Integrating miRNA profiling and machine learning for improved thyroid cancer diagnosis.

BACKGROUND: The incidence of thyroid cancer is rising significantly worldwide. It is particularly important to improve the diagnostic accuracy of thyroid cancer. This study aims to integrate miRNAs from fine needle aspiration (FNA) samples of thyroid nodules with machine learning algorithms to improve the accuracy of thyroid cancer diagnosis. METHODS: Differentially expressed miRNA profiles between thyroid cancer and benign thyroid tissues were obtained from the GEO (Gene Expression Omnibus) database (Series GSE116196). Following data preprocessing, least absolute shrinkage and selection operator (LASSO) regression was applied to identify the miRNAs that contributed most to thyroid cancer classification. Subsequently, advanced machine learning algorithms such as logistic regression (GLM), random forest (RF), support vector machine (SVM), XGBoost (XGB), and K-nearest neighbors (KNN) were employed to construct diagnostic models on the training set. Model performance was evaluated on the testing set by receiver operating characteristic (ROC) curve analysis, and the model with the highest area under the curve (AUC) was selected as the top-performing model. Finally, the diagnostic performance of the selected model was independently validated using an external dataset. RESULTS: Significantly upregulated and downregulated miRNAs in thyroid cancer were identified through DESeq2 analysis. LASSO regression was applied to select the miRNAs with the optimal discriminatory contribution to thyroid nodule characterization, yielding a panel of 13 candidate miRNAs, including hsa-miR-222 (1), hsa-miR-221 (1), and hsa-miR-424 (1). These 13 miRNAs were subsequently used to build disease prediction models using machine learning algorithms. By plotting ROC curves for these machine learning algorithm models and calculating their AUC values, the SVM demonstrated the best performance, with an AUC of 0.788 (95%CI: 0.698-0.873). External validation of the SVM model achieved an accuracy of 80.0%. CONCLUSION: This novel integration of miRNA from fine needle aspiration of thyroid nodules and machine learning algorithm holds considerable promise for the clinical translation of miRNA-based diagnostics, enhancing diagnostic precision for indeterminate thyroid nodules.