Speaker
Description
The detection of microcalcifications within breast tissue, composed of fibroglandular and adipose components, is a key indicator in the diagnosis of breast cancer. Differentiating between type 1 (calcium oxalate, CaOx) and type 2 (hydroxyapatite, HA) microcalcifications is clinically important, as they are associated with benign and malignant lesions, respectively. Conventional energy-integrating detector (EID)-based mammography lacks the spectral capability for material decomposition. Photon-counting detectors (PCDs) provide energy-resolved imaging for material decomposition, but their high cost and limited availability restrict clinical use. We propose a method that uses a modified U-Net to generate PCD-like spectral projections from EID data for material decomposition. Reference PCD data were generated via CdTe-based PCD simulation using the Photon Counting Toolkit (PcTk), configured to replicate realistic detector response and energy binning. The generated spectral projections were decomposed into CaOx, HA, and soft tissue maps. Preliminary results showed effective separation of soft tissue and microcalcifications with high selectivity. Quantitative results and further discussion will be provided in the full paper.
