Related Experiment Video
Updated: Sep 12, 2025
10:39
A Label-Free Segmentation Approach for Intravital Imaging of Mammary Tumor Microenvironment
Published on: May 24, 2022
2.4K
Training-Free Breast Ultrasound Image Segmentation With Retrieval-Based SAM2
IEEE Transactions on Bio-Medical Engineering
|August 5, 2025
Summary
TFSeg, a novel Training-Free framework, enhances breast cancer ultrasound segmentation using Segment Anything Model 2 (SAM2). It achieves high precision, offering an efficient alternative to supervised methods for improved diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Breast cancer early detection is vital for prognosis.
- Ultrasound interpretation relies on expertise, posing challenges.
- Supervised segmentation models require extensive data and training.
Purpose of the Study:
- To introduce TFSeg, a Training-Free segmentation framework.
- To overcome limitations of supervised models in medical image segmentation.
- To improve breast cancer ultrasound analysis.
Main Methods:
- TFSeg utilizes Segment Anything Model 2 (SAM2).
- It employs image retrieval to generate mask prompts.
- The framework avoids model re-training and hyper-parameter tuning.
Main Results:
- TFSeg achieved a Dice score of 82.66% on breast ultrasound images.
- The method demonstrated superior precision (87.93%) compared to other models.
- Performance surpassed most supervised and training-free segmentation approaches.
Conclusions:
- TFSeg offers an efficient and effective solution for breast cancer ultrasound segmentation.
- The framework leverages image retrieval and sequence generation with SAM2.
- It shows potential for broader applications in medical image analysis.
