Cell Microscopy Analysis

The Short: Microscopy Analysis program which makes use of machine learning techniques. The Long: The goal of this research project is to apply neural networks and machine learning to scientific problems related to machine learning and large data sets. Specifically, to track the locations of hundreds or thousands of cells in microscopy images for studies which occur on a longer time scale (24+ hours). Current research that would benefit from such a project includes that of Professor Bo Sun at Oregon State, who studies the effect of cancer cells on cellular communication. Due to the longevity of the study, staining the cells is unfeasible as it leads to premature cell death, making automated cell tracking impossible. Cell stain is contained within the nucleus (the location of DNA). This makes locating the center of the cell comparably easy. However, without a stained nucleus, cell features lack contrast, making traditional image analysis impossible. This problem is currently being rectified by graduate students performing the analysis by hand. This problem is rectified by neural networks. Neural networks mimic the human brain’s method of learning and storing information; that is, a collection of entities form pathways between each other. This allows for a training schedule consisting of both human guided and autonomous learning. Using these techniques, computer programs are able to perform image analysis and other tasks comparable to or above that of a human. Machine learning and neural networks currently perform image analysis of handwritten numbers, language translation, and are able to defeat experts at many video and board games. The use of such techniques in cell identification would allow the computer to perform image analysis of microscopy images at a level comparable to that of a graduate student.