Visual Natural Language Processing Of Medical Images For Enhanced Value

Abstract

Nearly all medical image analysis done in modern healthcare settings is performed qualitatively by the visual inspection of a medical professional. Not only is this an inefficient process, but such analyses are vulnerable to human bias and human error. In computer science, the field of image processing exploits the numerical underpinnings of digital images, calling on a wealth of mathematical algorithms to efficiently analyze images with operations such as segmentation and feature extraction. The contents of this thesis document the development of a medical image analysis software tool, called Fractal Eyes. Fractal Eyes effectively learns a medical image dataset, by extracting critical features, using mutual information to quantify the correspondence of two unrelated image features, and using a neural network to identify whether an unlabeled image belongs to a particular dataset. The goal of the Fractal Eyes tool is to aid medical professionals in the use of medical images, by introducing a quantitative basis for analysis. This project was completed as part of the Senior Design class in the College of Engineering, and is the product of Team 18075.