Ultrasound-Based Radiomics and Deep Learning for Prediction of Lateral and Central Lymph Node Metastasis and BRAF Gene Mutation in Papillary Thyroid Carcinoma.
Authors
Affiliations (5)
Affiliations (5)
- Department of Radiation and Medical Oncology, Wenzhou Medical University First Affiliated Hospital, Wenzhou, China.
- Department of Thyroid Surgery, Wenzhou Medical University First Affiliated Hospital, Wenzhou, China.
- Department of Information, Wenzhou Medical University First Affiliated Hospital, Wenzhou, China.
- Alberta College, Wenzhou Medical University, Wenzhou, China.
- School of Public Health, Wenzhou Medical University, Wenzhou, China.
Abstract
IntroductionPredicting B-Raf proto-oncogene, serine/threonine kinase (BRAF) gene mutations, lateral lymph node metastasis (LLNM) and central lymph node metastasis (CLNM) is crucial for selecting preoperative surgical approaches and predicting survival outcomes in patients with papillary thyroid carcinoma (PTC). In this study, features extracted from ultrasound images were combined with clinical characteristics to construct a nomogram model for predicting BRAF gene mutations, LLNM and CLNM.MethodsIn this retrospective study, 580 patients with pathologically confirmed PTC were enrolled between January 2020 and August 2023 from two medical centers, which was divided into training set, internal validation set and external test set, respectively. Radiomics and deep learning (DL) features were extracted from ultrasound images. Radiomics and DL models were constructed based on the selected features to generate radiomics score (R Score) and DL score (D Score), respectively. Finally, nomogram was developed by integrating R Score, D Score with clinical characteristics to predict BRAF gene mutations, LLNM and CLNM.ResultsIn the study of predicting BRAF mutations, LLNM, and CLNM in PTC, the R model achieved area under curves (AUCs) of 0.70 vs. 0.65 vs. 0.70, 0.77 vs. 0.74 vs. 0.64, and 0.81 vs. 0.74 vs. 0.69 in the training, internal validation and external test sets, respectively. A nomogram achieved AUCs of 0.69 vs. 0.71, 0.75 vs. 0.67 and 0.78 vs. 0.78 in the prediction of BRAF mutations, LLNM, and CLNM for PTC in the internal validation and external test sets, respectively.ConclusionsThis study demonstrates that ultrasound-based model integrating radiomics features, DL features, and clinical factors can effectively predict BRAF mutations, LLNM, and CLNM in patients with PTC. The proposed nomogram models exhibited superior predictive performance compared with radiomics-only and DL-only models.