Machine learning enables precise holographic characterization of colloidal materials in real time

Area of Science:

• Optics and photonics
• Colloidal science
• Machine learning applications

Background:

• Holographic particle characterization tracks and analyzes individual colloidal particles in fluid media.
• Applications span statistical physics, biopharmaceuticals, and medical diagnostics.
• Conventional methods rely on Lorenz-Mie theory and optimization algorithms for data extraction.

Purpose of the Study:

• To develop an automated, end-to-end neural network solution for holographic particle characterization.
• To improve the speed, precision, and accuracy of analyzing colloidal particle data.
• To create a tool that can bootstrap conventional optimization algorithms for demanding applications.

Main Methods:

• Development of CATCH (Characterizing and Tracking Colloids Holographically), an end-to-end neural network.
• Utilizing in-line holographic video microscopy for data acquisition.
• Training the neural network to learn a representation of Lorenz-Mie theory.

Main Results:

• CATCH provides fast, precise, and accurate predictions for particle characterization.
• The neural network's performance is suitable for high-throughput applications.
• CATCH can reliably initialize conventional optimization algorithms.

Conclusions:

• CATCH offers a powerful, efficient solution for holographic particle characterization.
• The neural network's compact representation of light scattering theory suggests simplified future models.
• This advancement has implications for both fundamental research and applied product development.