# Prediction model for extrathyroidal extension in thyroid papillary carcinoma based on ultrasound radiomics

## TL;DR

This study develops a preoperative model to predict extrathyroidal extension in thyroid cancer using ultrasound radiomics, showing strong clinical potential.

## Contribution

A novel XGB-based radiomics model for predicting extrathyroidal extension in papillary thyroid carcinoma using ultrasound data.

## Key findings

- The XGB model achieved an AUC of 0.841 in the test set and 0.814 in the external validation set.
- Six key radiomic features were identified as significant predictors of extrathyroidal extension.
- The model demonstrates clinical utility for guiding therapeutic strategies in papillary thyroid carcinoma.

## Abstract

This study aimed to construct preoperative prediction models for extrathyroidal extension (ETE) in papillary thyroid carcinoma (PTC) based on ultrasonic radiomics and explore their clinical application value. This retrospective study included PTC patients treated across three centers from 2015 to 2023. Data for 609 cases from two centers were utilized for model construction and divided 4:1 into a training set (n = 487; 144 with ETE and 343 without ETE) and test set (n = 122; 58 with ETE and 64 without ETE). The external validation set comprised 109 PTC patients from the third center (n = 109; 55 with ETE and 54 without ETE). Image features were extracted using Pyradiomics. Feature selection and dimensionality reduction were performed using the least absolute shrinkage and selection operator and principal component analysis to construct radiomics models. Model performance was evaluated by receiver operating characteristic (ROC) curve analysis, and clinical benefit was assessed by decision curve analysis. A total of 806 radiomics features were extracted from the training set data. After feature selection and dimensionality reduction, six significant features were included in the models, including one gray-level size zone matrix feature, one shape feature, one first-order feature, one gray-level run-length matrix feature, and two gray-level co-occurrence matrix features. The extreme gradient boosting (XGB) model showed the best performance in both the test and external validation sets, with area under the ROC curve values of 0.841 and 0.814, respectively. In conclusion, the XGB preoperative ETE prediction model for PTC based on ultrasonic radiomics offers good clinical application value for decision-making regarding therapeutic strategies.

The online version contains supplementary material available at 10.1038/s41598-025-19908-5.