Machine learning-based model for predicting contralateral central lymph node metastasis in papillary thyroid...

Area of Science:

• Oncology
• Surgical Oncology
• Medical Informatics

Background:

• Papillary thyroid carcinoma (PTC) originating from the thyroid isthmus has a high risk of metastasis to contralateral central lymph nodes (Cont-CLNs).
• Accurate risk assessment is crucial for personalized treatment strategies in PTC patients.

Purpose of the Study:

• To develop and validate an individualized predictive model for Cont-CLNs metastasis in PTC with isthmus proximity.
• To utilize machine learning algorithms for enhanced prediction accuracy.

Main Methods:

• Retrospective analysis of 1,672 PTC patients, with a focus on 397 patients with isthmus proximity.
• Feature selection using Boruta algorithm and LASSO regression.
• Development and evaluation of seven machine learning models, including random forest, with SHAP for interpretability.

Main Results:

• The incidence of Cont-CLNs metastasis was significantly higher in PTC with isthmus proximity (33%) compared to non-isthmic PTC (12%).
• Key predictors identified include preoperative CT and ultrasound assessments, extrathyroidal extension, ipsilateral CLNs metastasis, and tumor size.
• The random forest model achieved high performance (AUC 0.942 training, 0.861 validation), with preoperative CT assessment being the most influential predictor.

Conclusions:

• A machine learning-based model effectively predicts Cont-CLNs metastasis risk in PTC with isthmus proximity.
• This model serves as a valuable tool for personalized surgical decision-making, optimizing lymph node dissection extent.