Univ. Grenoble Alpes, Grenoble, France, CEA LETI MINATEC Campus, Grenoble, France, Université de Lyon, CREATIS, CNRS UMR5220, Inserm U1044, INSA-Lyon, Université Claude Bernard Lyon 1, Centre Léon Bérard, France

Breast density is known for being a parameter used in breast cancer risk models and might be measured during mammography exams. Automated methods based on grayscale values already exist, but they only allow a density estimation over the whole breast. Material decomposition approaches make it possible to define the breast density at each pixel based on a dual energy exposure. Furthermore, the use of contrast enhanced digital mammography exams is increasing and helps physicians to highlight blood vessels with higher permeability that may be related to malignant lesions. Nevertheless, this technique also requires two x-ray exposures. In this paper, we propose to take benefit of energy sensitive x-ray detectors that give access to several energy measurements in a single shot acquisition to both quantify breast density and iodine during a single mammography exam. To this purpose, a spectral mammography acquisition chain has been simulated and different material decomposition methods have been proposed and investigated using a simulated phantom under clinically achievable settings with an average absorbed dose by the mammary gland of 0.93 mGy (mimicking a 45 mm thick breast made of 50% fibroglandular tissue) using a 49 kVp with a tungsten anode, an inherent filtration of 0.8 mmBe and an additional filtration of 1.2 mmAl. Breast density is commonly defined as the ratio between glandular tissue and the sum of adipose and glandular tissues thicknesses. Standard polynomial material decomposition methods have been extended to make them compliant with 3-energy measurements in order to allow a three-material decomposition: adipose and glandular tissues as well as iodine concentration. Those methods allow to estimate iodine concentration with an average accuracy inferior to 0.15 mg ml-1 and breast density with a precision inferior to 8% (root-mean-square errors). Nevertheless, these methods are difficult to extend to more energy channels due to the least-square model inversion. To achieve better results, a maximum log-likelihood approach has been introduced to take into consideration more spectral information while improving accuracy and precision on measurements. Moreover, the use of the proposed method permits to accurately detect the absence of iodine and significantly improves the estimation of breast density compared to polynomial approaches. We also highlighted that there is a strong interest to increase the number of energy thresholds from 3 to 6, a less significant gain is still noticeable with more energy bins. © 2017 IOP Publishing Ltd.

breast density, breast imaging, contrast-enhanced spectral mammography, material decomposition