SwinCell: a 3D transformer and flow-based framework for improved cell segmentation.

Area of Science:

• Computational Biology
• Biomedical Imaging
• Cell Biology

Background:

• Accurate three-dimensional (3D) cellular image segmentation is crucial for understanding cell structure and function.
• Challenges in 3D segmentation include complex contextual information, anisotropic image properties, and sensitivity to internal cellular structures, often leading to segmentation errors, especially in dense tissues.

Purpose of the Study:

• To introduce SwinCell, a novel 3D transformer-based framework designed to improve the accuracy of 3D cellular image segmentation.
• To address the challenges of segmenting dense cells and tissues by leveraging contextual information and local feature recognition.

Main Methods:

• Development of SwinCell, a framework utilizing the Swin-transformer architecture for 3D cellular segmentation.
• The framework predicts cell flow and differentiates individual cell instances within 3D images.

Main Results:

• SwinCell demonstrated utility in segmenting nuclei, colon tissue cells, and densely cultured cells.
• The framework effectively balances detailed local feature recognition with understanding of broader contextual information.
• Extensive testing on public and in-house 3D cell imaging datasets confirmed SwinCell's effectiveness in segmenting dense cell populations.

Conclusions:

• SwinCell is a valuable tool for 3D segmentation in cellular analysis, offering improved accuracy for dense cell populations.
• The framework has the potential to expedite research in cell biology and tissue engineering by providing reliable segmentation of complex 3D cellular structures.