Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2022 Oct 26;14(21):5266.
doi: 10.3390/cancers14215266.

Evaluation of Cervical Lymph Node Metastasis in Papillary Thyroid Carcinoma Using Clinical-Ultrasound Radiomic Machine Learning-Based Model

Enock Adjei Agyekum  1   2 , Yong-Zhen Ren  1   2 , Xian Wang  1 , Sashana Sashakay Cranston  2 , Yu-Guo Wang  3 , Jun Wang  4 , Debora Akortia  5 , Fei-Ju Xu  1 , Leticia Gomashie  6 , Qing Zhang  1 , Dongmei Zhang  1 , Xiaoqin Qian  1
Affiliations

Evaluation of Cervical Lymph Node Metastasis in Papillary Thyroid Carcinoma Using Clinical-Ultrasound Radiomic Machine Learning-Based Model

Enock Adjei Agyekum et al. Cancers (Basel). .

Abstract

We aim to develop a clinical-ultrasound radiomic (USR) model based on USR features and clinical factors for the evaluation of cervical lymph node metastasis (CLNM) in patients with papillary thyroid carcinoma (PTC). This retrospective study used routine clinical and US data from 205 PTC patients. According to the pathology results, the enrolled patients were divided into a non-CLNM group and a CLNM group. All patients were randomly divided into a training cohort (n = 143) and a validation cohort (n = 62). A total of 1046 USR features of lesion areas were extracted. The features were reduced using Pearson's Correlation Coefficient (PCC) and Recursive Feature Elimination (RFE) with stratified 15-fold cross-validation. Several machine learning classifiers were employed to build a Clinical model based on clinical variables, a USR model based solely on extracted USR features, and a Clinical-USR model based on the combination of clinical variables and USR features. The Clinical-USR model could discriminate between PTC patients with CLNM and PTC patients without CLNM in the training (AUC, 0.78) and validation cohorts (AUC, 0.71). When compared to the Clinical model, the USR model had higher AUCs in the validation (0.74 vs. 0.63) cohorts. The Clinical-USR model demonstrated higher AUC values in the validation cohort (0.71 vs. 0.63) compared to the Clinical model. The newly developed Clinical-USR model is feasible for predicting CLNM in patients with PTC.

Keywords: cervical lymph node metastasis; machine learning; papillary thyroid carcinoma; thyroid neoplasms; ultrasound radiomics.

PubMed Disclaimer

Conflict of interest statement

The authors declare no conflict of interest.

Figures

Figure 7
Figure 7
ROC curve of the classifiers used in building the USR model. ROC, receiver operating characteristic. (A) Support vector classifier with radial basis function kernel. (B) Support vector classifier with linear kernel. (C) Logistic regression classifier. (D) Linear discriminant analysis.
Figure 1
Figure 1
Schematic diagram of the patient selection. PTC, papillary thyroid carcinoma.
Figure 2
Figure 2
Schematic diagram of the radiomic workflow in building the machine learning models.
Figure 3
Figure 3
Selected features after RFE. (A) In the Clinical model, features were reduced to four features in the training cohort. (B) In the USR model, features were reduced to nine features in the training cohort. (C) In the Clinical-USR model, features were reduced to five features in the training cohort.
Figure 3
Figure 3
Selected features after RFE. (A) In the Clinical model, features were reduced to four features in the training cohort. (B) In the USR model, features were reduced to nine features in the training cohort. (C) In the Clinical-USR model, features were reduced to five features in the training cohort.
Figure 4
Figure 4
Recursive feature elimination (RFE) with 15-fold cross-validation; number of features selected vs. cross-validation score. (A) Clinical model. (B) USR model. (C) Clinical-USR model.
Figure 4
Figure 4
Recursive feature elimination (RFE) with 15-fold cross-validation; number of features selected vs. cross-validation score. (A) Clinical model. (B) USR model. (C) Clinical-USR model.
Figure 5
Figure 5
ROC curve of the classifiers used in building the Clinical model. ROC, receiver operating characteristic. (A) Support vector classifier with radial basis function kernel. (B) Support vector classifier with linear kernel. (C) Logistic regression classifier. (D) Linear discriminant analysis.
Figure 6
Figure 6
ROC curve of the US reported status of CLNM by an experienced sonographer. ROC, receiver operating characteristic.
Figure 8
Figure 8
ROC curve of the classifiers used in building the Clinical-USR model. ROC, receiver operating characteristic. (A) Support vector classifier with radial basis function kernel. (B) Support vector classifier with linear kernel. (C) Logistic regression classifier. (D) Linear discriminant analysis.
Figure 9
Figure 9
Confusion matrix. The 2 × 2 contingency table reports the number of true positives, false positives, false negatives, and true negatives: Training cohort (A) and validation cohort (B).
Figure 10
Figure 10
Precision–recall curve of the Clinical-USR model.
See this image and copyright information in PMC

References

    1. Bonjoc K.-J., Young H., Warner S., Gernon T., Maghami E., Chaudhry A. Thyroid cancer diagnosis in the era of precision imaging. J. Thorac. Dis. 2020;12:5128–5139. doi: 10.21037/jtd.2019.08.37. - DOI - PMC - PubMed
    1. Qiu Z., Li H., Wang J., Sun C. miR-146a and miR-146b in the diagnosis and prognosis of papillary thyroid carcinoma. Oncol. Rep. 2017;38:2735–2740. doi: 10.3892/or.2017.5994. - DOI - PMC - PubMed
    1. Lonjou C., Damiola F., Moissonnier M., Durand G., Malakhova I., Masyakin V., Calvez-Kelm L., Cardis E., Byrnes G., Kesminiene A., et al. Investigation of DNA re-pair-related SNPs underlying susceptibility to papillary thyroid carcinoma reveals MGMT as a novel candidate gene in Belarusian children exposed to radiation. BMC Cancer. 2017;17:328. doi: 10.1186/s12885-017-3314-5. - DOI - PMC - PubMed
    1. Schneider D.F., Chen H. New developments in the diagnosis and treatment of thyroid cancer. CA Cancer J. Clin. 2013;63:373–394. doi: 10.3322/caac.21195. - DOI - PMC - PubMed
    1. Lopez-Campistrous A., Adewuyi E.E., Benesch M.G., Ko Y.M., Lai R., Thiesen A., Dewald J., Wang P., Chu K., Ghosh S., et al. PDGFRα Regulates Follicular Cell Differentiation Driving Treatment Resistance and Disease Recurrence in Papillary Thyroid Cancer. eBioMedicine. 2016;12:86–97. doi: 10.1016/j.ebiom.2016.09.007. - DOI - PMC - PubMed

LinkOut - more resources