Related Experiment Video
Updated: Mar 19, 2026
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Classifier Model Based on Machine Learning Algorithms: Application to Differential Diagnosis of Suspicious Thyroid
Hongxun Wu1, Zhaohong Deng2, Bingjie Zhang1
11 Department of Ultrasound, Jiangyuan Hospital Affiliated to Jiangsu Institute of Nuclear Medicine (Key Laboratory of Nuclear Medicine, Ministry of Health/Jiangsu Key Laboratory of Molecular Nuclear Medicine), 20 Qianrong Rd, Wuxi, Jiangsu 214063, China.
Machine learning models were developed to detect malignant thyroid nodules. An experienced radiologist achieved higher accuracy than machine learning, though the radial basis function neural network showed the best performance among algorithms.
Area of Science:
- Medical imaging and diagnostics
- Machine learning in healthcare
- Oncology and endocrinology
Background:
- Thyroid nodules are common, and differentiating benign from malignant ones is crucial for appropriate patient management.
- Accurate diagnosis relies on imaging interpretation and histopathology, but can be challenging.
Purpose of the Study:
- To construct and evaluate machine learning classifier models for distinguishing malignant from benign thyroid nodules.
- To compare the diagnostic performance of these models against human radiologists.
Main Methods:
- Utilized ultrasound images from 970 patients with histopathologically confirmed thyroid nodules.
- Developed machine learning models using statistically significant variables identified by an experienced radiologist.
- Compared model performance with radiologists' diagnoses using Receiver Operating Characteristic (ROC) curve analysis.
Main Results:
- The experienced radiologist achieved the highest predictive accuracy (88.66%) and AUC (0.9135).
- The radial basis function (RBF) neural network (NN) demonstrated the highest sensitivity (92.31%) among machine learning models.
- Machine learning algorithms generally underperformed compared to the experienced radiologist's readings.
Conclusions:
- Machine learning algorithms showed potential but did not surpass the diagnostic performance of an experienced radiologist in this study.
- The RBF-NN model exhibited superior performance compared to other tested machine learning algorithms for thyroid nodule malignancy prediction.
