Intelligent machine vision-based breast self-examination in android platform

Date of Publication


Document Type


Degree Name

Doctor of Philosophy in Electronics and Communications Engineering

Subject Categories

Electrical and Electronics | Engineering


Gokongwei College of Engineering


Electronics And Communications Engg

Thesis Adviser

Elmer P. Dadios

Defense Panel Chair

Laurence A. Gan Lim

Defense Panel Member

Edwin J. Calilung
Celso B. Co
Raouf Naguib
Raymund C. Sison


Breast cancer is a serious global concern, especially in low- and middle- income countries where health personnel and monetary resources are generally limited. Among the major challenges is the late presentation of the disease in which most women consult the physician during the advance stages of the disease resulting to higher mortality in general. In this context, the general aim of this dissertation is to develop a mobile technology based breast self-examination (BSE) as a clinical enhancement tool or decision support system to aid in raising breast cancer awareness and promotion of health monitoring among its users, to facilitate early reporting of unusual breast changes to clinicians. In particular, the main goal is to design and develop a machine vision-based application for BSE in Android platform, with the capability to guide, monitor, and evaluate the users performance. This study involves characterization, implementation and/or development of machine learning and vision algorithms for BSE motion analysis including the formulation of a quantitative metric for the BSE practice. Among the final output consists of a smartphone or tablet application that monitors and guides the BSE performance with its built-in camera. The mobile platform is specifically chosen for ubiquitous user access, which is among the core advantages of mobile health (mHealth) or the practice of medical and public health through the usage of mobile devices.

For system development methodology, a throwaway prototyping approach was utilized to deal with unclear requirements at the beginning, unfamiliar technology, and reliability requirements. Throwaway prototyping involves planning, analysis, analysis-design-designprototype-implementation (ADDPI) cycle, design, and implementation. Initially, due to the scarcity of existing literature on the proposed machine vision-based BSE system, its scope, requirements and its feasibility were largely based on consultation with domain experts, i.e. oncologists, physicians, health informatics researchers, and potential users. The planning and analysis phase determined 3 (three) sub-problems for enabling BSE motion analysis, namely, (1) Areola detection for breast presence, (2) fingertip region detection & tracking, and (3) palpation detection & depth estimation. Consequently, ADDPI cycle considered the aforementioned sub-problems and characterized various machine learning/vision algorithms for their solutions, i.e. Local Binary Pattern (LBP)/ Haar Cascade Classifiers, Bag-of-features using Support Vector Machine (SVM), Multilayer Perceptron (MLP), and K-Nearest Neighbors (KNN), and visual object tracking (VOT) methods, e.g. Lukas-Kanade (LK) Optical flow-based methods, Kernelized Correlation Filter (KCF), Normalized Cross-Correlation (NCC), Tracking-Learning-Detection (TLD), and Multiple Instance Learning (MIL) trackers. The final design and implementation includes: (1) an LBP cascade classifier for Areola detection that achieved a testing F1-score of 0.98, running real-time at 17 fps on a low-end smartphone device (2) Histogram of Gradients (HOG) based KCF tracker for fingertip region tracking with the best precision running at about 5.6 fps on a low-end smartphone device.

Abstract Format






Accession Number


Shelf Location

Archives, The Learning Commons, 12F Henry Sy Sr. Hall

Physical Description

1 computer optical disc; 4 3/4 in


Breast—Examination; Breast—Cancer; Awareness; Mobile communication systems--Design

This document is currently not available here.