Feasibility analysis of radiomics and deep learning models in predicting the efficacy of 131I therapy for papillary thyroid cancer
10.3760/cma.j.cn321828-20240904-00312
- VernacularTitle:影像组学与深度学习模型预测甲状腺乳头状癌 131I治疗疗效的可行性分析
- Author:
Lele ZHANG
1
;
Lu LU
;
Zhao GE
;
Ning LI
;
Jinquan HUANG
;
Xingyu MU
;
Wei FU
Author Information
1. 桂林医学院附属医院核医学科，桂林 541000
- Publication Type:Journal Article
- Keywords:
Thyroid neoplasms;
Radiotherapy;
Iodine radioisotopes;
Thyroglobulin;
Radiomics;
Deep learning;
Forecasting
- From:
Chinese Journal of Nuclear Medicine and Molecular Imaging
2025;45(9):543-548
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To explore the application value of radiomics, deep learning, and their combined models in predicting the efficacy of radioiodine adjuvant therapy in patients with papillary thyroid cancer (PTC).Methods:A retrospective analysis was conducted on the clinical and imaging data of 131 PTC patients (38 males, 93 females; age 41(33, 48) years) who received first 131I treatment at the Affiliated Hospital of Guilin Medical University from January 2018 to March 2023. Patients were randomly divided into a training set ( n=105) and a test set ( n=26) at the ratio of 8∶2. Multivariate logistic regression analysis was used to screen clinical features to determine independent predictors affecting the efficacy of 131I therapy. Radiomics and deep learning features were extracted from the enhanced CT scans and were combined by using the extremely randomized trees (ExtraTrees) algorithm to construct radiomics, deep learning, and combined models. The predictive abilities of the models were evaluated by AUC, and the Delong test was applied to compare the difference between AUCs. Results:Higher pre-ablation stimulated thyroglobulin (ps-Tg) levels (odds ratio( OR)=1.060, 95% CI: 1.025-1.095, P=0.004) and bilateral lesions ( OR=5.085, 95% CI: 1.452-17.814, P=0.033) were independent predictors of the efficacy of 131I therapy in intermediate to high-risk PTC patients. In the training set, the radiomics model (AUC=0.853) and combined model (AUC=0.880) significantly outperformed the deep learning model (AUC=0.711; Z values: 2.48, 3.09, P values: 0.013, 0.002), while there was no statistically significant difference between the radiomics and combined models ( Z=0.51, P=0.610). In the test set, AUCs of the radiomics, deep learning, and combined models were 0.746, 0.624, and 0.876, respectively, and the AUC of the combined model was higher than that of the radiomics model or deep learning model ( Z values: 2.05, 1.99, P values: 0.040, 0.047). Conclusion:The combined model demonstrates superior performance over the standalone radiomics model and deep learning model in predicting the efficacy of 131I treatment in PTC patients.
