Omnipose: a high-precision morphology-independent solution for bacterial cell segmentation

Area of Science:

• Microscopy and image analysis
• Computational biology
• Bacteriology

Background:

• Accurate single-cell segmentation is crucial for quantitative microscopy.
• Current segmentation algorithms struggle with diverse bacterial morphologies and imaging conditions.
• Limitations in segmentation hinder precise measurement of cellular phenomena.

Purpose of the Study:

• Introduce Omnipose, a novel deep neural network for robust cell image segmentation.
• Address limitations of existing algorithms, including Cellpose, in segmenting challenging cell types.
• Demonstrate Omnipose's utility across various biological samples and imaging modalities.

Main Methods:

• Developed Omnipose, a deep neural network utilizing unique outputs like the gradient of the distance field.
• Tested Omnipose on mixed bacterial cultures, antibiotic-treated cells, and cells with varied morphologies.
• Evaluated performance on non-bacterial subjects, diverse imaging techniques, and 3D objects.

Main Results:

• Omnipose achieved unprecedented segmentation accuracy on diverse bacterial cell types.
• Successfully segmented cells with elongated, branched, or complex morphologies.
• Demonstrated broad applicability to non-bacterial samples, varied imaging, and 3D data.

Conclusions:

• Omnipose offers a powerful and versatile solution for accurate cell image segmentation.
• Enables precise characterization of extreme morphological phenotypes, such as those in interbacterial antagonism.
• Represents a significant advancement for quantitative analysis of cellular imaging data.