Craig Henderson

2003 - 2005

In March 2003, I enrolled at the University of the West of England in Bristol, UK to study part-time for a Doctor of Philosophy research degree in Medical Image Analysis, working within Faculty of Computing, Engineering and Mathematical Sciences.


It has been suggested that asymmetry in bilateral mammograms is indicative of breast cancer. This project aims to investigate the significance of asymmetric features in such images with a view to improving the detection and differentiation of malignant masses, microcalcifications and other early indications. The provisional title of my thesis is Algorithms for Mammogram Analysis: A focus on Breast Tissue Asymmetry. My advisory team consisted of Dr. Tony Solomonides (Director of Studies), Dr. Jim Smith and Praminda Caleb-Solly from the Faculty. The former CEO of Mirada Solutions Ltd (now part of Siemens Medical Imaging), Dr. Ralph Highnam [2], has agreed to act as an advisor to this PhD project. Mirada is an Oxford University spin off company that specialise in mammographic image processing [5].

Project Overview

Computer Aided Mammography (CAM) generally uses Image Processing techniques on a single digitised mammogram at a time to identify regions of interest (ROIs). These ROIs are then used as inputs to a connectionist or evolutionary engine which attempts to classify them as normal or suspect pathology. It must be noted that the job of the CAM system is not to diagnose cancer (i.e. to differentiate benign from malignant). Rather, it is to aid the Radiologist in determining whether the x-ray image presents a disruption in the normal breast pattern and whether such a disruption may present itself as an abnormal process that requires further investigations [6].

There has been significant research in Computer Aided Mammography, largely focussed on the analysis of a mammogram image in isolation, or temporal image registration. My research will focus on the correlation and differential analysis of diagnostic features in asymmetric x-ray films, using two breasts to determine the normal breast pattern and to identify disruptions. This will include bilateral registration, asymmetric feature detection and modelling and the attribution of a significance quotient on each feature. I hope to identify and document the potential importance of bilateral comparisons in the diagnosis of early stage carcinoma through the development and application of an automated detection system. [4; further references therein].

Segmentation and bilateral registration

Segmentation of a mammogram to extract the breast region and concentrate further processing on the breast area in isolation of the film background is a well-documented problem, with many solutions published. I aim to review a range of what are considered to be the better approaches, and conclude by choosing a subset to use in this study. The registration of bilateral mammograms is a less well-documented research area, and the work will involve a review of the few existing approaches and further development to achieve the quality of result required for further progress.

Asymmetric feature detection and modelling

I aim to develop feature detection algorithms using the asymmetric properties of the bilateral images. Comparison of, e.g., conformally mapped or transformed images, with or without local averaging, will be considered to identify bilateral differences. It is anticipated that there will be a relatively large set of features that represent gross bilateral asymmetry. However, useful indicative features are more likely to be identified in local tissue asymmetry. Once relevant features have been identified in the image spatial domain, a feature model will be developed to analyse degree of asymmetry.

Significance attribution and feature classification

Using the abstract model of asymmetric features, and with the ability to cross-reference the features back to the source image domain, a significance quotient may be attributed to each feature. This quotient could be seen as a weight value that could give some indication as to the significance of the feature in the detection of microcalcifications or other pre-carcinoma.
Final analysis can use machine learning techniques to improve, with experience, the classification of images based on these features. I propose to use well established methods, including those developed by MammoGrid collaborators in Italy, rather than necessarily to invent new ones. Taking inputs from the specific feature attributes such as size and shape and the abstract model information of asymmetric details, I expect the system to be able to learn and adapt to achieve an acceptable level of classification accuracy.

Bibliographic References

1. Amendolia, S.R. et al. The CALMA project: a CAD tool in breast radiography Computing in High Energy and Nuclear Physics, 2000
2. Highnam, R. and Brady, M. Mammographic Image Analysis, Kluwer Academic Publishers Group 1999
3. Papadopoulos, A., Fotiadis, D.I. and Likas, A. An automatic microcalcification detection system based on a hybrid neural network classifier, Artificial Intelligence in Medicine 25 (2002) 149-167
4. Wirth, M.A. A Nonrigid Approach to Medical Image Registration: Matching Images of the Breast
5. Mirada Solutions Ltd, VirtualMammo™ (software)
6. Interactive Mammography Analysis Web Tutorial: Mammogram Analysis (online)
7. 2002 Diagnostic In-training Exam Answer Module: Section II #51-55: Breast
8. Breast Cancer Risk Factors