Related Experiment Video
Updated: Aug 16, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.9K
Classification for thyroid nodule using ViT with contrastive learning in ultrasound images
Jiawei Sun1, Bobo Wu2, Tong Zhao2
1The Affiliated Changzhou NO.2 People's Hospital of Nanjing Medical University, Changzhou 213003, China; Jiangsu Province Engineering Research Center of Medical Physics, Changzhou 213003, China; Center of Medical Physics, Nanjing Medical University, Changzhou 213003, China.
Computers in Biology and Medicine
|December 24, 2022
Summary
A new AI model, TC-ViT, uses Vision-Transformer and contrast learning to accurately classify thyroid nodules (TI-RADS 3) from ultrasound images, improving diagnostic accuracy and reducing unnecessary biopsies.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Distinguishing benign thyroid nodules (especially TI-RADS 3) from malignant ones is challenging, leading to diagnostic inaccuracies.
- Inconsistent interpretations and unnecessary biopsies are common due to feature ambiguity.
Purpose of the Study:
- To develop an accurate AI model for classifying TI-RADS 3 and malignant thyroid nodules.
- To enhance diagnostic accuracy and specificity of biopsy recommendations using deep learning.
Main Methods:
- A Vision-Transformer-based (ViT) model, TC-ViT, was developed incorporating contrast learning.
- Region of Interest (ROI) from nodule images were used to enhance local features within the ViT.
- Contrast learning minimized representation distances between similar nodule categories.
Main Results:
- The TC-ViT model achieved an accuracy of 86.9% in classifying thyroid nodules.
- Evaluation metrics demonstrated superior performance compared to other deep learning networks.
- The model effectively classifies TI-RADS 3 and malignant nodules from ultrasound images.
Conclusions:
- TC-ViT offers a promising approach for automatic and accurate classification of thyroid nodules.
- This AI tool can aid in computer-aided diagnosis, improving comprehensive analysis and diagnostic precision.
- The developed model has the potential to reduce overdiagnosis and unnecessary invasive procedures.
