Theses (Articles)

S. McAvoy

 

Abstract

Breast cancer is the most common cancer in Canadian women and early detection dramatically increases a woman's chance of survival. Until recently, womens' ages were considered the single most influential risk factor for developing breast cancer. Today, the density of fibroglandular tissue within the breast is considered just as important a risk factor as age. Because of this, accuracy and consistency while estimating tissue density is paramount. Currently, radiologists use the BI-RADS classification system to place mammographic images into one of four different categories. However, inter-observer variance has been shown to be as high as 30% and the methodology can be highly subjective. Many computer vision algorithms have been developed to automatically quantify breast density but only a few of these algorithms take advantage of the latest digital mammographic imaging technology. One algorithm, specifically designed to use digital mammography images, is explored in detail. Its ability to quantify and classify fibroglandular breast tissue is demonstrated and its accuracy is shown to be consistent with experienced radiologists. Finally, a modification to dramatically improve the running time is shown to have minimal effect on the overall accuracy of the algorithm.

 

Link to full PDF

T. Hoegg

 

Abstract

Although there are many known factors associated with an increased risk of breast cancer, age remains the main eligibility criterion for the current Screening Mammography Program of British Columbia (SMP BC). In the light of recent controversy surrounding regular breast screening, the ability of targeted screening for high-risk women is of current interest as it has the potential to sustain high cancer detection rates, reduce the number of false positive results while controlling the overall costs. Multiple models have been proposed to estimate a woman’s risk of breast cancer given her current risk factor profile, the model introduced by Gail et al. in 1989 being particularly popular. Using five-year follow-up data of 223,399 SMP BC participants, we investigate whether the probability estimates of the Gail model adequately stratify the British Columbian population into groups of high and low-risk women and, hence, provide a basis for a personalized access criterion into the Screening Mammography Program. Further, we built a breast cancer risk model for British Columbia with the goal to include a stronger set of predictor variables and improve the outcome stratification of the Gail model. Neither the Gail model nor the new risk prediction model based on SMP BC participants showed adequate stratification properties. Overall, effect sizes of all covariates were too small to clearly separate risk estimates of future breast cancer cases and non-cases. It is questionable whether changes in screening policies should be based on breast cancer risk prediction models.

 

Link to full PDF