Machine Learning-Based Shear Wave Elastography Elastic Index (SWEEI) in Predicting Cervical Lymph Node Metastasis of Papillary Thyroid Microcarcinoma: A Comparative Analysis of Five Practical Prediction Models

Abstract

Purpose

Although many factors determine the prognosis of papillary thyroid carcinoma (PTC), cervical lymph node metastasis (CLNM) is one of the most terrible factors. In view of this, this study aimed to build a CLNM prediction model for papillary thyroid microcarcinoma (PTMC) with the help of machine learning algorithm.

Methods

We retrospectively analyzed 387 PTMC patients hospitalized in the Department of Medical Oncology, Enshi Tujia and Miao Autonomous Prefecture Central Hospital from January 1, 2015, to January 31, 2022. Based on supervised learning algorithms, namely random forest classifier (RFC), artificial neural network(ANN), support vector machine(SVM), decision tree(DT), and extreme gradient boosting gradient(XGboost) algorithm, the LNM prediction model was constructed, and the prediction efficiency of ML-based model was evaluated via receiver operating characteristic curve(ROC) and decision curve analysis(DCA).

Results

Finally, a total of 24 baseline variables were included in the supervised learning algorithm. According to the iterative analysis results, the pulsatility index(PI), resistance index(RI), peak systolic blood flow velocity(PSBV), systolic acceleration time(SAT), and shear wave elastography elastic index(SWEEI), such as average value(Emean), maximum value(Emax), and minimum value(Emix) were candidate predictors. Among the five supervised learning models, RFC had the strongest prediction efficiency with area under curve(AUC) of 0.889 (95% CI: 0.838–0.940) and 0.878 (95% CI: 0.821–0.935) in the training set and testing set, respectively. While ANN, DT, SVM and XGboost had prediction efficiency between 0.767 (95% CI: 0.716–0.818) and 0.854 (95% CI: 0.803–0.905) in the training set, and ranged from 0.762 (95% CI: 0.705–0.819) to 0.861 (95% CI: 0.804–0.918) in the testing set.

Conclusion

We have successfully constructed an ML-based prediction model, which can accurately classify the LNM risk of patients with PTMC. In particular, the RFC model can help tailor clinical decisions of treatment and surveillance.