Era of Hope 2011 Investigator Vignette

Title: Image-Based Biomarkers for Breast Cancer Risk

Investigator: Fengshan Liu, PhD, and Charlie D. Wilson, PhD; Delaware State University

Dr. Liu: The current grant obtained from DoD in the area of minority training in the breast cancer. With this grant the goal is to train a group of Delaware State University faculty to do breast cancer research. This group of people come from many different areas; in fact there's members from mathematics, biology, computer science, and also faculty members from University of Pennsylvania work with us. So the eventual goal is to establish breast cancer research programs at Delaware State University, which we have never done before.

Dr. Wilson: This grant has several prongs. One of the facets of it is to establish software to allow us to analyze the images. Most of the mammograms these days are digital which allow us to use computational tools to analyze them for texture and about 25 other features and extract those features and analyze whether there's differences between different populations such as between minority groups and Caucasian groups.

So, one of the aims of this is to develop computer models to address the problem of mammography of large breasts. Many times there are several images needed to capture the entire breast and then those have to be put together computationally to analyze the breast as a whole. It's like building a jigsaw puzzle-a 3-D puzzle, by taking some features that are common in one but absent in the other and using those as markers for alignment to get a total picture.

And so one of the goals is to develop what we call phantoms which are computerized simulations of breasts as you might see on a mammogram and-and develop the programs to allow analysis of those breasts for multiple images. And also to extract those features we talked about such as density, texture. The images we required-over 11,000 images will be processed through a computational pipeline and do-retrieve data and then that data will be used to analyze the images and get some feature extractions and-and determine whether or not there are differences between different groups as far as mammography and how that might be a predictor for cancer risk.

Here we see an example of a computerized phantom; a computer generated image that mimics a mammogram and we use those to look at different sections and what features might be in that section because you want to be able to distinguish what may be a tumor versus what may be a normal breast structure versus what may be fatty tissue. So by using computers to extract that information, we can more accurately simulate analysis of real mammograms by looking at computerized phantoms of the breasts.

Here is just a flowchart where these images again are sent -- around 11,000. We send them through a series of computational steps, and we can extract features such as the density of the region or entropy or number of other factors that we can use to compare different breasts from different sources. And this allows us to be able to establish those procedures at Delaware State University from U-Penn and become independent in our analysis of these mammograms.

Several different programs are being used for this, different platforms, so one of the challenges is getting together what we call scripts that allow different programs to talk to each other and that's being done. Right now we're working with Despina's group, a group at U-Penn; we also are starting to develop the tools to analyze this thing and get this-the future ones to extract. So in the near future we should be able to automatically send these things-these 11,000 cases through the analytic pipeline and generate data, a lot of data. Each image will generate 52 sets of data. And so you can imagine with 11,000-it's quite a bit-but we'll use the computer tools later on to analyze those and look for the differences between the African American population and Caucasian population. And if they are confirmed through further analysis; they may serve as markers-you can see someone with the certain features and it may serve as predictive value to see whether or not the risk of having breast cancer is greater because of some digital features we might be able to extract from the mammogram.

Dr. Liu: The DoD supports new ideas, some idea even risky ideas. The idea is to get more people involved and more institutions involved particularly if the institutions have never done this before to put new ideas, new brains-not only from one certain area; so we need multiple groups together to work to find out new ideas, new measures to solve the breast cancer.