Deep learning for single-shot autofocus microscopy

Henry Pinkard, Zachary Phillips, Arman Babakhani, Daniel Fletcher and Laura Waller

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2019-161
December 1, 2019

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2019/EECS-2019-161.pdf

Maintaining an in-focus image over long time scales is an essential and non-trivial task for a vari- ety of microscopy applications. Here, we describe a fast and robust auto-focusing method that is compatible with a wide range of existing micro- scopes. It requires only the addition of one or a few off-axis illumination sources (e.g. LEDs), and can predict the focus correction from a single im- age with this illumination. We designed a neural network architecture, the fully connected Fourier neural network (FCFNN), that exploits an under- standing of the physics of the illumination in or- der to make accurate predictions with 2-3 orders of magnitude fewer learned parameters and less memory usage than existing state-of-the-art ar- chitectures, allowing it to be trained without any specialized hardware. We provide an open-source implementation of our method, in order to enable fast and inexpensive autofocus compatible with a variety of microscopes.

Advisor: Laura Waller


BibTeX citation:

@mastersthesis{Pinkard:EECS-2019-161,
    Author = {Pinkard, Henry and Phillips, Zachary and Babakhani, Arman and Fletcher, Daniel and Waller, Laura},
    Title = {Deep learning for single-shot autofocus microscopy},
    School = {EECS Department, University of California, Berkeley},
    Year = {2019},
    Month = {Dec},
    URL = {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2019/EECS-2019-161.html},
    Number = {UCB/EECS-2019-161},
    Abstract = {Maintaining an in-focus image over long time scales is an essential and non-trivial task for a vari- ety of microscopy applications. Here, we describe a fast and robust auto-focusing method that is compatible with a wide range of existing micro- scopes. It requires only the addition of one or a few off-axis illumination sources (e.g. LEDs), and can predict the focus correction from a single im- age with this illumination. We designed a neural network architecture, the fully connected Fourier neural network (FCFNN), that exploits an under- standing of the physics of the illumination in or- der to make accurate predictions with 2-3 orders of magnitude fewer learned parameters and less memory usage than existing state-of-the-art ar- chitectures, allowing it to be trained without any specialized hardware. We provide an open-source implementation of our method, in order to enable fast and inexpensive autofocus compatible with a variety of microscopes.}
}

EndNote citation:

%0 Thesis
%A Pinkard, Henry
%A Phillips, Zachary
%A Babakhani, Arman
%A Fletcher, Daniel
%A Waller, Laura
%T Deep learning for single-shot autofocus microscopy
%I EECS Department, University of California, Berkeley
%D 2019
%8 December 1
%@ UCB/EECS-2019-161
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2019/EECS-2019-161.html
%F Pinkard:EECS-2019-161