Autofocus of phase-constrast images of yeast cells using machine learningYan, April Yujiemachine learningimage processingconvolutional neural networkbiophysicsmicroscopic autofocusingcell imagingfocus measuresPhysicsB SMicroscopic autofocusing is an essential technique for long-period image acquisition process. The hardware autofocus of the optical microscope detects small drifting distances of the sample slide. However, hardware autofocus fails especially when samples have unevenly coated cover slip, which makes the drifting distance exceed the limit of the optical detection system. My research aims to develop autofocusing software, to distinguish axial distances of phase-contrast yeast cell images (40-fold magnification). I first explored a deep Convolutional Neural Network, and then built other classification and regression by extracting features with different focus measure methods. The classification models trained by focus-measure features can do a quick preliminary check of in-focus and out-of-focus images with over 99% accuracy. The shallow neural network with regression and selected combination of focus measure was able to distinguish different z-stacks taken from 0 micron to 17 microns with 1-micron step- size, both above and below focal plane. The RMSE of the best validation reached 0.33 um and the best prediction accuracy was about 87% on the independent dataset, both of which outperformed the original method. Lu Bai, Thesis SupervisorRichard Wallace Robinett, Thesis Honors Advisoropen_access2021-03-25T17:41:04Zhttps://honors.libraries.psu.edu/catalog/6979yxy5293