Related Experiment Video
Updated: Aug 5, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Ultrasound-based thyroid nodule segmentation with deep hybrid convolutional network
Fan Lu1, Hao Sun1, Binbin Jiang2
1School of Future Science and Engineering, Soochow University, Suzhou, Jiangsu, China.
Medical Physics
|August 3, 2026
Summary
This study introduces UTNseg, a novel deep learning model for automatic thyroid nodule segmentation in ultrasound images. UTNseg significantly improves segmentation accuracy, addressing challenges like blurred boundaries and low contrast for better clinical assessment.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Thyroid nodule segmentation in ultrasound images is crucial for disease assessment.
- Challenges include blurred boundaries, echo heterogeneity, and low contrast, hindering precise identification.
Purpose of the Study:
- To develop a deep hybrid convolutional network, UTNseg, for precise automatic segmentation of ultrasound-based thyroid nodules.
- To overcome existing limitations in thyroid ultrasound image analysis.
Main Methods:
- UTNseg integrates four novel modules: spatial-channel feature strong capture (STransformer), redundant feature map utilization (RMU), composite convolution (CConv), and lightweight edge refinement (LER).
- The model was evaluated on two public datasets (TN3K and DDTI) and compared against six state-of-the-art methods using Dice Similarity Coefficient (DSC) and Hausdorff Distance (HD95).
Main Results:
- UTNseg achieved superior performance, with DSC scores of 79.52% (TN3K), 86.21% (DDTI), and 80.77% (hybrid dataset).
- HD95 scores were 16.32 (TN3K), 12.76 (DDTI), and 14.56 (hybrid dataset).
- Statistical analysis confirmed significant improvements (p < 0.05) over existing methods with medium to large effect sizes.
Conclusions:
- UTNseg demonstrates excellent and reliable performance for thyroid nodule segmentation.
- The model's improvements are statistically significant and practically important for clinical applications.
