Related Experiment Video
Updated: Sep 17, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.0K
Ultrasound-based classification of follicular thyroid Cancer using deep convolutional neural networks with transfer
Enock Adjei Agyekum1,2, Zhang Yuzhi3, Yu Fang4,5,6
1Department of Ultrasound, Affiliated People's Hospital of Jiangsu University, Zhenjiang, 212002, China.
Scientific Reports
|July 1, 2025
Summary
Convolutional neural network (CNN) models effectively distinguish follicular thyroid carcinoma from adenoma. These AI models outperformed established ultrasound risk stratification systems, offering improved diagnostic potential for follicular neoplasms.
Area of Science:
- Artificial Intelligence in Medical Imaging
- Thyroid Nodule Diagnosis
- Machine Learning for Pathology
Background:
- Distinguishing follicular thyroid carcinoma (FTC) from follicular thyroid adenoma (FTA) is challenging, impacting patient management.
- Current ultrasound-based risk stratification systems like ACR-TIRADS and C-TIRADS have limitations in differentiating these conditions.
- Deep learning, specifically convolutional neural networks (CNNs), shows promise for enhancing diagnostic accuracy in medical imaging.
Purpose of the Study:
- To develop and validate CNN models for differentiating FTC from FTA using preoperative thyroid ultrasound images.
- To compare the diagnostic performance of the developed CNN models against established ACR-TIRADS and C-TIRADS risk stratification systems.
- To assess the clinical utility and calibration of the CNN models for guiding the management of follicular thyroid neoplasms.
Main Methods:
- Retrospective analysis of 327 patients with FTC or FTA who underwent preoperative thyroid ultrasound.
- Development and testing of five pre-trained CNN models (VGG16, ResNet101, MobileNetV2, ResNet152, ResNet50) on a training (n=263) and test (n=64) cohort.
- Performance evaluation using receiver operating characteristic (ROC) curves, decision curve analysis, and calibration curves; comparison with ACR-TIRADS and C-TIRADS.
Main Results:
- CNN models achieved areas under the ROC curve (AUC) between 0.64 and 0.77 for distinguishing FTC from FTA.
- The ResNet152 model exhibited the highest AUC of 0.77 (95% CI, 0.67-0.87).
- Developed CNN models demonstrated superior performance compared to ACR-TIRADS and C-TIRADS, with favorable clinical value and calibration.
Conclusions:
- CNN models, particularly ResNet152, show significant potential for accurately differentiating FTC from FTA.
- The developed AI models outperform current ultrasound risk stratification systems, offering a more reliable diagnostic tool.
- These CNN models can potentially improve the clinical management of patients with follicular thyroid neoplasms.
Related Concept Videos
01:25Ultrasound II: Endoscopic Ultrasound and FibroScan
209
Endoscopic Ultrasound (EUS) and FibroScan are valuable diagnostic tools in gastroenterology and hepatology, each with specific applications and techniques.
Endoscopic Ultrasound (EUS):
Endoscopic Ultrasound (EUS):
209
01:23The Thyroid Gland
4.4K
The thyroid gland is a small, butterfly-shaped gland located in the neck and covers the anterior surface of the trachea. The gland has two lateral lobes connected by a thin tissue mass called the isthmus. Internally, each lobe comprises many small spherical structures known as thyroid follicles, surrounded by a network of blood vessels.
The follicles have a central cavity lined by simple cuboidal to squamous epithelial cells called follicular cells. These cells produce the glycoprotein...
The follicles have a central cavity lined by simple cuboidal to squamous epithelial cells called follicular cells. These cells produce the glycoprotein...
4.4K