Using transcripts to refine image based cell segmentation with FastReseg.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics (ST) cell segmentation accuracy is crucial for unbiased biological interpretation.
• Existing image-based segmentation methods often contain inaccuracies that affect spatial analysis.
• Cellular proximity and overlap in 2D can lead to segmentation errors like spatial doublets.

Purpose of the Study:

• To introduce FastReseg, a novel algorithm for refining spatial transcriptomics cell segmentation.
• To enhance segmentation accuracy by integrating image and transcriptomic data.
• To address computational challenges and improve the interpretability of ST data.

Main Methods:

• Developed a novel algorithm, FastReseg, to refine existing image-based segmentations.
• Utilized transcriptomic data to correct segmentation inaccuracies without altering cell boundaries.
• Implemented a transcript scoring system based on log-likelihood ratios to identify and correct spatial doublets.
• Designed a modular workflow to handle large datasets and computational challenges.

Main Results:

• FastReseg refines cell segmentation accuracy by combining image and 3D transcriptomic data.
• The algorithm effectively identifies and corrects spatial doublets caused by cell proximity or overlap.
• Reduced circularity in boundary derivation and addressed computational demands for large datasets.
• Demonstrated improved quality and interpretability of spatial transcriptomics data.

Conclusions:

• FastReseg offers a scalable and efficient solution for improving spatial transcriptomics cell segmentation.
• The algorithm enhances biological interpretation by providing more accurate spatial data.
• FastReseg's modularity ensures compatibility with future advancements in segmentation technology.