Three Dimensional Holographic Microscopy Using Convolution Neural Networks

Peng, Lindsey  1  ;   Razvi, Erum  2  ;   Christopher Yip  1, 2, 3  ;   Howell, Lynn  2  

1.   Department of Chemistry and Chemical Engineering, University of Toronto;    2.   Department of Biochemistry, University of Toronto;    3.   Department of Biomedical Engineering, University of Toronto  

Once established in the patient with cystic fibrosis , P. aeruginosa can be especially difficult to treat as most drugs cannot penetrate the biofilm the bacteria grows in. There are interest in studying the development of biofilm under various conditions to understand the roles of various components in sustaining its structure. Digital holographic microscopic (DHM) has been previously applied to study swimming behavior of Pseudomonas aeruginosa to analyze trajectories of the bacterium. In this thesis, efforts are underway to use DHM analyze the bacteria interactions over a longer period of time in studying formation of biofilm. One particular challenge in analyzing the the dynamic is the amount of computational time required for numerical reconstruction. I propose a alternative method using convolution neural network that will classify diffraction pattern into corresponding z position. Ultimately, real time imaging using DHM can be achieved in one forward end-to-end deep learning algorithm.