Files in this item



application/pdfECE499-Sp2019-chemaly.pdf (1MB)Restricted to U of Illinois
(no description provided)PDF


Title:A Machine Learning Segmentation for Cell Tracking and Analysis
Author(s):Chemaly, Anthony
Contributor(s):Popescu, Gabriel
Subject(s):Machine Learning
Cell Tracking
Abstract:With the rapid increases in hardware capability in recent years, machine learning is becoming more prevalent in both academia and industry. Part of this research was to segment cells with standard methods (thresholding, dilation/erosion, and watershed transform) and then train a neural network to segment unedited images. The purpose of this was to see if a neural net can indeed segment images and if so to what degree. If this is successful, the network will be able to detect cancer more accurately than current detection methods. We trained the network to segment cells with upwards of 1,000 images and made adjustments so it would be as close to the standard segmentation results as possible. CHO (Chinese hamster ovary) cells were used to collect data. These cells were selected because they have a fairly circular structure which makes it easier for the network to locate and segment each cell. The other half of the research was using cell cluster tracking to determine if cell clusters influence each other. Does one cluster’s growth rate affect the rates of other clusters in the surrounding area? If so, what is their sphere of influence? Using the images we segmented with standard methods, we analyzed the properties of the cells. Recording the parent of a given cell after separation proved to be a difficult challenge. Clusters of cells are hard to identify due to the varying mobility and growth rates of the cells. We recorded their mass, cluster location, lineage, and growth rates. In short, the work yielded two separate results. Machine learning resulted in a neural network being able to predict CHO cells with a high degree of accuracy given a binary mask and a raw image. Through cell analysis we showed the growth rates, mass, area, and lineage of our data.
Issue Date:2019-05
Date Available in IDEALS:2019-06-19

This item appears in the following Collection(s)

Item Statistics