Mammographic breast density (BD) is a significant breast cancer risk factor, second in magnitude only to inherited BRCA mutations. Most research studies generating this conclusion used an operator-assisted method (applied to digitized film) to estimate the percentage of BD (i.e. PD, the standard), which requires an expert technician to outline the breast region and define thresholds. Although clearly an invaluable research tool, this standard does not lend itself to automation, and is therefore not amenable for application in the clinical setting (i.e. large-scale implementation) for patient risk assessment and management. Our goal is to lay the foundation for translating the demonstrated research value of BD into the clinic by advancing our recent achievements in full field digital mammography (FFDM), the emerging standard modality for breast screening in the US. We developed a calibration system for FFDM using a specific unit that produced four significant findings: (1) a standardization technique that makes pixel values comparable across all images, (2) a new calibrated spatial variation BD measurement (or Vc) that offered a stronger measurement of risk than the standard, (3) Vc is a function of PD, another calibrated measure of BD that is also a significant risk factor, and other important risk covariates, i.e. high correlation but non-linear, and (4) demonstrated the variation measure (or V) applied to raw mammograms (or Vr) is also a significant breast cancer risk factor. In this proposed work we build on our calibration approach and apply it to different FFDM units. We will validate the Vc and Vr measures from different FFDM technology and make comparisons with our previous findings using a matched case-control study using both pre-existing and new FFDM datasets. Because differences in detector designs have the potential to alter spatial variation, it is imperative to assess these influences n the new V-metrics to demonstrate that breast cancer risk is not dependent upon the system design. We will quantify the gains derived from calibration by comparing Vc and Vr, because gains are derived at the expense of advanced image processing and analyses. We will determine the optimal breast density measure and representation (i.e. is calibration required), where optimal is defined by these attributes: automated, quantitative, reproducible, consistent across different imaging platforms, and offers risk prediction at least equivalent with that offere by PD. To meet our objectives, we use accepted techniques and introduce novel analysis strategies that include statistical learning to better capture the relationships between the import risk covariates. This work will provide a prescription for making the optimal BD measurement. The successful completion of this work will allow the full scale integration of BD into the clinica environment. Potential applications include personalized care of patients in terms of screening frequency, risk reduction interventions, and the identification of situations where mammography may be ineffective (i.e. where dense tissue significantly reduces either sensitivity or specificityof mammography).
Mammographic breast density is a significant breast cancer risk factor, second in magnitude only to inherited BRCA mutations. To date, the best way to calculate this measurement is not well understood. However, an operator-assisted measure is often used for research purposes. Due to measurement complexities breast density is not used for clinical based risk assessments. The focus of our work is full field digital mammography (FFDM). We use FFDM because the majority of accredited mammography facilities now operate FFDM units and FFDM will replace film systems eventually. Moreover, digital images from FFDM facilitate the processing methods. We are developing a method to measure this important factor from mammograms. We used experimental phantom imaging to adjust mammograms to make improved measures of breast density for full field digital mammography. Using advanced image processing techniques, we normalized mammograms so that they are comparable. This process allowed us to develop a new automated method of measuring breast density that shows stronger correlation with breast cancer than other measures. In this proposed work, we will explore the generalizations of our novel measure and evaluate it further with different types of mammography units. The work has important healthcare implications upon completion. This project will provide a technique that can be used in the clinic to (a) collect important serial brest density information, and (b) develop breast cancer risk models that include breast density.
|Fowler, Erin E E; Lu, Beibei; Heine, John J (2013) A comparison of calibration data from full field digital mammography units for breast density measurements. Biomed Eng Online 12:114|