Fast detection of slender bodies in high density microscopy data.

Area of Science:

• * Biology
• * Computer Science
• * Microscopy

Background:

• * Deep learning significantly enhances biological microscopy data analysis.
• * Tracking overlapping, slender organisms (e.g., nematodes, spermatozoa, flagella) in microscopy presents a significant challenge.
• * Existing methods struggle with low-resolution data and dense populations.

Purpose of the Study:

• * To develop an end-to-end deep learning approach for precise shape trajectory extraction of motile, overlapping slender bodies.
• * To address challenges in low-resolution microscopy and high-density organism tracking.
• * To demonstrate the method's generalizability and application in biological research.

Main Methods:

• * Developed an end-to-end deep learning model for trajectory extraction.
• * Employed a physics-based model for nematode motility in synthetic data generation.
• * Validated the model's performance in low-resolution settings and on dense populations of Caenorhabditis elegans.

Main Results:

• * Achieved precise shape trajectory extraction for generally motile and overlapping slender bodies.
• * Demonstrated fast detection and the ability to track thousands of organisms simultaneously.
• * Showcased successful generalization from purely synthetic data to experimental videos.

Conclusions:

• * The developed deep learning approach effectively overcomes challenges in tracking overlapping slender organisms in microscopy.
• * The method is robust in low-resolution conditions and scalable to large populations.
• * Training on synthetic data enables generalization to real-world biological experiments, offering a versatile tool for various applications.