Generative Models for Digital Holographic Microscopy

Abstract

In the past few years, the venerable field of holographic microscopy has been revitalized by computational data analysis. It is now possible to fit a generative (forward) model of scattering directly to experimentally obtained holograms. This approach enables precision measurements: it allows the motion of colloidal particles and biological organisms to be tracked with nanometer-scale precision and their optical properties inferred on a particle-by-particle basis. In this thesis, I discuss how the model-based inference approach to holographic microscopy is opening up new applications. I also discuss how it must evolve to meet the needs of new applications that demand lower systematic uncertainties and maximum precision. In this context, I present some new and previous results on how modeling the optical train of the microscope can enable better measurements of the positions of spherical and nonspherical colloidal particles. Finally, I discuss how machine learning might play a role in future advances. Though I do not exhaustively catalogue all the developments in this field, I hope that by presenting a few examples and some new results I can spotlight open questions and opportunities.