# Machine learning-based model for predicting contralateral central lymph node metastasis in papillary thyroid carcinoma with isthmus proximity

## TL;DR

This study develops a machine learning model to predict lymph node metastasis risk in thyroid cancer patients with tumors near the isthmus, aiding surgical decisions.

## Contribution

A novel machine learning model is developed and validated for predicting contralateral lymph node metastasis in isthmus-proximal papillary thyroid carcinoma.

## Key findings

- Patients with isthmus-proximal PTC had a significantly higher rate of contralateral lymph node metastasis (33%) compared to non-isthmic PTC (12%).
- The random forest model achieved an AUC of 0.861 in validation, with preoperative CT assessment identified as the most influential predictor.

## Abstract

Papillary thyroid carcinoma (PTC) originating from the isthmus exhibits a marked tendency for contralateral central lymph nodes (Cont-CLNs) metastasis. To accurately assess this risk, this study aims to establish and validate an individualized predictive model for contralateral central zone lymph node metastasis in PTC with isthmus proximity using machine learning algorithms.

This retrospective study analyzed 1,672 patients with PTC. Based on tumor location, patients were categorized into a group with PTC with isthmus proximity and a non-isthmic group to compare the incidence of Cont-CLNs metastasis. Subsequently, we focused on 397 patients with PTC with isthmus proximity, who were randomly allocated in a 7:3 ratio to a training set (n=279) and a validation set (n=118). Feature selection was performed using the Boruta algorithm and LASSO regression. Seven machine learning algorithms were then employed to construct prediction models. Model performance was evaluated using metrics including the AUC, sensitivity, and specificity. The optimal model was interpreted using the shapley additive explanations (SHAP) method.

This study included 1,672 patients with PTC. The rate of Cont-CLNs metastasis was significantly higher in patients with unilateral PTC with isthmus proximity (n=397) than in those with non-isthmic PTC (33% vs. 12%, P < 0.05). Feature selection using LASSO regression and the Boruta algorithm identified five key predictors: preoperative CT assessment, extrathyroidal extension, ipsilateral central lymph node (Ipsi-CLNs) metastasis, preoperative ultrasound assessment, and tumor size. Among the seven machine learning algorithms evaluated, the random forest model demonstrated the best overall performance, achieving the highest F1 score and AUC values of 0.942 in the training set and 0.861 in the validation set. SHAP interpretability analysis confirmed that preoperative CT assessment was the most influential predictor, and its impact pattern was highly consistent with established clinical knowledge.

The machine learning model developed in this study effectively predicts the risk of Cont-CLNs metastasis in patients with unilateral PTC with isthmus proximity, providing a valuable tool to support personalized surgical decision-making regarding the extent of lymph node dissection.