Biophysical simulation enables segmentation and nervous system atlas mapping for image first spatial omics.

Area of Science:

• Neuroscience
• Computational Biology
• Bioinformatics

Background:

• Spatial omics (SO) offers high-resolution molecular mapping in tissues.
• Accurate anatomical segmentation is crucial for SO analysis but challenging in complex neural tissues.
• Current methods require extensive manual annotations, limiting scalability.

Purpose of the Study:

• To develop an automated image-driven approach for spatial omics analysis in the nervous system.
• To overcome the bottleneck of manual annotation in neural tissue segmentation.
• To create transferable models for diverse species and tissue architectures.

Main Methods:

• Introduced SiDoLa-NS (Simulate, Don't Label-Nervous System), an image-driven framework.
• Generated synthetic tissue images based on biophysical properties to create training data.
• Trained supervised instance segmentation convolutional neural networks (CNNs) for nucleus and tissue structure segmentation.

Main Results:

• Achieved high precision and F1-scores (>0.95) for nucleus segmentation using synthetic data.
• Successfully identified macroscopic tissue structures in mouse brain (mAP50=0.869), spinal cord (mAP50=0.96), and pig sciatic nerve (mAP50=0.995).
• Demonstrated the potential for scalable training data generation and model transferability.

Conclusions:

• SiDoLa-NS effectively addresses the challenge of manual annotation in spatial omics of the nervous system.
• The framework enables accurate segmentation of neural tissue structures, accelerating SO applications.
• This approach facilitates the development of transferable models across different species and tissue types.