A robust transformer-based pipeline of 3D cell alignment, denoise and instance segmentation on electron microscopy...

Area of Science:

• Plant Biology
• Cell Biology
• Developmental Biology

Background:

• Germline cell differentiation is crucial for genetic transmission but poorly understood in plants.
• Investigating plant germline cell ultrastructure requires nanoscale 3D reconstruction of tissues.
• Electron microscopy is essential for high-resolution imaging of plant developmental processes.

Purpose of the Study:

• To present a comprehensive framework for reconstructing large-volume plant tissue from serial electron microscopy images.
• To develop and validate a deep learning-based pipeline (3DCADS) for accurate 3D cell segmentation and reconstruction.
• To analyze the morphological and topological characteristics of plant germline cells during development.

Main Methods:

• Serial electron microscopy imaging of plant tissues.
• A five-stage pipeline: image registration, denoising, semantic segmentation (Transformer network), instance segmentation (supervoxel clustering), and automated analysis.
• Deep learning model (3DCADS) for cell instance segmentation and 3D reconstruction.

Main Results:

• The 3DCADS pipeline achieved high accuracy in 3D cell reconstruction, outperforming baseline models.
• Successful reconstruction of early meiosis stages in Arabidopsis thaliana anthers.
• Generated topological connectivity networks and analyzed morphological parameters of germline cells.

Conclusions:

• The 3DCADS framework provides accurate and efficient 3D reconstruction of plant tissues for biological analysis.
• This method offers significant potential for quantitative analysis of plant cell development, including genetic variations.
• The study provides new insights into germline cell fate transition and stamen development in plants.