CATCH: Characterizing and Tracking Colloids Holographically Using Deep Neural Networks.

Area of Science:

• Colloidal science
• Optical microscopy
• Computational physics

Background:

• In-line holographic microscopy offers rich data on colloidal dispersions.
• Traditional analysis involves computationally intensive steps: feature localization, property estimation, and model optimization.
• These methods can be slow and challenging for complex or crowded samples.

Purpose of the Study:

• To develop a machine learning-based approach for analyzing colloidal particle holograms.
• To create a computationally efficient and accurate method for characterizing colloidal dispersions.
• To enable real-time analysis of holographic data.

Main Methods:

• Implemented an end-to-end system using deep convolutional neural networks.
• Developed Characterizing and Tracking Colloids Holographically (CATCH) software.
• Applied the CATCH system to analyze holograms of colloidal spheres in experiments.

Main Results:

• The CATCH system achieves real-time analysis speeds.
• Machine learning approach outperforms conventional algorithms, especially for heterogeneous and crowded samples.
• Demonstrated precise 3D position, size, and refractive index determination of colloidal particles.

Conclusions:

• Machine learning provides a powerful and efficient alternative to traditional methods for holographic particle analysis.
• CATCH enables rapid and accurate characterization of colloidal dispersions.
• This technology has potential for real-time monitoring and analysis in various applications.