Computer Vision Methods for Spatial Transcriptomics: A Survey

Area of Science:

• Computational Biology
• Bioinformatics
• Artificial Intelligence

Background:

• Spatial transcriptomics (ST) measures gene expression and localization in tissues.
• Current bioinformatics approaches often underutilize detailed tissue morphology.
• Limitations of ST include high cost, 2D analysis, and limited clinical translation.

Purpose of the Study:

• To systematically survey computer vision AI models applied to spatial transcriptomics analytics.
• To explore how AI can bridge the gap between histology and molecular data.
• To highlight the potential of AI to overcome ST limitations and advance research and clinical practice.

Main Methods:

• Categorization of AI models based on architectures, learning paradigms, tasks, and datasets.
• Review of AI techniques for integrating histological images with gene expression data.
• Analysis of AI's role in virtual sequencing and 3D tissue reconstruction for ST.

Main Results:

• Computer vision AI offers novel approaches beyond traditional bioinformatics for ST.
• AI enables 'virtual sequencing' from histology, reducing costs and integrating pathology insights.
• AI facilitates 3D reconstruction of tissues, advancing spatial omics analysis.

Conclusions:

• AI-powered computer vision is transforming spatial transcriptomics analytics.
• Vision-driven ST accelerates discovery in basic research and clinical translation.
• This survey provides a panoramic view of AI in ST and future directions.