Developing Measures of Interval Breast Cancer Risk
Benjamin John Hinton
One in eight women experiences breast cancer in their lives, making it the number one cancer affecting women. Interval cancers are cancers that are found by women in between normal screening mammogram intervals, and often represent a failure mode of mammography screening where it was unable to detect a lesion. While supplemental screening can help prevent these interval cancers, referring all women with interval cancer risk to supplemental screening would lead additional biopsies, stress, and false positives. A need exists to create a specific measure of interval risk that could help identify women that would truly benefit from supplemental screening.
The goal of this dissertation was to develop that measure of interval risk and is centered on the hypothesis that advanced computer vision methods can identify metrics of detectability that quantify risk of interval cancer more effectively than currently used metrics of breast density. I first present methods to better quantify composition of limbs, and trunk, as compositional measures have been shown to relate to cancer risk. I then present an algorithm that directly quantifies mammogram detectability by creating and inserting pseudo-lesions in digital mammograms and summarize its ability to quantify interval risk. I lastly present a method using deep learning to create and train classifier to identify risk of interval cancer. I describe the underlying hypotheses behind these methods, their effectiveness of identifying interval risk compared to current gold standards, strategies for improving their effectiveness, and future steps.
In conclusion, we have developed several methods that help to improve upon current measures of interval cancer risk. Further work to refine and develop these methods could be applied to improve risk models, identify groups of women at high risk of interval cancers, develop software to aid radiologists, and help radiologists identify women who would benefit from supplemental screening. As a result, these methods may be able to help prevent interval cancers, improve sensitivity of screening, and save lives. The following chapters outline these tools and methods, as well as ways they can be applied.
