Room: Track 1
Purpose: To quantify breast glandular tissue volume and volumetric glandular ratio using dual-energy mammography images acquired with a commercial unit.
Methods: The proposed method is based upon the existence of an invertible mapping between material thickness and pixel value in images acquired with two different X-ray spectra. To calibrate, images of variable thicknesses of aluminum (Al) and PMMA were acquired using dual-energy technique (Anode/Filter/X-ray tube voltage): Low Energy (LE)- W/Rh/31kV; High Energy (HE)- W/Al/45kV plus 5mm Al as external hardening filter. Linear and 8-parameter rational expressions were fitted to the material thickness, HE, and LE pixel value data set, which consisted of 43 points. Following the basis material decomposition formalism, for a given glandular/adipose tissue thickness there exists an equivalent Al and PMMA thickness. Once the mapping function is determined, it is possible to generate glandular/adipose thickness images from Al/PMMA thickness images and use the former for glandular density calculation. The material decomposition algorithm was used to process 10 clinical images, yielding the following
results: glandular, adipose and total thickness maps, glandular volume, total breast volume, and volumetric glandular ratio. A pre-processing routine involving downsampling and skin masking was performed on the clinical images.
Results: Glandular and total breast volume were calculated for the 10 images. The obtained volumetric glandular ratios were compared to existing models of glandular ratio as a function of compressed breast thickness, and an overall agreement was found (Mean absolute error = 5.8%).
Conclusions: This work explores the feasibility of using dual-energy imaging to quantify the glandular volume and volumetric glandular ratio. The reported results are considered positive. These are yet to be compared with validated measurements, and that will be the focus of the next actions. A detailed assessment of the global uncertainty and robustness associated with this methodology is also underway.
Funding Support, Disclosures, and Conflict of Interest: GPG acknowledges an M.Sc. fellowship from CONACYT, authors acknowledge partial funding from UNAM PAPIIT-IN103219.