Background: It is necessary to identify patients at risk of developing lymph node metastasis prior to papillary thyroid carcinoma (PTC) surgery. This can be challenging due to limiting factors, and an artificial intelligence algorithm may be a viable option. Objective: In this study, we aimed to evaluate whether combining an artificial intelligence algorithm (support vector machine and probabilistic neural network) and clinico-pathologic data can preoperatively predict lymph node metastasis of papillary thyroid carcinoma (PTC). Methods: We retrospectively examined 251 PTCs with lymph node metastasis and 194 PTCs without lymph node metastasis. The artificial intelligence algorithm included the support vector machine (SVM) and the probabilistic neural network (PNN). Results: The ACR TI-RADS (Thyroid Imaging, Reporting and Data System), number of tumours, no well-defined margin, lymph node status and rim calcification on ultrasonography (US), age, sex, tumour size, and presence of Hashimoto’s thyroiditis were significantly more frequent among PTCs with central lymph node metastasis than those without metastasis (P<0.05). The PNN classifier revealed an F1 score of 0.88 on the central lymph node metastasis test set. The SVM classifier revealed an F1 score of 0.93 on the lateral lymph node metastasis test set. Our study demonstrates that combining artificial intelligence algorithms and clinico-pathologic data can effectively predict the lymph node metastasis of papillary thyroid carcinoma prior to surgery.
Keywords: Artificial intelligence algorithm, papillary thyroid carcinoma, lymph node metastasis
Introduction
The incidence of thyroid cancer has grown rapidly over the last decade, making it the most common endocrine malignancy [1,2]. Along with advances in fine-needle aspiration biopsy (FNAB) and ultrasonography (US), early diagnosis and surgery of thyroid cancer have increased [3-5]. Papillary thyroid carcinoma (PTC) is the most common sub-type of thyroid carcinoma, accounting for 85% [6]. Although the prognosis for PTC is excellent, up to 10% of PTC patients suffer regional or distant metastatic recurrence after surgical treatment [7]. Lymph node metastasis is a very important predictor of local recurrence and distant metastasis of PTC. According to prior studies, patients with PTC generally have been found to have the following predictors for lymph node metastasis, which fall into three main categories: demographic, ultrasound, and pathological characteristics. The most commonly reported factors include age, sex, tumour size and location, multifocality, bilaterality, the presence of Hashimoto’s thyroiditis, the presence of BRAF (V600E) mutations, and lymph node abnormalities found on ultrasound [8-10]. Therefore, the development of a predictive model for PTC recurrence, such as an artificial intelligence algorithm, has the possibility to improve the clinical management of patients by aiding in risk stratification.
Artificial intelligence algorithms have been around for decades and, more recently, have been applied to medical research [11-13]. Artificial intelligence algorithms consist of SVM, PNN, artificial neural networks (ANNs), and various other intelligence algorithms [14-18]. Liu et al. found that, compared to pathologists, artificial intelligence algorithms were able to achieve higher tumour-level sensitivity [19]. Wang et al. discovered that combining artificial intelligence algorithms and CT images was able to predict EGFR-mutation status in lung adenocarcinoma patients [20]. Researchers demonstrated that, compared to dermatologists, deep neural networks were more accurate in identifying skin cancer [21]. All of these studies imply that artificial intelligence algorithms have proven to be applicable in medical testing and diagnostics.
In our study, we retrospectively examined 251 PTCs with lymph node metastasis and 194 PTCs without lymph node metastasis. We found that combining artificial intelligence algorithms and clinico-pathologic data was able to effectively predict lymph node metastasis of PTCs prior to surgery.
Materials and methods
Patient selection and data collection
This study included 445 patients with PTC who underwent radical thyroid cancer surgery at the First Affiliated Hospital of Wenzhou Medical University between January 2018 and June 2020 and who underwent preventive central lymph node dissection; of these, 128 patients underwent further lateral lymph node dissection. All pathological diagnoses were made by two experienced pathologists. None of the patients had a history of neck surgery or neck irradiation. This study retrospectively analysed electronic pathology and clinical data and collected clinico-pathologic data. The research protocol was approved by the Ethics Committee of the First Affiliated Hospital of Wenzhou Medical University (approval no. 2018-40).
Ultrasound examination and image evaluation
The ultrasound examinations were carried out with the help of Acuson Sequoia and 128XP sonographic scanners (Siemens Medical Solutions, Mountain View, CA, USA) using 8-13 MHz linear probes. Selection criteria for investigating the risk of central lymph node metastasis are indicated in Table 1, and selection criteria for analysing the risk of lateral lymph node metastasis are shown in Table 2. Largest tumour diameter size was measured using B-mode ultrasound. Multifocal PTMC was defined as the presence of more than one lesion in a given paraffin section; the lesion with the largest diameter or the most suspicious nodule was analysed. The margin was categorized as well defined or not well defined. B-mode ultrasound was used to determine the state of the lymph node, skin, and medullary junction and whether to explore enlarged lymph nodes. Based on preoperative ultrasound images, tumour location was classified into four domains, i.e., upper, middle, lower, and isthmus.
Table 1.
Baseline characteristics of the 445 patients with TPC with and without metastatic central lymph nodes
| Characteristics | The status of metastatic central lymph nodes | P | |
|---|---|---|---|
|
| |||
| Positive (n=251) | Negative (n=194) | ||
| Age (y) | 44.28±12.67 | 48.03±11.28 | 0.001 |
| Sex | <0.0001 | ||
| Female | 167 | 166 | |
| Male | 84 | 28 | |
| Tumour size (mm) | 12.93±8.28 | 8.86±5.16 | <0.0001 |
| Hashimoto’s thyroiditis | 0.024 | ||
| Yes | 46 | 53 | |
| No | 205 | 141 | |
| No. of tumour | 0.002 | ||
| Unilateral unifocal | 146 | 144 | |
| Unilateral multifocal | 25 | 10 | |
| Bilateral multifocal | 80 | 40 | |
| Tumour location | 0.548 | ||
| Lower | 61 | 52 | |
| Upper/middle/isthmus | 190 | 142 | |
| Margin | 0.001 | ||
| Well defined | 151 | 147 | |
| Not well defined | 100 | 47 | |
| Calcification | <0.0001 | ||
| Present | 211 | 123 | |
| Absent | 40 | 71 | |
| Lymph node status | <0.0001 | ||
| Clear junction of LN, skin and marrow | 16 | 10 | |
| Unclear junction of LN, skin and marrow | 82 | 19 | |
| Absent | 153 | 165 | |
| TI-RADS | <0.0001 | ||
| III | 7 | 7 | |
| IVa | 76 | 75 | |
| IVb | 59 | 72 | |
| IVc | 66 | 22 | |
| V | 21 | 7 | |
| VI | 22 | 11 | |
Table 2.
Multivariate logistic regression analysis associated with central lymph node metastases of PTC
| Clinico-pathologic characteristics | OR | 95% CI | P |
|---|---|---|---|
| Age | 0.979 | 0.960-0.997 | 0.026 |
| Sex | |||
| Male | 3.24 | 1.872-5.606 | <0.001 |
| Tumour size | 1.088 | 1.043-1.136 | <0.001 |
| Hashimoto’s thyroiditis | |||
| Yes | 0.847 | 0.498-1.439 | 0.539 |
| No. of tumour | |||
| Unilateral unifocal | 0.002 | ||
| Unilateral multifocal | 2.359 | 1.019-5.461 | 0.045 |
| Bilateral multifocal | 2.345 | 1.398-3.934 | 0.001 |
| Margina | |||
| Not well defined | 1.657 | 1.018-2.696 | 0.042 |
| Calcificationa | |||
| Present | 1.886 | 1.116-3.187 | 0.018 |
| Lymph node statusa | |||
| Absent | 0.007 | ||
| Clear junction of LN, skin and marrow | 0.954 | 0.378-2.409 | 0.920 |
| Unclear junction of LN, skin and marrow | 2.73 | 1.440-5.174 | 0.002 |
| TI-RADSa | |||
| III | 0.162 | ||
| IVa | 2.38 | 0.618-9.168 | 0.207 |
| IVb | 1.427 | 0.367-5.555 | 0.608 |
| IVc | 3.063 | 0.741-12.668 | 0.122 |
| V | 2.481 | 0.462-13.323 | 0.289 |
| VI | 2.943 | 0.632-13.695 | 0.169 |
Thyroid nodules observed under Ultrasound.
Model development
We retrospectively used support vector machine (SVM) and probabilistic neural network (PNN) algorithms for the development of statistical models to predict central lymph node metastasis and lateral lymph node metastasis based on clinico-pathologic factors and B-mode ultrasound features. We used the MATLAB 2014a modelling platform to establish the SVM model using the RBF kernel function, and the PNN model using the newpnn function. Based on the Kennard-Stone algorithm (KS), central lymph node metastasis is divided into 350 training sets and 95 test sets, and the lateral lymph node metastasis is divided into 90 training sets and 38 test sets. Also, during the development of the SVM model, a grid search method was used to debug the model. When the highest verification accuracy rate is reached, the obtained c and g are the best parameters. In the grid search method, the global search is performed at intervals of 0.5, and the ranges of c and g are both 2-10 to 210. The RBF kernel function is K(xi,xj) = exp(-gIIxi-xjII)2. The kernel function parameter is g, which controls the range of the input space. The above optimization problem is transformed into (Equation 1):
 |
The SVM prediction problem depends on two important parameters, c and g. The parameter c, which is common to all SVM kernels, can compete with the simplicity of the decision-making surface and can be a valuable conversion to the misclassification of training samples. A smaller c makes the decision surface smoother, while a larger c tends to correctly classify all training samples. The variable g defines how much impact a single training sample can have. A larger g will affect other samples more. We used the newpnn function that comes with the MATLAB 2014a modelling platform to build the PNN model.
Newpnn function net = newpnn (P.T. spread)
P: R*Q matrix, including Q input vectors of length R.
T: S*Q matrix, including Q target vectors.
Spread: scalar, representing the spreading speed of the probabilistic neural network; the default value is 0.1. At the same time, we used the F1 score and kappa coefficient to evaluate the predictive performance of the classifier. Our patients were divided into two groups. A positive value indicated positive lymph node metastasis, and a negative value indicated negative lymph node metastasis.
Statistical analysis and combined model
PTCs with and without lymph nodes metastasis were compared based on patient demographic information. The application of an independent two-sample t-test was built to carry out the evaluation of records in a normal distribution. Categorical variables were analysed using the chi-square test, while continuous variables were analysed using Student’s t-test. Variables with P<0.05 in the univariate analysis were advanced to a multivariate analysis by means of forward stepwise selection. The results are presented as odds ratio (OR) and 95% CI as well as P-value. All P-values were two-sided. Furthermore, the P-value <0.05 was considered statistically significant. Statistical analysis was carried out using SPSS software version 22.0 (SPSS, Chicago, IL, USA).
We retrospectively used support vector machine (SVM) and probabilistic neural network (PNN) algorithms for the development of statistical models to predict central lymph node metastasis and lateral lymph node metastasis based on clinico-pathologic factors and B-mode ultrasound features. We used the MATLAB 2014a modelling platform to establish the SVM model using the RBF kernel function and the PNN model using the newpnn function. Also, during the development of the SVM model, a grid search method was used to debug the model. Our patients were divided into two groups. A positive value indicated the presence of lymph node metastasis and a negative value indicated the absence of lymph node metastasis.
Results
Relationship of central lymph node metastasis of PTC with clinico-pathologic and US features
In our study, we enrolled 445 PTCs, including 251 PTCs with central lymph node metastasis and 194 PTCs without central lymph node metastasis. To understand the central lymph node metastasis of PTCs, we investigated the relationship of central lymph node metastasis with clinico-pathologic factors and ultrasound features (Table 1). The results indicated that PTCs with central lymph node metastasis occurred at a younger age (P=0.001); with a larger tumour size (P<0.001); in the presence of calcification (P<0.001); with no well-defined margin (P=0.001); with unclear junction of lymph node, skin and marrow (P<0.001); at higher levels of TI-RADS (P≤0.001); and with a higher number of thyroid cancer nodules (P=0.002) than PTCs without central lymph node metastasis. Men (P<0.001) and PTCs with concurrent Hashimoto’s thyroiditis (P=0.024) had a greater chance of central lymph node metastasis. We used multivariate logistic regression to determine whether these elements were independently associated with central lymph node metastasis (Table 2). Our analyses suggest that men (P<0.001); younger age (P=0.026); larger tumour size (P<0.001); no well-defined margin (P=0.001); an unclear junction of lymph node, skin and marrow (P=0.007); and the presence of calcification (P=0.018) were independently predictive of central lymph node metastasis.
Relationship of central lymph node metastasis of PTC with clinico-pathologic and US features
A total of 128 PTCs underwent both central and lateral lymph node dissection. To better comprehend lateral lymph node metastasis of PTCs, we further explored the relationship between lateral lymph node metastasis and clinico-pathologic factors and US features (Table 3). Our results reveal that PTCs associated with Hashimoto’s thyroiditis (P=0.004) may negatively predict lateral lymph node metastasis, whereas larger tumour size (P=0.045), no well-defined margin (P=0.02), and unclear junction of lymph node, skin and marrow (P<0.001) have a higher probability of lateral lymph node metastasis. In addition, men with central lymph node metastasis have increased odds of having lateral lymph node metastasis (P<0.001). Multivariate logistic regression was used to obtain a deeper understanding (Table 4) and revealed that unclear junction of lymph node, skin and marrow (P=0.034) and the presence of central lymph node metastasis (P<0.001) were independently predictive of lateral lymph node metastasis.
Table 3.
Baseline characteristics of the 128 patients with TPC with and without metastatic lateral lymph nodes
| Characteristics | The status of metastatic lateral lymph nodes | P | |
|---|---|---|---|
|
| |||
| Positive (n=87) | Negative (n=41) | ||
| Age | 44.08±14.34 | 45.95±11.77 | 0.468 |
| Sex | 0.108 | ||
| Female | 58 | 33 | |
| Male | 29 | 8 | |
| Tumour size | 16.24±10.59 | 12.46±7.99 | 0.045 |
| Hashimoto’s thyroiditis | 0.004 | ||
| Yes | 14 | 16 | |
| No | 73 | 25 | |
| No. of tumour | 0.378 | ||
| Unilateral unifocal | 55 | 31 | |
| Unilateral multifocal | 6 | 2 | |
| Bilateral multifocal | 26 | 8 | |
| Central lymph nodes metastasis | <0.001 | ||
| Yes | 77 | 14 | |
| No | 10 | 27 | |
| Tumour location | 0.132 | ||
| Upper | 23 | 4 | |
| Middle | 44 | 22 | |
| Lower | 16 | 12 | |
| Isthmus | 4 | 3 | |
| Margin | 0.02 | ||
| Well defined | 47 | 31 | |
| Not well defined | 40 | 10 | |
| Calcification | 0.757 | ||
| Present | 76 | 6 | |
| Absent | 11 | 11 | |
| Lymph node status | 0.001 | ||
| Clear junction of skin and marrow | 5 | 8 | |
| Unclear junction of skin and marrow | 69 | 19 | |
| Absence | 13 | 14 | |
| TI-RADS | 0.201 | ||
| III | 1 | 1 | |
| IVa | 6 | 7 | |
| IVb | 22 | 12 | |
| IVc | 39 | 12 | |
| V | 14 | 4 | |
| VI | 5 | 5 | |
Table 4.
Multivariate logistic regression analysis associated with lateral lymph node metastases of PTC
| Clinico-pathologic characteristics | OR | 95% CI | P |
|---|---|---|---|
| Age | 1.016 | 0.977-1.056 | 0.43 |
| Sex | |||
| Male | 2.396 | 0.672-8.541 | 0.178 |
| Tumour size | 1.029 | 0.968-1.093 | 0.359 |
| Hashimoto’s thyroiditis | |||
| Yes | 0.327 | 0.104-1.032 | 0.057 |
| Central lymph nodes metastasis | |||
| Yes | 14.956 | 4.673-47.861 | <0.001 |
| Margina | |||
| Not well defined | 2.157 | 0.161-1.339 | 0.155 |
| Lymph node statusa | |||
| Absence | 0.034 | ||
| Clear junction of skin and marrow | 0.091 | 0.011-0.731 | 0.024 |
| Unclear junction of skin and marrow | 0.849 | 0.234-3.078 | 0.804 |
Thyroid nodules observed under Ultrasound.
The predictive model
We constructed a model to predict central lymph node metastasis using a PNN algorithm. We used the Kennard-Stone (KS) algorithm, among others, to divide 445 patients into a training set of 350 cases and a test set of 95 cases. In the test set, 51 cases were pathologically diagnosed as lymph node positive, and 44 cases were pathologically diagnosed as lymph node negative. The prediction results are shown in Table 5. When spread is 0.5, accuracy is 0.88, sensitivity is 0.80, and specificity is 0.98. The PNN classifier showed a high F1 score and kappa on the test folds (F1 score =0.88, kappa =0.84). The sensitivity and specificity of the PNN model were found to be 80.4% and 97.7%, respectively (Table 6). The model had a satisfactory predictive value when compared to other single predictors.
Table 5.
Diagnostic performance of the predictive factors for central lymph node metastasis of PTC using PNN
| Predictors | ||
|---|---|---|
| Training performance | Accuracy | 1 |
| Sensitivity | 1 | |
| Specificity | 1 | |
| Test performance | Accuracy | 0.884211 |
| Sensitivity | 0.803921 | |
| Specificity | 0.977273 | |
| F1 score | 0.8842 | |
| Kappa | 0.8439 | |
Table 6.
The predictive value of the PNN model and independent predictors
| Independent predictors | Sensitivity (%) | Specificity (%) |
|---|---|---|
| Sex | 33.5 | 85.6 |
| Tumour location | 24.3 | 73.2 |
| No. of tumour | 41.8 | 74.2 |
| Margin | 39.8 | 75.8 |
| Calcification | 84.1 | 36.6 |
| Lymph node status | 39 | 85.1 |
| PNN model | 80.4 | 97.7 |
The SVM algorithm was used to predict lateral lymph node metastasis of PTCs. Using the KS algorithm, 128 patients were divided into the training set (90 patients) and the test set (38 patients). Among the test set, 27 patients were pathologically diagnosed as lymph node positive, and 11 patients were pathologically diagnosed as lymph node negative. The prediction results are shown in Table 7. Accuracy was 0.95, sensitivity was 0.96, specificity was 0.91, the best c was 0.75786, and the best g was 0.14359. The performance of the SVM classifier showed the F1 score to be 0.93 and the kappa coefficient to be 0.94. The sensitivity and specificity for the SVM model were found to be 100% and 81.8%, respectively (Table 8). The model had a satisfactory predictive value when compared to other single predictors.
Table 7.
Diagnostic performance of the predictive factors for lateral lymph node metastasis of PTC using SVM
| Predictors | ||
|---|---|---|
| Training performance | Accuracy | 0.84444 |
| Sensitivity | 0.96667 | |
| Specificity | 0.6 | |
| Test performance | Accuracy | 0.947368 |
| Sensitivity | 1 | |
| Specificity | 0.8181 | |
| F1 score | 0.9321 | |
| Kappa | 0.9435 | |
Table 8.
The predictive value for the SVM model and independent predictors
| Independent predictors | Sensitivity (%) | Specificity (%) |
|---|---|---|
| Hashimoto’s thyroiditis | 83.9 | 39 |
| Central lymph nodes metastasis | 88.5 | 65.9 |
| SVM model | 100 | 81.8 |
Discussion
With the advancement of preoperative sonography and ultrasound-guided FNAB, the diagnosis and treatment of PTC have developed rapidly in recent years [22-24]. Although most of the PTCs had a slow clinical course together with a good prognosis, the incidence of lymph node metastasis was identified as an important poor prognostic indicator [25,26]. Although lymph node dissection for PTC may be one way to decrease the probability of recurrence and metastasis, an aggressive treatment approach for PTC may lead to negative results, such as injury to the laryngeal nerve leading to damage to the voice [27,28]. Hay et al. reported that aggressive treatment has nothing to do with prolonged survival or reduced recurrence in PTC [26]. This is the reason that some PTC patients can choose to be monitored rather than aggressively treated. Although PTC is a disease with an excellent outcome, patients presenting with metastatic disease do have poor outcomes [29]. In clinical practice, the decision for surgery mainly depends on the individual doctor or the patient. To better treat patients, we urgently need a clinical approach that can predict whether lymph nodes have metastasized prior to surgery.
Numerous articles have elucidated multiple and varied factors related to lymph node metastasis [30-33]. Medas et al. found that the incidence of lymph node metastasis in PTMC is considerable and that tumour size appears to be the most significant predictive factor for lymph node metastasis [10]. Kim et al. reported that, as the size of thyroid nodules increase, so does the likelihood of lymph node metastasis [34]. In addition, Wang et al. have found that no well-defined margin was a significant ultrasound feature of aggressive PTC [35]. Tian et al. found that age, sex, ultrasound-reported LN status and ultrasound signature were the strongest predictors of CLNM in PTC patients, which has aided in the development of a predictive nomogram [36]. In our research, men, younger age, larger tumour size, the presence of calcifications, no well-defined margin, an unclear junction of the lymph node, skin and marrow, higher levels of TI-RADS, greater numbers of thyroid cancer nodules, and PTCs associated with Hashimoto’s thyroiditis were significantly linked to central lymph node metastasis. Multivariate logistic regression revealed that men, younger age, larger tumour size, no well-defined margin, an unclear junction of the lymph node, skin and marrow, and the presence of calcification were independently predictive of central lymph node metastasis. Furthermore, PTCs associated with Hashimoto’s thyroiditis (P=0.004) may be an inhibitor of lateral lymph node metastasis, whereas larger tumour size (P=0.045), no well-defined margin (P=0.02), and an unclear junction of the lymph node, skin and marrow (P<0.001) have a higher probability of lateral lymph node metastasis. Multivariate logistic regression showed that an unclear junction of the lymph node, skin and marrow and the presence of central lymph node metastasis were independently predictive of lateral lymph node metastasis.
The diagnosis of lymph node metastasis showed low sensitivity by preoperative sonography. We combined clinico-pathologic features and ultrasonographic findings to enhance the diagnostic accuracy of lymph node metastasis by PNN and SVM algorithms. In our study, we used one set of features for the PNN classifier (on average 10 features across different folds) from 455 samples (251 PTCs with central lymph node metastasis and 194 PTCs without metastasis). By contrast, we used a different set of features for the SVM classifier from 128 samples (87 PTCs with lateral lymph node metastasis and 41 PTCs without lateral lymph node metastasis). The PNN classifier showed a high F1 score and kappa coefficient on the test folds (F1 score =0.88, kappa =0.84), while the performance of SVM classifiers showed an F1 score of 0.93 and kappa of 0.94.
Despite these findings, there remain limitations to our study. First, in those patients with multifocal nodules, we investigated the most suspicious or the largest lesions only in those instances where data about other nodules were not available. Second, to generate more convincing results, the quantity of the validated set and test set should be expanded.
Conclusion
In summary, our study demonstrated that combining an artificial intelligence algorithm and clinico-pathologic data can effectively predict lymph node metastasis of papillary thyroid carcinoma prior to surgery.
Acknowledgements
Written, informed consent was obtained from each individual participant.
Disclosure of conflict of interest
None.
References
- 1.Siegel RL, Miller KD, Fuchs HE, Jemal A. Cancer statistics, 2021. CA Cancer J Clin. 2021;71:7–33. doi: 10.3322/caac.21654. [DOI] [PubMed] [Google Scholar]
- 2.Ciavardelli D, Bellomo M, Consalvo A, Crescimanno C, Vella V. Metabolic alterations of thyroid cancer as potential therapeutic targets. Biomed Res Int. 2017;2017:2545031. doi: 10.1155/2017/2545031. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Nikiforova MN, Mercurio S, Wald AI, Barbi de Moura M, Callenberg K, Santana-Santos L, Gooding WE, Yip L, Ferris RL, Nikiforov YE. Analytical performance of the ThyroSeq v3 genomic classifier for cancer diagnosis in thyroid nodules. Cancer. 2018;124:1682–1690. doi: 10.1002/cncr.31245. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Valabrega S, Santolamazza G, Romanelli F, Scapicchio G, D’Angelo F, Bellotti C, Aurello P, Izzo L, Giovagnoli MR, Trimboli P. Cancer rate of the indeterminate lesions at low or high risk according to Italian system for reporting of thyroid FNA. Front Endocrinol (Lausanne) 2018;9:371. doi: 10.3389/fendo.2018.00371. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Hu Z, Yang B, Li T, Li J. Thyroid cancer detection by ultrasound molecular imaging with shp2-targeted perfluorocarbon nanoparticles. Contrast Media Mol Imaging. 2018;2018:8710862. doi: 10.1155/2018/8710862. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Randle RW, Balentine CJ, Leverson GE, Havlena JA, Sippel RS, Schneider DF, Pitt SC. Trends in the presentation, treatment, and survival of patients with medullary thyroid cancer over the past 30 years. Surgery. 2017;161:137–146. doi: 10.1016/j.surg.2016.04.053. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Aragon Han P, Weng CH, Khawaja HT, Nagarajan N, Schneider EB, Umbricht CB, Witwer KW, Zeiger MA. MicroRNA expression and association with clinicopathologic features in papillary thyroid cancer: a systematic review. Thyroid. 2015;25:1322–1329. doi: 10.1089/thy.2015.0193. [DOI] [PubMed] [Google Scholar]
- 8.Sahin S, Daglar G, Menekse E, Cavdarli B, Baglan T. The effect of BRAF (V600E) mutation on lymph node involvement in papillary thyroid cancer. Turk J Surg. 2020;36:249–255. doi: 10.47717/turkjsurg.2020.4696. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Yu J, Deng Y, Liu T, Zhou J, Jia X, Xiao T, Zhou S, Li J, Guo Y, Wang Y, Zhou J, Chang C. Lymph node metastasis prediction of papillary thyroid carcinoma based on transfer learning radiomics. Nat Commun. 2020;11:4807. doi: 10.1038/s41467-020-18497-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Medas F, Canu GL, Cappellacci F, Boi F, Lai ML, Erdas E, Calo PG. Predictive factors of lymph node metastasis in patients with papillary microcarcinoma of the thyroid: retrospective analysis on 293 cases. Front Endocrinol (Lausanne) 2020;11:551. doi: 10.3389/fendo.2020.00551. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Heremans ERM, Chen AS, Wang X, Cheng J, Xu F, Martinez AE, Lazaridis G, Van Huffel S, Chen JDZ. Artificial neural network-based automatic detection of food intake for neuromodulation in treating obesity and diabetes. Obes Surg. 2020;30:2547–2557. doi: 10.1007/s11695-020-04511-6. [DOI] [PubMed] [Google Scholar]
- 12.Abouzari M, Goshtasbi K, Sarna B, Khosravi P, Reutershan T, Mostaghni N, Lin HW, Djalilian HR. Prediction of vestibular schwannoma recurrence using artificial neural network. Laryngoscope Investig Otolaryngol. 2020;5:278–285. doi: 10.1002/lio2.362. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Ahmad I, Muhammad Asad S. Predictions of coronavirus COVID-19 distinct cases in Pakistan through an artificial neural network. Epidemiol Infect. 2020;148:e222. doi: 10.1017/S0950268820002174. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Wang X, Luo F, Qian Y, Ranzi G. A personalized electronic movie recommendation system based on support vector machine and improved particle swarm optimization. PLoS One. 2016;11:e0165868. doi: 10.1371/journal.pone.0165868. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Pedretti G, Milo V, Ambrogio S, Carboni R, Bianchi S, Calderoni A, Ramaswamy N, Spinelli AS, Ielmini D. Memristive neural network for on-line learning and tracking with brain-inspired spike timing dependent plasticity. Sci Rep. 2017;7:5288. doi: 10.1038/s41598-017-05480-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Guo J, Chen Q, Lam CW, Wang C, Wong VK, Xu F, Jiang Z, Zhang W. Application of artificial neural network to investigate the effects of 5-fluorouracil on ribonucleotides and deoxyribonucleotides in HepG2 cells. Sci Rep. 2015;5:16861. doi: 10.1038/srep16861. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Liu S, Yang R, Yang J, Yi T, Song H, Jiang M, Tripathi DK, Ma M, Chen Q. Differentiating thamnocalamus munro from fargesia franchet emend. Yi (Bambusoideae, Poaceae): novel evidence from morphological and neural-network analyses. Sci Rep. 2017;7:4192. doi: 10.1038/s41598-017-04613-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Toma C, Manganaro A, Raitano G, Marzo M, Gadaleta D, Baderna D, Roncaglioni A, Kramer N, Benfenati E. QSAR models for human carcinogenicity: an assessment based on oral and inhalation slope factors. Molecules. 2020;26:127. doi: 10.3390/molecules26010127. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Liu Y, Kohlberger T, Norouzi M, Dahl GE, Smith JL, Mohtashamian A, Olson N, Peng LH, Hipp JD, Stumpe MC. Artificial intelligence-based breast cancer nodal metastasis detection: insights into the black box for pathologists. Arch Pathol Lab Med. 2019;143:859–868. doi: 10.5858/arpa.2018-0147-OA. [DOI] [PubMed] [Google Scholar]
- 20.Wang S, Shi J, Ye Z, Dong D, Yu D, Zhou M, Liu Y, Gevaert O, Wang K, Zhu Y, Zhou H, Liu Z, Tian J. Predicting EGFR mutation status in lung adenocarcinoma on computed tomography image using deep learning. Eur Respir J. 2019;53:1800986. doi: 10.1183/13993003.00986-2018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM, Thrun S. Dermatologist-level classification of skin cancer with deep neural networks. Nature. 2017;542:115–118. doi: 10.1038/nature21056. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Sulejmanovic M, Cickusic AJ, Salkic S. The value of fine-needle aspiration biopsy (FNAB) in differential diagnosis of scintigraphic cold thyroid nodule. Acta Inform Med. 2019;27:114–118. doi: 10.5455/aim.2019.27.114-118. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Karadeniz E, Yur M, Temiz A, Akcay MN. Malignancy risk for thyroid nodules larger than 4 cm and diagnostic reliability of ultrasound-guided FNAB results. Turk J Surg. 2019;35:13–18. doi: 10.5578/turkjsurg.4029. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Mahmoudian-Sani MR, Amrollahi-Sharifabadi M, Taheri A, Hosseini SM, Tahmasebi K, Mobini GR. Diagnostic value of the candidate microRNAs in thyroid fine-needle aspiration biopsy (FNAB) samples. Horm Mol Biol Clin Investig. 2020;41 doi: 10.1515/hmbci-2019-0027. [DOI] [PubMed] [Google Scholar]
- 25.Wu Y, Rao K, Liu J, Han C, Gong L, Chong Y, Liu Z, Xu X. Machine learning algorithms for the prediction of central lymph node metastasis in patients with papillary thyroid cancer. Front Endocrinol (Lausanne) 2020;11:577537. doi: 10.3389/fendo.2020.577537. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Hay ID, Hutchinson ME, Gonzalez-Losada T, McIver B, Reinalda ME, Grant CS, Thompson GB, Sebo TJ, Goellner JR. Papillary thyroid microcarcinoma: a study of 900 cases observed in a 60-year period. Surgery. 2008;144:980–987. doi: 10.1016/j.surg.2008.08.035. discussion 987-988. [DOI] [PubMed] [Google Scholar]
- 27.Conzo G, Tartaglia E, Avenia N, Calo PG, de Bellis A, Esposito K, Gambardella C, Iorio S, Pasquali D, Santini L, Sinisi MA, Sinisi AA, Testini M, Polistena A, Bellastella G. Role of prophylactic central compartment lymph node dissection in clinically N0 differentiated thyroid cancer patients: analysis of risk factors and review of modern trends. World J Surg Oncol. 2016;14:149. doi: 10.1186/s12957-016-0879-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Conzo G, Mauriello C, Docimo G, Gambardella C, Thomas G, Cavallo F, Tartaglia E, Napolitano S, Varriale R, Rossetti G, Fei L, Santini L. Clinicopathological pattern of lymph node recurrence of papillary thyroid cancer. Implications for surgery. Int J Surg. 2014;12(Suppl 1):S194–197. doi: 10.1016/j.ijsu.2014.05.010. [DOI] [PubMed] [Google Scholar]
- 29.Ricarte-Filho J, Ganly I, Rivera M, Katabi N, Fu W, Shaha A, Tuttle RM, Fagin JA, Ghossein R. Papillary thyroid carcinomas with cervical lymph node metastases can be stratified into clinically relevant prognostic categories using oncogenic BRAF, the number of nodal metastases, and extra-nodal extension. Thyroid. 2012;22:575–584. doi: 10.1089/thy.2011.0431. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Zou Q, Ma S, Zhou X. Association of sonographic features and clinicopathologic factors of papillary thyroid microcarcinoma for prevalence of lymph node metastasis: a retrospective analysis. Arch Endocrinol Metab. 2020:2359-3997000000297. doi: 10.20945/2359-3997000000297. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Jiwang L, Yahong L, Kai L, Bo H, Yuejiao Z, Haotian W, Tao Y. Clinicopathologic factors and preoperative ultrasonographic characteristics for predicting central lymph node metastasis in papillary thyroid microcarcinoma: a single center retrospective study. Braz J Otorhinolaryngol. 2020 doi: 10.1016/j.bjorl.2020.05.004. S1808-8694(20)30062-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Yu QA, Ma DK, Liu KP, Wang P, Xie CM, Wu YH, Dai WJ, Jiang HC. Clinicopathologic risk factors for right paraesophageal lymph node metastasis in patients with papillary thyroid carcinoma. J Endocrinol Invest. 2018;41:1333–1338. doi: 10.1007/s40618-018-0874-4. [DOI] [PubMed] [Google Scholar]
- 33.Zhu Y, Zheng K, Zhang H, Chen L, Xue J, Ding M, Wu K, Wang Z, Kong L, Chen X. The clinicopathologic differences of central lymph node metastasis in predicting lateral lymph node metastasis and prognosis in papillary thyroid cancer associated with or without Hashimoto’s thyroiditis. Tumour Biol. 2016;37:8037–8045. doi: 10.1007/s13277-015-4706-2. [DOI] [PubMed] [Google Scholar]
- 34.Kim BY, Jung CH, Kim JW, Lee SW, Kim CH, Kang SK, Mok JO. Impact of clinicopathologic factors on subclinical central lymph node metastasis in papillary thyroid microcarcinoma. Yonsei Med J. 2012;53:924–930. doi: 10.3349/ymj.2012.53.5.924. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Wang WH, Xu SY, Zhan WW. Clinicopathologic factors and thyroid nodule sonographic features for predicting central lymph node metastasis in papillary thyroid microcarcinoma: a retrospective study of 1204 patients. J Ultrasound Med. 2016;35:2475–2481. doi: 10.7863/ultra.15.10012. [DOI] [PubMed] [Google Scholar]
- 36.Tian X, Song Q, Xie F, Ren L, Zhang Y, Tang J, Zhang Y, Jin Z, Zhu Y, Zhang M, Luo Y. Papillary thyroid carcinoma: an ultrasound-based nomogram improves the prediction of lymph node metastases in the central compartment. Eur Radiol. 2020;30:5881–5893. doi: 10.1007/s00330-020-06906-6. [DOI] [PubMed] [Google Scholar]