“Multi-view Imaging under the microscope: Polarized ray tracing and deep learning for light field microscopy”

Datum

20.02.2020

Zeit

14:00 - 15:00

Sprecher

Rudolf Oldenbourg

Zugehörigkeit

Marine Biological Laboratory, Chicago, USA

Sprache

en

Hauptthema

Biologie

Host

Ivo Sbalzarini

Beschreibung

For the past several years, we have worked on advancing polarized light microscopy from 2D to its full 3-dimensional glory. To this end, we work with multi-view imaging techniques, like the diSPIM in Abhishek Kumar’s lab at MBL and the light field microscope in my lab. We have built the hardware, collected the images, but we are still short of appropriate computational algorithms to reconstruct the 3D density and orientation of fluorophores attached to actin, microtubule, or membrane structures, whose polarized fluorescence is recorded in a stack of diSPIM images or in a single light field image. Therefore, Ben Preusser and I are delighted to be able to visit the group of Ivo Sbalzarini to explore computational approaches for reconstructing objects based on multi-view image data, specifically light field images. We are bringing with us experimental light field images of simple test samples, like giant unilamellar vesicles (GUVs) whose membrane contained fluorescent dyes. I also developed a ray tracing program that generates polarized light field images of phantom objects that resemble real-world objects like the GUVs, filament networks, etc. Using the ray tracing program, we can create a plethora of synthetic light field images that are paired with their phantom objects. We plan to explore if and how we can use the many pairs of phantom objects and their light field images to train a neural network that will then assist in the reconstruction of experimental light field images of unknown objects. During our visit to your beautiful city and world-class institute, we hope to explore the potential architecture and training procedures for a neural network that will ultimately be capable of reconstructing real-world objects, like living cells and tissues that were observed with the polarized light field microscope. At the same time we wish to learn more about adaptive particle representation, a data compression scheme developed in your lab that seems well suited for light field images, whose object resolution varies across a reconstructed volume. In this half-hour talk, I will give an overview of the light field imaging technique, its modeling using ray tracing algorithms, and some examples of light field data of real-world and phantom objects.

Letztmalig verändert: 21.02.2020, 00:10:02