Uncertainty-Driven Edge Prompt Generation Network for Medical Image Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Foundational image segmentation models like SAM excel in natural images but struggle with medical imaging's complexities.
• Current SAM-based medical segmentation methods require extensive manual input (points, boxes, masks), limiting flexibility and adaptability.
• Edge blurring in medical images presents a significant challenge for accurate segmentation.

Purpose of the Study:

• To develop an adaptable and flexible medical image segmentation method that overcomes the limitations of manual guidance.
• To address the challenge of edge blurring in medical images for improved segmentation quality.
• To introduce UDEG-Net, an uncertainty-driven edge prompt generation network for robust medical image segmentation.

Main Methods:

• Fine-tuning the SAM encoder using Low-Rank Adaptation (LoRA) to enhance feature extraction for medical images.
• Developing an automatic edge prompt generator to overcome interactive prompt limitations and improve structural representation.
• Implementing evidence-based uncertainty estimation and a progressive uncertainty-driven loss to guide prompt generation.

Main Results:

• UDESG-Net demonstrated superior performance compared to state-of-the-art methods in medical image segmentation.
• The proposed method effectively handles edge blurring and improves segmentation accuracy.
• Experimental validation was conducted on three public and one private medical image datasets.

Conclusions:

• UDESG-Net offers a significant advancement in automated medical image segmentation.
• The uncertainty-driven edge prompt generation approach enhances adaptability and robustness.
• This method holds promise for improving diagnostic accuracy and efficiency in medical imaging workflows.