Fourier Channel Attention Powered Lightweight Network for Image Segmentation.

Area of Science:

• Medical Image Analysis
• Computer Vision
• Biomedical Engineering

Background:

• Accurate image segmentation is crucial for quantitative analysis in biomedical research.
• Existing segmentation methods often require significant computational resources and parameters.
• Fourier Channel Attention (FCA) has shown promise in image super-resolution but its application in semantic segmentation is less explored.

Purpose of the Study:

• To develop a lightweight and accurate network for biomedical image segmentation.
• To investigate the efficacy of combining Fourier Channel Attention (FCA) with the U-Net architecture.
• To improve the segmentation of pathological images, specifically nuclei and glands.

Main Methods:

• Introduced FRUNet, a lightweight network based on U-Net, incorporating Fourier Channel Attention (FCA) blocks and residual units.
• FCA Block adaptively weights frequency information, focusing on high-frequency details in biomedical images.
• The U-Net's skip connections facilitate the fusion of encoder and decoder information.

Main Results:

• FRUNet demonstrated superior performance compared to existing advanced medical image segmentation methods on three public datasets.
• The proposed method achieved higher accuracy while utilizing fewer network parameters.
• FRUNet particularly excelled in the semantic segmentation of pathological sections, including nuclei and glands.

Conclusions:

• FRUNet offers an effective and efficient solution for accurate biomedical image segmentation.
• The integration of FCA with U-Net significantly enhances segmentation accuracy, especially for complex pathological structures.
• This lightweight architecture presents a promising advancement for quantitative analysis in medical imaging.