Publication
Optical microscopy is an essential tool for biologists, who are often faced with the need to overcome the spatial and temporal resolution limitations of their devices to capture finer details. As upgrading imaging hardware is expensive, computational methods offer cost-efficient ways to enhance the capabilities of the devices by combining acquisition techniques and post-processing. Despite the freedom offered by mechanized stages and variable intensity lights in modern microscopes, existing custom control solutions do not offer sufficient flexibility to explore various subframe illumination and motion patterns. Moreover, the high computational cost of 3D simulations makes it difficult to characterize imaging setups at scale.
The contributions in this thesis address the above issue by introducing illumination control hardware that allows exploiting the optomechanical components commonly found in most imaging platforms to increase their spatial and temporal resolutions through computational imaging approaches. A complete simulation environment accelerated by parallel computing complements the hardware setup. I illustrate the strengths of this framework by imaging biological samples in three applications: focal sheet scanning optical projection tomography, multichannel low-speed cardiac imaging, and optomechanical modulation tomography.
As my first contribution, I extend the OpenSPIM microscope, an open light sheet device, to enable sub-frame control of illumination modulation in conjunction with the camera and stage. This allows implementing complex acquisition procedures that require a precisely timed control of the microscope's optomechanical components. Then, I show that illumination shaping using focal plane scanning increases the resolution of optical projection tomography images. This result is supported by my simulation framework, which allows testing a range of optical settings at scale thanks to parallel programming. Next, I take advantage of the subframe illumination capabilities of my system and introduce a method to reconstruct multichannel videos with virtually increased frame rate from single-channel cardiac fluorescence imaging using a low frame rate camera. The proposed method uses a paired acquisition approach that alternates between illumination modalities to allow reconstructing videos with improved temporal resolution in post-processing. Finally, I investigate a compressed sensing method that combines the spatial and temporal aspects of illumination shaping to reconstruct high-resolution volumes from few images. I formulate an efficient 1+2D regularization function that offers high reconstruction fidelity while being fast to compute using parallel computing.
In conclusion, this thesis provides hardware and software tools to implement spatiotemporal light modulation methods with good cost efficiency. It shows that the combination of long camera integration times with active illumination modulation or sample movement can be exp
Edoardo Charbon, Claudio Bruschini, Arin Can Ülkü, Yichen Feng