Machine Learning to Improve
Microscope Images
Ph.D. Student Sean Turner

|
Expensive microscope objectives may be required in some microscopy
imaging when cheaper objectives introduce optical aberrations that
degrade the acquired image to an unacceptable level.
|
We are trying to
create unaberrated images from images acquired with optical
aberrations, starting with defocus and spherical aberration. We are
using Fourier Optics to simulate the creation of the unaberrated and
aberrated images.
We are currently using a phase diverse aberrated image set, which is
generated by intentionally introducing known aberrations, as input to
a variety of machine learning models to try and learn to do aberration
correction. We also want to introduce domain knowledge about the
optical imaging process and compare against naive machine learning
models.
|
|
Structured Illumination to Measure
Monomer Orientation
Ph.D. Student Mahsa Azizi  |
A fluorescent tag attached to a collagen monomer enables
measurement of its location. Two tags could enable
measurement of orientation.
|
However tags bind to
locations other than the ends so their spacing is
below the resolution of an optical microscope.
Structured illumination microscopy (SIM) utilizes multiple
sinusoidal fringe patterns to produce a computed complex image
where the phase is related to the tag location. Two orthogonal
patterns could determine location of a tag in two dimensions.
Surprisingly images of phase from two tags with only one
direction of the pattern produce a symmetry pattern that exhibits
a symmetry that is coupled to the direction of a line between the
tags, even when the spacing of the tags is below the resolution
limit of the microscope. Fourier and Radon transforms can be
used to automate the process of determining monomer orientation
in many cases for very small spacing of the tags.
|
|
Confocal Interferometry to Measure
Fibril Diameter
Ph.D. Student Eric Hall  |
Collagen monomers combine to form fibrils that are typically
a few hundred nanometers in diameter. Measuring the diameter of
these fibrils is important in the study of collagen growth and
remodeling, but they are below the resolution of optical
microscopes. Interference between scattering from the front and
back surfaces will vary with diameter and wavelength.
|
The behavior is complicated by the effects of diffraction at these
small dimensions, by the Gaussian illumination of the confocal
microscope, and by the possibility of a fibril resting on the
coverslip. Finite-Difference Time-Domain calculations can predict
the confocal backscatter accurately. With multiple wavelengths the
diameter can be inferred from confocal backscatter at multiple
wavelengths.
|
|
Dynamic Light Scattering to Measure
Rotational Dynamics
MS Student Ava Giorgianni
 |
Measuring rotation of objects is important in many
fields such as Polarization fluorescence microscopy, biomotors,
flow measurement, and bio-assembly. The scattering of light from
most non-spherical objects will vary with their orientation, and
will be periodic in time with period related to the rotation
speed and the structure of the object.
|
Temporal correlation or
spectral analysis can be used to understand the rotational
dynamics.
For simple objects with a narrow range of rotation frequencies,
spectral analysis is straightforward. In contrast when the
rotation is driven by diffusion, analysis becomes more
complicated. Further complications include interaction between
translation and rotation such as in non-uniform flows, and the
complicated motion of non-rigid bodies.
|
|