Date of Award
Spring 2022
Document Type
Honors Thesis
Department/Major
Computer Science
First Advisor
KC Santosh, PhD
Second Advisor
Douglas R Goodman, PhD
Third Advisor
Arun Singh, PhD
Keywords
Convolutional Neural Network, Binary Image Classification, COVID-19, Chest X-rays
Subject Categories
Artificial Intelligence and Robotics
Abstract
The COVID-19 pandemic has had a large effect on almost every facet of life. As COVID-19 was a disease only discovered in recent history, there is comparatively little data on the disease, how we detect it, and how we cure it. Deep learning is a powerful tool that can be used to learn to classify information in ways that humans might not be able to. This allows computers to learn on relatively little data and provide exceptional results. In this paper, I propose a novel convolutional neural network (CNN) for the detection of COVID-19 from chest X-rays called basicConv. This network consists of five sets of convolution and pooling layers, a flatten layer, and two dense layers with a total of approximately 9 million parameters. This network achieves an accuracy of 95.8 percent, which is comparable to other high-performing image classification networks. This provides a promising launching point for future research and developing a network that achieves an accuracy higher than that of the leading classification networks. It also demonstrates the incredible power of convolution.
Recommended Citation
Henderson, Joshua Elliot, "Convolutional Neural Network for COVID-19 Detection in Chest X-Rays" (2022). Honors Thesis. 254.
https://red.library.usd.edu/honors-thesis/254