Taming large vision model for medical image segmentation via Dual Visual Prompt Tuning.

Area of Science:

• Medical Image Analysis
• Computer Vision
• Artificial Intelligence

Background:

• The Segment Anything Model (SAM) shows promise in natural image segmentation but struggles with medical imaging due to unique target characteristics, noise, and limited fine-tuning data.
• Manual prompting in SAM is inefficient for medical applications, requiring laborious adaptation.

Purpose of the Study:

• To introduce Dual Visual Prompt Tuning (DVPT), an automated strategy to enhance SAM's performance for medical image segmentation.
• To address limitations of SAM in medical imaging, including noise, artifacts, and the need for efficient prompting.

Main Methods:

• DVPT utilizes a fully automatic prompting paradigm with two key components: the Local Feature Prompt Tuning (LFPT) module for detailed anatomical structures and the Global Guiding Prompt (GGP) encoder for noise mitigation and boundary clarification.
• Both local and global prompts are integrated into the mask decoder of SAM.

Main Results:

• DVPT demonstrated superior segmentation accuracy across three distinct medical image segmentation tasks.
• The method consistently outperformed existing state-of-the-art approaches in experimental evaluations.

Conclusions:

• DVPT significantly enhances SAM's capability for medical image segmentation, offering a more accurate and efficient solution.
• This advancement contributes to improved computer-assisted diagnostics and promotes progress in healthcare technology.