27 nov 2023 -- 15:30
Abstract.
The amount of fibroglandular tissue within the breast is one of the strongest risk factors for breast cancer. It is often measured using breast x-rays (mammograms) done during routine breast cancer screening appointment. Mammographic density is often measured visually by trained readers as the proportion of radio-opaque tissue on the mammogram. Breast density decreases with age so it is possible to predict a woman’s age from her breast density. We aim to improve risk prediction of future breast cancer by using this to estimate ‘breast age’ from the mammogram, and compared it to a woman’s actual age at the time of the mammogram. We hypothesize that women whose breast age is younger than their actual age will be at increased risk of breast cancer. The issue that we consider in this talk is how to measure breast age. We do
so harnessing Persistent Homology (PH) based on information from mammogra- phy images, that is then fed into a traditional computer visionmachine learning
pipeline. Improved risk assessment may be clinically useful because there are in- creasing moves to replace one-size-fits-all mammography screening for breast cancer
by risk-adapted screening, where both the frequency of screening and the modality are chosen based on risk of breast cancer. PH is a tool used in topological data analysis (TDA) to study qualitative features of data. It is robust to perturbations of the input data, independent of dimensions and coordinates, and provides a compact representation of the qualitative features of the input. It can be encoded in the form of Persistence Barcodes (PB). In this talk, we use an image-based approach to build several PB representation for each mammogram, based on a landmark selection method known as local binary patterns, that encode different types of local texture and structure from a digital image. We will present some results of a classification experiment where we predict the age of a patient from their mammogram.
This work is part of the CRUK funded project (reference 49757A28689) “An Ar- tificial Intelligence System for Real-time Risk Assessment at Mammography Screen- ing (Mammo AI)”.