Ensemble Deep Learning Object Detection Fusion for Cell Tracking, Mitosis, and Lineage.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Machine Learning

Background:

• Manual cell tracking is time-consuming and error-prone.
• Accurate cell tracking is crucial for understanding biological processes and disease diagnosis.

Purpose of the Study:

• To develop an automated deep learning-based method for cell detection, tracking, and motility analysis.
• To improve robustness across different cell lines and model cell lineage and proliferation.

Main Methods:

• Developed EDNet, an ensemble deep learning approach for 2D cell detection, outperforming YOLO and FasterRCNN.
• Integrated EDNet with the M2Track algorithm for multiobject tracking, mitosis detection, and cell lineage graph generation.

Main Results:

• Achieved state-of-the-art performance on the CTMCv1 dataset with MOTA score of 50.6% and TRA score of 52.5%.
• Demonstrated comparable or superior performance to human tracking in muscle stem cell motility studies.

Conclusions:

• EDNet offers a robust and efficient solution for automated cell tracking and motility analysis.
• The method has the potential to significantly advance biomedical research and medical diagnosis.