Scale selection and machine learning based cell segmentation and tracking in time lapse microscopy

Area of Science:

• Biomedical Engineering
• Computational Biology
• Microscopy Image Analysis

Background:

• Manual cell tracking is time-consuming and lacks reproducibility.
• Automated cell tracking faces challenges with intensity variations and resolution limits.

Purpose of the Study:

• To develop a comprehensive methodology for automated cell segmentation and tracking.
• To improve the accuracy and generalizability of cell motion analysis.

Main Methods:

• Utilized multi-scale space-time interest point detection for automatic scale selection and segmentation.
• Employed a neural network with class prototype balancing for cell region detection.
• Applied a graph-based framework for track generation and cell event detection.

Main Results:

• The proposed method demonstrated competitive performance against top techniques in the Cell Tracking Challenge (CTC).
• Achieved excellent generalization across diverse cell types, sizes, and imaging modalities.
• Validated through rigorous evaluation on time-lapse microscopy sequences.

Conclusions:

• The developed methodology offers an effective and robust solution for automated cell segmentation and tracking.
• Provides a valuable tool for disease mechanism research and treatment evaluation.
• Publicly available code facilitates wider adoption and further development.