Author ORCID Identifier

https://orcid.org/0000-0002-8917-7683

Date of Award

8-13-2019

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Dr. Yanqing Zhang

Abstract

Artificial Intelligence (AI) has impacted our lives in many meaningful ways. For our research, we focus on improving disease diagnosis systems by analyzing medical images using AI, specifically deep learning technologies. The recent advances in deep learning technologies are leading to enhanced performance for medical image analysis and computer-aided disease diagnosis. In this dissertation, we explore a major research area in medical image analysis - Image classification. Image classification is the process to assign an image a label from a fixed set of categories. For our research, we focus on the problem of Alzheimer's Disease (AD) diagnosis from 3D structural Magnetic Resonance Imaging (sMRI) and Positron Emission Tomography (PET) brain scans.

Alzheimer's Disease is a severe neurological disorder. In this dissertation, we address challenges related to Alzheimer's Disease diagnosis and propose several models for improved diagnosis. We focus on analyzing the 3D Structural MRI (sMRI) and Positron Emission Tomography (PET) brain scans to identify the current stage of Alzheimer's Disease: Normal Controls (CN), Mild Cognitive Impairment (MCI), and Alzheimer's Disease (AD). This dissertation demonstrates ways to improve the performance of a Convolutional Neural Network (CNN) for Alzheimer's Disease diagnosis. Besides, we present approaches to solve the class-imbalance problem and improving classification performance with limited training data for medical image analysis. To understand the decision of the CNN, we present methods to visualize the behavior of a CNN model for disease diagnosis. As a case study, we analyzed brain PET scans of AD and CN patients to see how CNN discriminates among data samples of different classes.

Additionally, this dissertation proposes a novel approach to generate synthetic medical images using Generative Adversarial Networks (GANs). Working with the limited dataset and small amount of annotated samples makes it difficult to develop a robust automated disease diagnosis model. Our proposed model can solve such issue and generate brain MRI and PET images for three different stages of Alzheimer's Disease - Normal Control (CN), Mild Cognitive Impairment (MCI), and Alzheimer's Disease (AD). Our proposed approach can be generalized to create synthetic data for other medical image analysis problems and help to develop better disease diagnosis model.

File Upload Confirmation

1

Share

COinS