Presenter: Tyson Hoyt
Authors: Tyson Hoyt, Dr. Vern Hart
Faculty Advisor: Vern Hart
Institution: Utah Valley University
Our group has successfully developed an algorithm capable of auto-segmenting optical microscopy images of adherent mammalian cells, based on the diffraction patterns from a laser diode. Our recent attempt at reconstructing both phase and amplitude variations in a sample, using holographic techniques, requires the preparation of imaging specimens with controlled densities and distributions. The training of an artificially intelligent neural network is improved as these concentrations are controlled more precisely (e.g., 99.5% cell type A and 0.5% cell type B). As a first attempt at training our neural network, a line of COS-7 cells (African Green Monkey kidney fibroblasts) were cultured with polystyrene microspheres, used to simulate trace amounts of a foreign cell for algorithm calibration. Each cancer line exhibits a unique diffraction pattern due to external morphology and intracytoplasmic features. The polystyrene microspheres are comparable in size (10–20 µm), shape (approximately round), and index of refraction (1.33) to the cells but lack intracellular morphology. This allows the network to learn specific distinguishing features for increased classification accuracy. Cells of this type are not typically cultured in combination with inorganic materials. As such, unique growth protocols will be presented, including sterilization and the removal of microspheres during media replacement and passaging, in addition to formulations of sphere concentration as a function of cell density.