Author ORCID Identifier

https://orcid.org/0009-0002-0571-0274

Document Type

Thesis

Date of Award

2023

Degree Name

Master of Science (MS)

Department

Computer Science

First Advisor

KC Santosh

Abstract

In recent years, the utilization of graph-based deep learning has gained prominence, yet its potential in the realm of medical diagnosis remains relatively unexplored. Convolutional Neural Network (CNN) has achieved state-of-the-art performance in areas such as computer vision, particularly for grid-like data such as images. However, they require a huge dataset to achieve top level of performance and challenge arises when learning from the inherent irregular/unordered nature of physiological data. In this thesis, the research primarily focuses on abnormality screening: classification of Chest X-Ray (CXR) as Tuberculosis positive or negative, using Graph Neural Networks (GNN) that uses Region Adjacency Graphs (RAGs), and each superpixel serves as a dedicated graph node. For graph classification, provided that the different classes are distinct enough GNN often classify graphs using just the graph structures. This study delves into the inquiry of whether the incorporation of node features, such as coordinate points and pixel intensity, along with structured data representing graph can enhance the learning process. By integration of residual and concatenation structures, this methodology adeptly captures essential features and relationships among superpixels, thereby contributing to advancements in tuberculosis identification. We achieved the best performance: accuracy of 0.80 and AUC of 0.79, through the union of state-of-the-art neural network architectures and innovative graph-based representations. This work introduces a new perspective to medical image analysis.

Subject Categories

Computer Sciences

Keywords

Graph Neural Networks, GNN, Region Adjacency Graphs, RAGs, Chest X-Ray

Number of Pages

61

Publisher

University of South Dakota

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.