Computer Aided Breast Cancer Risk Assessment using Shape and Texture of Breast Parenchyma in Mammography

PhD Defence by Gopal Caremore


The fundamental goal of this dissertation is to investigate the potential of mammographic parenchymal texture as an imaging biomarker of (i) General risk of developing breast cancer, (ii) Effect specific drugs involving various HRT treatments, and (iii) Estrogen Receptor sub type specific risk.

In first part of the work, the significance of mammographic parenchymal texture in general risk assessment is introduced and existing approaches are surveyed. An accurate and consistent mammographic texture resemblance (MTR) marker is developed and evaluated on various public/private datasets. Texture features based on multiscale such as Fractals, Structure Tensors, and Gaussian Derivatives (n-jet) are studied.

Second part is focused on improving the performance of MTR by developing an anatomically oriented breast coordinate system. Its potentials are investigated by comparing its performance on existing mammogram registration techniques in longitudinal study. In this part we also introduce the committee based machine learning approach that shows the improvement in classification accuracy.

In third part, we introduce a nested cross-validation framework that automatically selects the inner and outer scale of probability map of post processed mammogram in breast coordinate system. This will help to understand and investigate the region on mammogram that experiences the maximum effect of carcinogenesis in parenchymal tissue structure during the development of breast cancer in case-control study design.

Forth part aimed at qualifying the MTR marker as an effect specific measure in clinical trials involving various HRT treatments. Here we study the coherence properties of structure tensor and structure enhancing diffusion that will help radiologist in visualizing the structural pattern change in one-to-one correspondence between temporal mammograms of populations with placebo and HRT treatment populations.

In last part, we investigate the potential of MTR as a surrogate marker of risk to develop Estrogen Receptor subtype-specific breast cancer, compared to standard mammographic density measures. There by to establish a non-invasive biomarker during screening process to identify woman who would benefit most from SERM (selective Estrogen Receptor Modulators) Chemo-prevention.

Conclusively, the whole intension of this dissertation is to help in identifying the woman who has higher risk of developing general/subtype specific breast cancer, based on fully automatic mammographic parenchymal texture marker in screening mammograms. Such identification leads to better allocation of screening resources and thereby earlier cancer detections and lower mortality/unnecessary biopsy rate.

Assessment Committee:

Chairman: Christian Igel, Department of Computer Science, Copenhagen University
Member 1: Professor Michael Brady, Department of Oncology, Oxford, England
Member 2: Associate Professor Carla van Gils, Julius Center for Health Sciences and Primary Care, Epidemiology UMC Utrecht

Academic supervisor:

Mads Nielsen and co supervisor Sami Brandt

For an electronic copy of the thesis, please contact Dina Riis Egholm,