Breast Cancer Risk Assessment using Mammographic Image Texture Analysis

Author: Kelly XiZhao Li

XiZhao Li, Kelly, 2014 Breast Cancer Risk Assessment using Mammographic Image Texture Analysis, Flinders University, School of Computer Science, Engineering and Mathematics

This electronic version is made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. This thesis may incorporate third party material which has been used by the author pursuant to Fair Dealing exceptions. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.

Abstract

Breast cancer is one of the most common cancers among women and early detection plays an important role in reducing the mortality and morbidity due to breast cancer. Importantly, early breast cancer detection is facilitated by accurate breast cancer risk assessment. This thesis aims to develop computer methods for analyzing tissue texture in screening mammograms in order to assess the risk of breast cancer. According to the literature, the breast density is a strong indicator of breast cancer risk and is independent of non-mammographic risk factors (age, race, family history, etc.). In addition, texture from screening mammograms is also considered to play an important role in predicting breast cancer risk. However, the contribution of texture alone to breast cancer risk is unclear and the role of texture for assessing breast cancer risk over time is also unknown. The focus of this thesis is on studying the role of texture, independent of density, in breast cancer risk assessment. In this thesis, the emphasis is on characterizing texture through the use of textons. Textons can be described as ubiquitous local texture patterns. The distribution of conventional textons (referred to as first-order textons in this thesis) has been shown to characterize texture in visual images and has been successful in tasks such as separating regions corresponding to grass from regions representing trees or animals. An important contribution of this thesis is the introduction of higher-order textons. The notion of higher-order textons is to extend the power of the first-order textons. Higher-order textons allow quantitative analysis of commonly occurring patterns of patterns, offering a mechanism for understanding more complex texture structure in images. In this thesis, textons and higher-order textons are used to distinguish mammograms from women having a high risk of breast cancer from women having a low risk of breast cancer. A number of experiments were conducted to determine the best implementation of textons and higher-order textons for breast cancer risk assessment. Results indicate that texture analysis based on higher-order textons predicts risk at least as well as any method currently available for estimating breast cancer risk from mammograms. Risk of breast cancer can be measured using texture at least four years prior to the cancer becoming apparent mammographically. In addition, a number of discoveries were made in the course of the study. Texture features from CC view mammograms (top view) perform better than texture features from MLO view mammograms (side view). Better risk assessment is obtained by measuring texture over the full breast than any particular local region of the breast. Texture features calculated from 3x3 local neighborhoods perform as good or better than texture features based on larger patches. Texture information relevant to breast cancer risk is more pronounced in the breast in which cancer eventually occurs than in the breast without known cancer of the same woman. These discoveries have potential impact on the fields of image analysis and computer-aided mammography and so form natural seeds for future work.

Keywords: Breast cancer,Risk assessment,Computer-aided,Texture features,Mammographic images
Subject: Computer Science thesis

Thesis type: Doctor of Philosophy
Completed: 2014
School: School of Computer Science, Engineering and Mathematics
Supervisor: Murk J. Bottema