Related Experiment Video
Updated: Sep 17, 2025
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
9.0K
Cross-domain subcortical brain structure segmentation algorithm based on low-rank adaptation fine-tuning SAM
Yuan Sui1, Qian Hu2, Yujie Zhang3
1School of Computer Science and Engineering, Northeastern University, Shenyang, Liaoning, 110169, China.
BMC Medical Imaging
|July 2, 2025
Summary
This study introduces a new method to improve brain MRI segmentation using Low-Rank Adaptation (LoRA) to fine-tune the Segment Anything Model (SAM). This approach enhances accuracy for deep brain structures, reducing annotation costs for clinicians.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuroscience
Background:
- Accurate brain MRI segmentation is vital for diagnosing and treating neurological conditions.
- Deep learning models struggle with subcortical brain structure segmentation due to domain differences.
Purpose of the Study:
- To develop an efficient and accurate subcortical brain structure segmentation algorithm for MRI.
- To adapt large-scale foundation models for specialized medical imaging tasks.
Main Methods:
- Fine-tuning the Segment Anything Model (SAM) using Low-Rank Adaptation (LoRA).
- Freezing the SAM image encoder and applying LoRA to its weights.
- Fine-tuning SAM's prompt encoder and mask decoder with adaptive prompt learning.
Main Results:
- The fine-tuned model uses only 6.39% of the original SAM's parameters.
- Adaptive prompt learning enhances segmentation accuracy for arbitrary brain MRI scans.
- The method demonstrates generalization across diverse MRI datasets and segmentation scenarios.
Conclusions:
- This interactive approach offers intelligent segmentation for deep brain structures, overcoming data limitations.
- The algorithm effectively reduces manual annotation costs in medical image segmentation.
- The proposed method shows superior generalization and effectiveness compared to existing algorithms.
