CMTT-JTracker: a fully test-time adaptive framework serving automated cell lineage construction.

Area of Science:

• Computational Biology
• Image Analysis
• Machine Learning

Background:

• Automated cellular activity monitoring relies on accurate cell tracking.
• Existing methods often struggle to balance computational efficiency, accuracy, and generalizability across different cell datasets.
• Robust cell tracking is crucial for various biological research applications.

Purpose of the Study:

• To develop a novel, efficient, and accurate cell tracking framework.
• To improve the generalizability of cell tracking methods across diverse biological datasets.
• To enhance automated cellular activity monitoring capabilities.

Main Methods:

• Developed a central-metric fully test-time adaptive framework for cell tracking (CMTT-JTracker).
• Designed a CMTT mechanism for pre-segmentation, extracting information at multiple resolutions without retraining.
• Utilized a multi-task learning network with spatial attention for simultaneous cell detection and re-identification.

Main Results:

• CMTT-JTracker demonstrated superior biological and tracking performance compared to existing benchmarks.
• Achieved high Multiple Object Tracking Accuracy (MOTA) scores: 0.894 on Fluo-N2DH-SIM+ and 0.850 on PhC-C2DL-PSC.
• The CMTT segmentation unit alone outperformed state-of-the-art methods, especially in dense cell scenarios, with Dice coefficients ranging from 0.758 to 0.928.

Conclusions:

• CMTT-JTracker offers a significant advancement in automated cell tracking, balancing efficiency and accuracy.
• The framework shows robust generalizability across varied cell imaging datasets.
• The CMTT mechanism provides a powerful tool for cell segmentation, particularly in challenging dense cell environments.