Development of a Population of Digital Anthropomorphic Phantoms with Simulated Acquisitions for use in Deep Learning Improvement of DMSA Quantification and Estimation of Attenuation Maps from Emission Reconstructions in DMSA Pediatric SPECT Imaging

PEDIATRICS have increased tissue radio-sensitivity and longer potential lifespan when compared to adults. Therefore, their risk of developing health problems such as cancer per unit administered activity (AA) is significantly higher than adults [1] , emphasizing the need to the reduce activity administered. Tc-99m dimercaptosuccinic acid (DMSA) SPECT imaging is performed to evaluate children with pyelonephritis and scarring to the kidneys resulting from infection [2] . CT scans are not acquired during the DMSA imaging protocol to reduce radiation exposure to pediatric patients. Instead, typically attenuation maps are formed by filling segmentations from the emission data with uniform attenuation coefficients. This process is inaccurate and can lead to reduced image quantitation and quality, which motivates our work to develop a deep learning (DL) method to estimate these attenuation maps from emission data [3] .