DEGAST3D: Learning Deformable 3D Graph Similarity to Track Plant Cells in Unregistered Time Lapse Images.

Area of Science:

• Plant biology
• Cellular imaging
• Computational biology

Background:

• Accurate tracking of plant cells in 3D microscopy is challenging due to dense packing, varied growth, and image noise.
• Existing methods struggle with complex cellular structures and imaging artifacts in deep tissue layers.

Purpose of the Study:

• To introduce DEGAST3D, a novel learning-based method for precise 3D plant cell tracking in time-lapse images.
• To improve cell division detection and 3D image registration algorithms for plant tissues.

Main Methods:

• Developed DEGAST3D, a method that leverages 3D graph similarity for tracking plant cells.
• Proposed a new algorithm for detecting cell divisions within 3D plant tissues.
• Implemented an effective 3D registration technique to handle unregistered time-lapse images.

Main Results:

• DEGAST3D's cell pair matching improved precision by 6.83%, recall by 5.96%, and F1-score by 6.40% over baselines.
• The novel cell division detection improved recall by 15.38% and F1-score by 14.78% on a public dataset.
• The method effectively handles challenges like dense cell packing and image noise in 3D plant tissues.

Conclusions:

• DEGAST3D offers a robust solution for 3D plant cell tracking, improving accuracy and efficiency.
• The advancements in cell division detection and registration contribute significantly to plant science research.
• This work provides a valuable tool for analyzing plant development and cellular dynamics from microscopy data.