Bayesian tissue decomposition method for spectral mammography

Breast density evaluation is an active topic in the breast imaging field. This can be accurately measured separating fibroglandular tissue from adipose tissue using material decomposition techniques. In this work, we present a Bayesian approach, based on a Poisson maximum likelihood material decomposition method and that includes an adjustable prior, known from the compressed breast thickness during a screening exam. Since some density variations in biological tissues may appear and the measured compressed thickness is not perfectly known, the proposed method moderates the prior upon the confidence on the thickness value. This work on spectral mammography has been carried out with energy-sensitive photon counting detectors that allow multi-measurements from a single acquisition unlike dual-energy digital mammography. The proposed approach has been benchmarked with a maximum likelihood method through a simulation process that included a detector response function of 60 μm CdTe X-ray detector. Our method has been investigated in severe conditions such as contrast-enhanced mammography. Hence, a 3-material decomposition base has been used, made of PMMA, water (to mimic adipose and fibroglandular tissues) and iodine. A phantom mimicking breast densities ranging from 10% to 100% with an iodine concentration set to 7 mg/mL was used as test case. In our specific conditions (49 kVp tungsten anode, 5 mAs, mean glandular dose = 0.93 mGy, ROI = 1.04 mm
²
), the Bayesian approach reduced bias and noise on both breast density and iodine concentrations. Finally, the error on iodine concentration was below 0.1 mg/mL and around 5% on breast density, that is 1.8 times lower compared to standard maximum likelihood approach.