Research Grants Awarded
Tomosynthesis Texture Biomarkers For Computer-Assisted Risk Estimation (Care) Of Breast Cancer
Several studies have demonstrated a relationship between parenchymal patterns in mammograms and breast cancer risk. Growing evidence suggests that there is a strong association between extensive mammographic breast density and a higher risk of developing breast cancer. Computerized analysis of digitized mammograms has also shown the potential to distinguish the parenchymal patterns of BRCA1/2 gene mutation carriers using image texture. The analysis of parenchymal patterns in breast images provides the foundation for developing robust image biomarkers that can be used to improve cancer risk estimation. Currently, breast cancer risk assessment is limited due to the constraints imposed both by the existing epidemiological risk estimation models, and by the breast imaging methods that have been considered to date.
The current gold standards for breast cancer risk estimation, the Gail and Claus models, are based primarily on non-modifiable risk factors; for this reason they lack the flexibility to adjust risk levels after risk reduction interventions. In addition, being designed from population statistics, they are not accurate in predicting cancer incidence for individual women. Recent studies have demonstrated the potential to improve the Gail model by including mammographic density descriptors; parenchymal patterns are considered as a surrogate of modifiable risk factors such as hormonal levels, diet, and body mass index. However, mammograms are 2D projection images in which the 3D fibroglandular tissue is superimposed. As it is likely that breast cancer risk is related to the volumetric distribution and amount of glandular tissue, 3D imaging of the breast could provide more realistic means for parenchymal analysis. Tomosynthesis is a novel x-ray imaging modality in which 3D images of the breast are reconstructed from a limited number of projection images. The 3D nature of tomosynthesis allows for more accurate estimation of parenchymal properties.
The aim of the proposed translational research project is to develop tomosynthesis biomarkers that can be used to improve breast cancer risk estimation in clinical practice. Our Computer-Assisted Risk Estimation (CARe) study aims to: (i) determine the discriminative value of individual parenchymal texture features as a surrogate of cancer risk, (ii) develop a better model for breast cancer risk estimation using tomosynthesis texture and demographic data, and (iii) investigate the potential of tomosynthesis to provide superior risk assessment compared to mammography. The proposed research plan is designed as a pilot study with the intention to develop sufficient preliminary evidence to justify a large prospective cohort clinical trial; following a large cohort of women over an extended time period (i.e. number of years) will allow us to fully evaluate the predictive value of tomosynthesis biomarkers for cancer risk estimation. Our long-term goal is to test the hypothesis that tomosynthesis risk biomarkers combined with demographic and clinical information can outperform the current Gail and Claus models for cancer risk estimation.
Three specific aims are proposed: (SA1) Compute texture features from tomosynthesis images and corresponding digital mammograms; images from two on-going clinical trials in our department will be used retrospectively. (SA2) Develop and optimize models for cancer risk assessment. Linear discriminant analysis and artificial neural networks will be implemented. The models will have as input texture features alone or a combination of texture and demographic data. (SA3) Test hypotheses to evaluate superiority of tomosynthesis in breast cancer risk estimation. Receiver Operating Characteristic (ROC) curve analysis will be performed to evaluate the superiority of tomosynthesis based models.
A cross-sectional analytical study will be performed. Due to the limited time frame of the proposed study, our analysis will be based on cancer prevalence as the closest estimate of cancer risk. Only the breasts contralateral to cancer will be included in the analysis for women with cancer; the parenchymal properties of the unaffected breast will be considered as a surrogate for breast cancer risk. Women with bilateral breast cancer will be excluded.
Personalized diagnosis and treatment are currently being considered as the future of medicine. Developing breast cancer risk biomarkers based on tomosynthesis parenchymal texture could be of great clinical use for customizing detection, tailoring individual treatment and forming preventive strategies, especially for women associated with a higher risk. Identifying women that are of high risk for developing breast cancer is of primary significance in preventing breast cancer; modern risk-reduction therapies can inhibit the development of the disease. The current standard Gail and Claus models for risk prediction are not able to re-assess a woman?s risk after participating in risk reduction interventions. For this reason, the development of image biomarkers that are indicative of modifiable cancer risk factors has great potential to improve the accuracy of cancer risk estimation. A refined breast cancer risk estimation model would provide the prospect for reducing breast cancer mortality during the next decade; a cure of breast cancer begins with the effective prevention of the disease. The improved performance and low cost of breast tomosynthesis will likely fuel the rapid and broad dissemination of tomosynthesis as a breast cancer screening modality. Therefore, a refined cancer risk estimator based on tomosynthesis parenchymal texture would be of great clinical advantage.
Currently, there are two broadly recognized approaches for estimating a woman?s risk of developing breast cancer. The first approach, also implemented in the National Cancer Institute (NCI) on-line Breast Cancer Risk Assessment Tool, is based on epidemiological risk assessment models developed by Gail; these models estimate a woman?s lifetime probability of developing breast cancer. The second approach estimates a woman?s breast cancer risk by measuring breast density from mammograms. Both these approaches have potential problems. With the exception of childbirth, the Gail model is based primarily on non-modifiable risk factors, such as history of menstrual period, first degree relatives with cancer, and prior biopsies; for example, once a lifetime probability of cancer is estimated for a woman, it is difficult to accurately estimate how much a woman?s risk is altered by risk reduction therapies. In addition, it has been shown that the Gail model is not accurate in predicting which individual women will actually develop cancer. In the approach for estimating breast cancer risk by measuring breast density, mammography imposes restrictions. Mammograms are 2D images in which the breast tissues overlap; for this reason, breast density cannot be accurately measured. It is anticipated that 3D breast images should provide a better way of measuring density.
Digital breast tomosynthesis is a new, affordable, x-ray imaging technique that creates a 3D image of the breast; this imaging technique is very likely to replace mammography in the near future. Tomosynthesis images are 3D images that can show the actual distribution of breast tissues inside the entire volume of the breast; for this reason tomosynthesis should allow for more accurate measures of breast density. In addition, scientists have recently shown that a woman?s breast density can reflect changes in underlying factors that are associated with breast cancer risk, such as changes in diet, hormones, and breast cancer genes. For this reason, breast density measures could be used by doctors to estimate adjustments to a woman?s breast cancer risk after she has been treated with risk reduction therapies. Having the advantage of 3D imaging, tomosynthesis allows us to study these changes in breast density more accurately. For all these reasons, we believe that tomosynthesis breast density analysis could provide better estimation of breast cancer risk.
The goal of the proposed Computer-Assisted Risk Estimation (CARe) study is to investigate the ways to improve breast cancer risk estimation by analyzing tomosynthesis images. Currently, several clinical trials on breast tomosynthesis are being conducted in our department. These data will be used to analyze breast density in tomosynthesis images by using computerized analysis to calculate tomosynthesis image texture measures; we will compare to breast density measures in mammography. We will use tomosynthesis image texture analysis to develop a better method to estimate a woman?s breast cancer risk; our goal is to develop breast cancer risk biomarkers from tomosynthesis images that can be used to improve the Gail models for breast cancer risk estimation.
Dr. Kontos, who will be the postdoctoral fellow to perform this work, is a woman scientist who has stated that her life goal is to fight breast cancer. Although still very young in her career, she has won more than a few prestigious awards for her research. Dr. Kontos? goal is to develop biomarkers of breast cancer risk using computer analysis of breast images. Dr. Maidment and Dr. Conant, who will supervise Dr. Kontos, are known world-wide for their discoveries in breast imaging and their clinical studies on breast cancer.
The ability to improve breast cancer risk estimation is critical for the prevention of breast cancer, particular for women that are in high risk. By identifying high-risk women during breast screening procedures it would be possible to perform cancer risk reduction therapies that can prevent the development of breast cancer. Physicians would also be able to recommend the best modality and the best frequency of screening for women at high risk to help detect cancer at a very early stage; this would enable more effective treatment and save more lives. This project has a great potential to lead to the reduction of breast cancer deaths during the next decade; a cure of breast cancer begins with the effective prevention of the disease.