CelloType: a unified model for segmentation and classification of tissue images

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Accurate cell segmentation and classification are essential for interpreting spatial omics data.
• Traditional methods often use a two-stage approach, which can be inefficient and less accurate.

Purpose of the Study:

• To introduce CelloType, an integrated end-to-end model for simultaneous cell segmentation and classification in image-based spatial omics data.
• To improve the accuracy and efficiency of spatial omics data analysis.

Main Methods:

• CelloType employs a multitask learning strategy, integrating segmentation and classification within a single framework.
• The model utilizes transformer-based deep learning for enhanced object detection, segmentation, and classification.
• It was evaluated on multiplexed fluorescence and spatial transcriptomic images.

Main Results:

• CelloType outperforms existing segmentation methods on diverse spatial omics image types.
• It surpasses state-of-the-art methods in cell type classification accuracy.
• Demonstrated utility for multiscale segmentation and classification of cellular and noncellular elements in tissue images.

Conclusions:

• CelloType offers enhanced accuracy and multitask learning capabilities for spatial omics data analysis.
• The model facilitates automated annotation of large-scale spatial omics datasets.
• CelloType represents a significant advancement in computational tools for spatial biology research.