Target area distillation and section attention segmentation network for accurate 3D medical image segmentation

Area of Science:

• Medical Image Analysis
• Computer Vision
• Artificial Intelligence

Background:

• 3D medical image segmentation is crucial for analysis.
• Attention mechanisms improve performance but have limitations due to small receptive fields.
• Radiologists use a multi-slice, multi-view approach for accurate segmentation.

Purpose of the Study:

• To propose a novel 3D medical image segmentation method that simulates radiologists' recognition process.
• To improve the accuracy of 3D medical image segmentation by exploiting 3D context information more effectively.
• To address the limitations of existing attention mechanisms in 3D segmentation.

Main Methods:

• Simulating radiologists' recognition process by exploiting 3D context information.
• Utilizing target region distillation to extract common segmented region information from similar anatomical structures.
• Introducing two optimizations: Target Area Distillation for initial target attention and Section Attention for large receptive field attention extraction across 2D sections.

Main Results:

• The proposed method demonstrated improved segmentation accuracy on the ImageCHD and COVID-19 datasets.
• Achieved a 2-4% improvement in Dice score compared to state-of-the-art methods.
• Code released publicly for reproducibility and further research.

Conclusions:

• The developed method effectively enhances 3D medical image segmentation by incorporating 3D context and simulating clinical practice.
• The proposed Target Area Distillation and Section Attention significantly improve segmentation performance.
• This approach offers a promising direction for more accurate and efficient medical image analysis.