Classification of masses in dense mammograms using Enhanced Local Ternary Pattern (ELTP)

Author: Maryam Hammad Almaeen

Almaeen, Maryam Hammad, 2020 Classification of masses in dense mammograms using Enhanced Local Ternary Pattern (ELTP), Flinders University, College of Science and Engineering

Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. 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

The World Health Organisation (WHO) has declared breast cancer as the second leading cause of cancer death in adult women worldwide after lung cancer. The possibility that breast cancer will result in a woman's death is 2.6%(1 in 38). The high mortality rate of breast cancer is due to imperfect detection techniques available. Technology used for diagnosis or mammography has utmost importance in clinical research, as a mammogram image provides a detailed information. This study proposes a technique that uses regions of interest to classify lesions present in mammogram. The proposed method utilises an extended local ternary pattern to extract the feature vector from regions of interest. In mammograms, information about texture plays a vital role in the classification of lesions. Therefore, the extended local ternary pattern is adopted in order to give information in depth texture features of the regions of interest (ROIs). To classify the lesions, different machine learning algorithms are used such support vector machine, k-nearest neighbours, and artificial neural networks classifier. The Digital Database for Screening Mammography (DDSM) is used which is publicly available. In total 101 mammograms are considered, out of 51 malignant mammograms and 50 benign mammograms. Then,1302 benign ROIs, and 1632 ROIs malignant are extracted. To standardise the ROIs, each ROI is kept fixed in size, which is 51*51. In this study, efforts are made to propose a model that can be used for effective classification of malignant and benign regions of interest, so that efficient and early diagnosis of breast cancer can be made possible. The proposed technique using KNN classifier achieved the highest sensitivity of 88.73%, and achieved AUC value of 93.89% with 17 patterns.

Keywords: classification, mass, dense mammogram, malignant, benign, ROIs, AUC, Enhanced Local Ternary Pattern, ELTP

Subject: Computer Science thesis

Thesis type: Masters
Completed: 2020
School: College of Science and Engineering
Supervisor: Dr Mariusz Bajger