CP-Net: Instance-aware part segmentation network for biological cell parsing.

Area of Science:

• Medical Image Analysis
• Computational Biology
• Cell Biology

Background:

• Accurate instance segmentation of biological cells is crucial for quantitative analysis of subcellular structures.
• Cellular measurements are vital for cell phenotyping and quality assessment, especially in diagnostics like sperm quality analysis.

Purpose of the Study:

• To introduce an instance-aware part segmentation network for distinguishing individual cells and segmenting their subcellular components.
• To develop a novel network, Cell Parsing Net (CP-Net), for precise instance-level cell parsing.

Main Methods:

• Proposed CP-Net utilizes an attention-based feature fusion module with instance masks as spatial cues to handle irregular cell shapes.
• A coarse-to-fine segmentation module enables hierarchical segmentation for accurate identification of tiny subcellular structures.
• A new sperm parsing dataset with 320 annotated images and five semantic subcellular part labels was created.

Main Results:

• CP-Net demonstrated superior performance in instance-level cell parsing compared to existing state-of-the-art methods.
• The attention-based fusion and coarse-to-fine modules effectively addressed contour misalignments and segmented small subcellular structures.

Conclusions:

• The developed CP-Net provides an effective solution for instance-aware subcellular part segmentation in medical imaging.
• This method holds significant potential for improving cell phenotyping and quantitative analysis in various biological and medical applications.