Abstract
Abstract
The purpose of this study was to investigate the use of a Gallium Arsenide (GaAs) photon-counting spectral mammography system to differentiate between Type I and Type II calcifications. Type I calcifications, consisting of calcium oxalate dihydrate (CO) or weddellite compounds are more often associated with benign lesions in the breast, and Type II calcifications containing hydroxyapatite (HA) are associated with both benign and malignant lesions in the breast. To be able to differentiate between these two calcification types, it is necessary to be able to estimate the full spectrum of the x-ray beam transmitted through the breast. We propose a novel method for estimating the energy-dependent x-ray transmission fraction of a beam using a photon counting detector with a limited number of energy bins. Using the estimated x-ray transmission through microcalcifications, it was observed that calcification type can be accurately estimated with machine learning. The study was carried out on a custom-built laboratory benchtop system using the SANTIS 0804 GaAs detector prototype system from DECTRIS Ltd with two energy thresholds enabled. Four energy thresholds detector was simulated by taking two separate acquisitions in which two energy thresholds were enabled for each acquisition and set at (12 keV, 21 keV) and then (29 keV, 36 keV). Measurements were performed using BR3D (CIRS, Norfolk, VA) breast imaging phantoms mimicking 100% adipose and 100% glandular tissues swirled together in an approximate 50/50 ratio by weight with the addition of in-house-developed synthetic microcalcifications. First, an inverse problem-based approach was used to estimate the full energy x-ray transmission fraction factor using known basis transmission factors from varying thicknesses of aluminum and polymethyl methacrylate (PMMA). Second, the classification of Type I and Type II calcifications was performed using the estimated energy-dependent transmission fraction factors for the pixels containing calcifications. The results were analyzed using receiver operating characteristic (ROC) analysis and demonstrated good discrimination performance with the area under the ROC curve greater than 84%. They indicated that GaAs photon-counting spectral mammography has potential use as a non-invasive method for discrimination between Type I and Type II calcifications. Results from this study suggested that GaAs-based spectral mammography could serve as a non-invasive measure for ruling out malignancy of calcifications found in the breast. Additional studies in more clinically realistic conditions involving breast tissues samples with smaller microcalcification specks should be performed to further explore the feasibility of this approach.