Related Experiment Video
Updated: Jun 25, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
AGSAM: Agent-Guided Segment Anything Model for Automatic Segmentation in Few-Shot Scenarios
Hao Zhou1, Yao He1, Xiaoxiao Cui2
1State Key Laboratory of Ophthalmology, Guangdong Provincial Key Laboratory of Ophthalmology and Visual Science, Zhongshan Ophthalmic Center, Sun Yat-Sen University, Guangzhou 510000, China.
Bioengineering (Basel, Switzerland)
|May 25, 2024
Summary
Agent-Guided SAM (AGSAM) automates medical image segmentation by generating prompts for the Segment Anything Model (SAM), overcoming limited annotated data challenges. This approach enhances diagnostic accuracy in clinical settings.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Pixel-level medical image segmentation is vital for disease diagnosis and monitoring.
- High-quality annotated data is scarce due to resource-intensive manual annotation, leading to few-shot learning challenges in clinical applications.
Purpose of the Study:
- To introduce Agent-Guided SAM (AGSAM), an automated method for medical image segmentation.
- To address the limitations of manual prompt engineering in few-shot medical image segmentation scenarios.
Main Methods:
- AGSAM leverages the Segment Anything Model (SAM) and its SAM-Med2D variant for automated prompt generation.
- A novel feature augmentation convolution module (FACM) is proposed to improve feature representation stability and model accuracy.
Main Results:
- AGSAM demonstrated consistent superiority over existing methods across various segmentation metrics.
- The method effectively overcomes the challenge of limited annotated data in medical image segmentation tasks.
Conclusions:
- AGSAM provides an effective solution for automated, high-quality medical image segmentation, particularly in data-scarce clinical environments.
- The automated prompt generation and feature augmentation contribute to improved accuracy and adaptability.
