Computational Hyperspectral Microscopy for Bioimaging

Neerja Aggarwal

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2025-83
May 16, 2025

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2025/EECS-2025-83.pdf

Hyperspectral imaging involves detecting the spectrum (intensity vs wavelength) of light emitted at each point in space. It has applications in biology such as fluorescence imaging of live cells and interferometry to see inside tissues. However, traditional hyperspectral systems often have to scan through this three-dimensional spatial-spectral datacube ($x,y, \lambda$) due to a 2D sensor, resulting in long acquisition times and large setups. Snapshot imaging fits the entire 3D datacube onto a 2D sensor at once but sacrifices resolution. Computational imaging involves the codesign of both optics and algorithms together to beat traditional tradeoffs. In this work, we present three imaging systems for various bioimaging applications that benefit from computational imaging to improve spectral imaging performance. In the first application, we redesigned a traditional spectrometer using a diffuser instead of a grating to diffract light. The resulting speckle pattern was captured using an image sensor and inverted to solve for the spectrum. This compact spectrometer was developed for optical coherence tomography, an interferometry technique for imaging eyes. In the second project for fluorescence microscopy, we used a diffuser to multiplex light onto a spectral filter array on an image sensor. We used compressed sensing to solve for more voxels in the hyperspectral data cube than pixels on the sensor. We developed a compact attachment for a traditional benchtop microscopy that enables live imaging on biological samples and demonstrate high fidelity reconstructions in experiment.

In the final project, we adapted a Fourier ptychography system for spectral imaging using a filter array. Fourier ptychography uses angled illumination to scan through the spatial Fourier plane and build up a higher resolution image. By placing the filter array in the Fourier plane, we can scanned the object’s spatial frequencies through each spectral filter to build up a high resolution spatio-spectral datacube. We investigated this idea via simulation and proposed an experimental setup that could be used for digital pathology.

Advisor: Laura Waller

"; ?>

BibTeX citation:

@phdthesis{Aggarwal:EECS-2025-83,
    Author = {Aggarwal, Neerja},
    Title = {Computational Hyperspectral Microscopy for Bioimaging},
    School = {EECS Department, University of California, Berkeley},
    Year = {2025},
    Month = {May},
    URL = {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2025/EECS-2025-83.html},
    Number = {UCB/EECS-2025-83},
    Abstract = {
Hyperspectral imaging involves detecting the spectrum (intensity vs wavelength) of light emitted at each point in space.  It has applications in biology such as fluorescence imaging of live cells and interferometry to see inside tissues.  However, traditional hyperspectral systems often have to scan through this three-dimensional spatial-spectral datacube ($x,y, \lambda$) due to a 2D sensor, resulting in long acquisition times and large setups.  Snapshot imaging fits the entire 3D datacube onto a 2D sensor at once but sacrifices resolution.  Computational imaging involves the codesign of both optics and algorithms together to beat traditional tradeoffs. In this work, we present three imaging systems for various bioimaging applications that benefit from computational imaging to improve spectral imaging performance. 
 
In the first application, we redesigned a traditional spectrometer using a diffuser instead of a grating to diffract light.  The resulting speckle pattern was captured using an image sensor and inverted to solve for the spectrum.  This compact spectrometer was developed for optical coherence tomography, an interferometry technique for imaging eyes.  
 
In the second project for fluorescence microscopy, we used a diffuser to multiplex light onto a spectral filter array on an image sensor.  We used compressed sensing to solve for more voxels in the hyperspectral data cube than pixels on the sensor.  We developed a compact attachment for a traditional benchtop microscopy that enables live imaging on biological samples and demonstrate high fidelity reconstructions in experiment.

In the final project, we adapted a Fourier ptychography system for spectral imaging using a filter array.  Fourier ptychography uses angled illumination to scan through the spatial Fourier plane and build up a higher resolution image.  By placing the filter array in the Fourier plane, we can scanned the object’s spatial frequencies through each spectral filter to build up a high resolution spatio-spectral datacube.  We investigated this idea via simulation and proposed an experimental setup that could be used for digital pathology.}
}

EndNote citation:

%0 Thesis
%A Aggarwal, Neerja
%T Computational Hyperspectral Microscopy for Bioimaging
%I EECS Department, University of California, Berkeley
%D 2025
%8 May 16
%@ UCB/EECS-2025-83
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2025/EECS-2025-83.html
%F Aggarwal:EECS-2025-83