Your new experience awaits. Try the new design now and help us make it even better

ORIGINAL RESEARCH article

Front. Endocrinol.

Sec. Thyroid Endocrinology

Volume 16 - 2025 | doi: 10.3389/fendo.2025.1639017

This article is part of the Research TopicExploring the Applications of Artificial Intelligence in Disease Screening, Diagnosis, Treatment, and NursingView all 9 articles

Multimodal radiomics model with triple -timepoint contrast-enhanced ultrasound for precise diagnosis of C-TIRADS 4 thyroid nodules

Provisionally accepted
Linlin  ShaoLinlin ShaoLili  ZhangLili ZhangLifang  LiuLifang LiuFangfang  SunFangfang SunHongYu  LiHongYu LiTongfeng  LiuTongfeng LiuFeng  HuFeng HuLirong  ZhaoLirong Zhao*
  • First Affiliated Hospital of Jilin University, Changchun, China

The final, formatted version of the article will be published soon.

Objective: This study aims to construct a multimodal radiomics model based on contrast-enhanced ultrasound (CEUS) radiomic features, combined with conventional ultrasonography (US) images and clinical data, to evaluate its diagnostic efficacy in differentiating benign and malignant thyroid nodules (TNs) classified as C-TIRADS 4, and to assess the clinical application value of the model.This retrospective study enrolled 135 patients with C-TIRADS 4 thyroid nodules who underwent concurrent US and CEUS before FNA/surgery. From each case, one US image and three CEUS key frames (2s post-perfusion, peak enhancement, 2s post-peak) were selected. Patients were randomly split into training (n=108) and test (n=27) cohorts (8:2 ratio). ROIs were manually delineated (3D-Slicer), with radiomics features extracted (PyRadiomics) and selected via mRMR and LASSO. Six CEUS radiomics-based machine learning models (KNN, SVM, RF, XGBoost, LightGBM, SGD) were developed and evaluated using AUC, accuracy, sensitivity, specificity, and F1-score. The optimal classifier was used to build US-only, US+CEUS, and clinical+US+CEUS models. Statistical comparisons employed DeLong tests, calibration curves, and DCA.The CEUS radiomics model demonstrated favorable diagnostic performance in differentiating benign and malignant C-TIRADS 4 thyroid nodules, with sensitivity, specificity, and accuracy of 0.875, 0.769, and 0.833, respectively. When CEUS radiomic features were combined with US features, the diagnostic performance of the CEUS radiomics model was comparable to that of the US+CEUS radiomics model (AUC: 0.813 vs. 0.829, P=0.005). Furthermore, the multimodal radiomics model integrating clinical data (clinical+US+CEUS radiomics model) achieved significantly improved diagnostic efficacy, with an AUC of 0.967, along with accuracy, sensitivity, specificity, and F1-score values of 0.815, 0.823, 0.792, and 0.884, respectively.Our study developed a high-performance multimodal diagnostic model through the innovative integration of radiomic features from three critical CEUS timepoints combined with conventional ultrasound and clinical data, establishing a novel decision-support tool for accurate noninvasive classification of C-TIRADS 4 thyroid nodules. The model's superior diagnostic performance (AUC 0.967) demonstrates the transformative potential of multimodal integration in overcoming single-modality limitations and enhancing clinical decision-making, positioning this approach as a promising solution to mitigate unnecessary diagnostic procedures and overtreatment.

Keywords: thyroid nodules, contrast-enhanced ultrasound, radiomics features, machine learning, C-TIRADS

Received: 01 Jun 2025; Accepted: 31 Jul 2025.

Copyright: © 2025 Shao, Zhang, Liu, Sun, Li, Liu, Hu and Zhao. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

* Correspondence: Lirong Zhao, First Affiliated Hospital of Jilin University, Changchun, China

Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.