Automated Quantitative Measures of Breast Density

AQMBD-Heine 2013

No coordinating investigator defined.

Breast density
No design specified.
Other, Specify
Breast and Gynecologic Cancers Research Group

Current Objective: This project’s goal is to determine the optimal breast density measure and representation using full field digital mammography (FFDM). We define optimal as: a method that is automated, quantitative, consistent across imaging platforms, reproducible, and offers prediction of breast cancer risk at least equivalent with that offered by PD. Translational Objective: We will lay the foundation with this proposed work to move breast density beyond research into the clinic. Potential applications include personalized care of patients in terms of screening frequency, due to individualized risk models, risk reduction interventions, and the identification of situations where mammography may be ineffective (i.e. where dense tissue significantly reduces either sensitivity or specificity of mammography) and subsequent use of Magnetic Resonance Imaging (MRI) may be more effective.

Background: Mammographic breast density (BD) is a significant breast cancer risk factor, second in magnitude only to inherited mutations in BRCA1/2. Most research in this area to date used an operator assisted approach to estimate the percentage of breast density (PD). Although an important research tool, PD is somewhat subjective, does not lend itself to automation, and is therefore not amenable for application in the clinical setting for patient risk assessment and management. Preliminary Data and Justification: We recently developed a calibration system to adjust for image acquisition technique differences for FFDM. After applying the calibration, inter-image pixel values are comparable. From this research, a new variation measure of breast density (or V) emerged that provided stronger breast cancer risk prediction than PD. We also showed that there is a non-linear functional relationship between V, PD, and several other relevant and highly correlated covariates. Analogous measures of V were also derived from both non-calibrated FFDM and digitized film images. These various non-calibrated V measurements showed both stronger (the majority of comparisons) and equivalent risk association than offered by PD, indicating the form of detection may impact this new measure more so than PD. Because V measures variation rather than relative brightness (as does PD), we have insufficient evidence to determine whether (a) V is independent of the detection system, or (b) depends on the calibration step, which if so will provide a benchmark (i.e. define the technical effort requirement) to automate the V measurement. There are three Specific Aims required to meet our objectives. These will be examined using FFDM data from a large matched case-control study of breast cancer. Specific Aim 1: A calibration methodology will be established for newer model FFDM units. Inter-FFDM system (i.e. inter-manufacturer) V-BD consistency will be assessed to demonstrate that V is invariant with respect to technological design differences. V-BD will be compared with PD to demonstrate it shows stronger association with breast cancer. Expected Outcome: We will validate the calibrated V measure (or Vc) with a larger dataset and demonstrate that Vc is not dependent upon the FFDM technological design. This is a crucial step in developing the optimal measure since risk should be independent of system design. Alternatively, this aim may demonstrate there is a preferable (technological-design) FFDM unit for risk prediction. Specific Aim 2: We will determine whether calibration is required for the development of the optimal breast density measurement by comparing intra and inter-machine (i.e. inter-manufacturer) calibrated and non-calibrated V measures. We will also compare the non-calibrated measures with PD. Expected Outcome: Because the calibration-system requires extensive experimentation and advanced image processing before it is possible to calibrate prospective mamm
Advanced image processing

Version 5.1.0