Related Experiment Video
Updated: Jul 25, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.2K
MFA-Net: multi-scale feature aggregation network with background-aware module for ultrasound segmentation of thyroid
Dongfen Ye1, Kun Lan2, Jianzhen Cheng3
1College of Electrical and Information Engineering, Quzhou University, Quzhou, China.
Quantitative Imaging in Medicine and Surgery
|December 10, 2025
Summary
This study introduces MFA-Net, a novel deep learning model for segmenting thyroid nodules in ultrasound images. MFA-Net effectively improves segmentation accuracy by capturing multi-scale features and reducing background noise.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Biomedical Engineering
Background:
- Thyroid nodule segmentation in ultrasound images is crucial for distinguishing benign from malignant cases.
- Challenges include weak edges, low contrast, and complex internal structures, hindering precise segmentation.
- Accurate segmentation is essential for reliable clinical diagnosis.
Purpose of the Study:
- To develop a robust multi-scale feature aggregation network (MFA-Net) for accurate thyroid nodule segmentation.
- To enhance segmentation performance by incorporating a background-aware module (BAM).
- To improve fine-grained details and global structure representation in ultrasound images.
Main Methods:
- Developed MFA-Net with a multi-scale feature aggregation module (MFAM) to capture multi-scale context.
- Integrated a background-aware module (BAM) to suppress background noise and differentiate nodules from surrounding tissues.
- Employed spatial and channel attentions within the residual decoder module (RDM) for refined segmentation.
Main Results:
- MFA-Net achieved high performance across multiple datasets, including TN3K (Dice: 0.8616), TG3K (Dice: 0.9857), DDTI (Dice: 0.7483), and BrainTumor (Dice: 0.8485).
- Quantitative metrics such as Intersection over Union (IoU), accuracy, and Matthews correlation coefficient (Mcc) also demonstrated strong results.
- Performance surpassed current leading models on these datasets.
Conclusions:
- MFA-Net demonstrates significant improvements in thyroid nodule segmentation accuracy and robustness.
- The MFAM, BAM, and RDM components proved effective and adaptable in various segmentation tasks.
- The findings confirm MFA-Net's potential for reliable clinical application in thyroid nodule diagnosis.
