Enhancing yeast cell tracking with a time-symmetric deep learning approach.

Area of Science:

• Cell biology
• Bioimage analysis
• Computational biology

Background:

• Accurate live cell tracking in video microscopy is crucial but challenging for current image processing methods.
• Existing deep-learning approaches often rely on consecutive frames, limiting generalizability.
• Challenges include handling artifacts and learning complex motion patterns.

Purpose of the Study:

• To develop a novel deep-learning-based cell tracking method.
• To overcome limitations of consecutive frame-based tracking.
• To enable learning of cell motion patterns without prior assumptions.

Main Methods:

• A novel deep-learning framework for cell tracking.
• Utilizes spatio-temporal neighborhood analysis instead of consecutive frames.
• Learns cell motion patterns intrinsically.

Main Results:

• Demonstrated efficacy through biologically motivated validation.
• Outperformed multiple state-of-the-art cell tracking methods.
• Successfully applied to budding yeast recordings and simulated samples.

Conclusions:

• The proposed method offers improved accuracy and generalizability for live cell tracking.
• Spatio-temporal neighborhood analysis is a promising approach for deep-learning-based tracking.
• The method shows potential for handling challenging microscopy data with artifacts.