Related Experiment Video
Updated: May 11, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.8K
A segmentation-based algorithm for classification of benign and malignancy Thyroid nodules with multi-feature
Zhiqiang Zheng1, Enhe Liang1, Yujie Zhang1
1School of Electronic Information Engineering, Inner Mongolia University, 235 Daxue West Road, Saihan District, Hohhot, 010021 Inner Mongolia China.
Biomedical Engineering Letters
|July 1, 2024
Summary
This study introduces a novel "segmentation + classification" AI model to enhance thyroid nodule ultrasonography screening. The advanced model improves diagnostic accuracy for classifying thyroid nodules, aiding physicians in distinguishing benign from malignant cases.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Thyroid nodule ultrasonography is crucial for routine screening.
- Accurate classification of thyroid nodules (benign vs. malignant) remains a clinical challenge.
- Existing AI models require improvement for reliable diagnostic assistance.
Purpose of the Study:
- To develop and validate a new diagnostic model for thyroid nodule ultrasonography.
- To improve the accuracy and consistency of thyroid nodule classification using AI.
- To integrate domain knowledge into an AI framework for enhanced medical diagnosis.
Main Methods:
- Proposed a Multi-scale segmentation network incorporating an Attention Gate and Atrous Spatial Pyramid Pooling (ASPP).
- Developed a three-branch classification network utilizing nodule image, regional image, and edge image features.
- Employed Coordinate attention (CA) mechanism and cross-level feature fusion for improved classification accuracy.
Main Results:
- The Multi-scale segmentation network achieved high performance: 94.27% mPA, 93.90% Dice, and 88.85% MIoU.
- The classification network reached 86.07% accuracy, 81.34% specificity, and 90.19% sensitivity.
- The proposed method outperformed several classical and recent AI models in comparative tests.
Conclusions:
- The 'segmentation + classification' model offers a promising auxiliary diagnostic tool for thyroid nodules.
- The model provides objective quantitative indicators, reducing subjective judgment bias in diagnosis.
- This AI approach enhances diagnostic consistency and accuracy, aiding physicians in assessing nodule nature.
