Related Experiment Video
Updated: Sep 24, 2025
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Thyroid nodule segmentation and classification in ultrasound images through intra- and inter-task consistent
Qingbo Kang1, Qicheng Lao2, Yiyue Li1
1West China Biomedical Big Data Center, West China Hospital, Sichuan University, Chengdu, Sichuan 610041, China; Med-X Center for Informatics, Sichuan University, Chengdu, Sichuan 610041, China; West China Hospital-SenseTime Joint Lab, Chengdu, Sichuan 610041, China.
This study introduces consistent learning to improve thyroid nodule segmentation and classification in ultrasound images. The new method effectively reduces prediction inconsistencies, enhancing diagnostic accuracy for computer-aided diagnosis systems.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer-Aided Diagnosis
Background:
- Thyroid nodule segmentation and classification are crucial for computer-aided diagnosis.
- Multi-task learning is a promising approach but suffers from prediction inconsistencies.
Purpose of the Study:
- To address intra- and inter-task inconsistencies in multi-task learning for thyroid nodule analysis.
- To improve the performance of both segmentation and classification tasks.
Main Methods:
- Proposed a novel intra- and inter-task consistent learning framework.
- Developed a multi-stage, multi-task learning network.
- Enforced consistent predictions across related tasks during training.
Main Results:
- Effectively eliminated intra-task and inter-task inconsistencies.
- Significantly improved performance in thyroid nodule segmentation.
- Enhanced accuracy in thyroid nodule classification.
Conclusions:
- The proposed consistent learning approach is effective for thyroid nodule analysis.
- This method improves the reliability and accuracy of computer-aided diagnosis systems.
- Addresses key challenges in joint segmentation and classification tasks.
