MSCA-Net: Multi-Modal Cell Segmentation for Spatial Transcriptomics

Area of Science:

• Computational Biology
• Genomics
• Biotechnology

Background:

• Single-cell spatial transcriptomics offers unprecedented cellular resolution but faces computational challenges in cell segmentation.
• Current methods often use single data modalities, neglecting complementary information for accurate boundary extraction.

Purpose of the Study:

• To develop an advanced computational framework for accurate cell boundary segmentation in high-resolution spatial transcriptomics.
• To integrate multi-modal data, including H&E staining images and transcriptomic features, for improved segmentation performance.

Main Methods:

• Proposed MSCA-Net (Multi-Scale Convolutional Attention U-Net), a deep learning framework.
• Integrated H&E staining images with selected transcriptomic features for cell segmentation.
• Evaluated MSCA-Net on dorsal root ganglia (DRG) neuron datasets.

Main Results:

• MSCA-Net achieved accurate cell boundary extraction by effectively integrating multi-modal data.
• The method demonstrated superior performance compared to existing state-of-the-art segmentation techniques.
• Reconstructed spatial transcriptomic slices enabled downstream analyses consistent with prior biological knowledge.

Conclusions:

• MSCA-Net provides a robust solution for cell segmentation in high-resolution spatial transcriptomics.
• The framework enhances the reliability of spatial transcriptomic data analysis for biological discovery.
• Integrating multi-modal data is crucial for advancing computational challenges in transcriptomics.