Speaker
Description
As part of the ICSC initiative (National Research Center in High-Performance Computing, Big Data, and Quantum Computing), funded by the PNRR (National Recovery and Resilience Plan), the INFN Milano computing center has expanded its capacity by deploying a bare-metal Kubernetes cluster, which hosts an HTCondor cluster running in Docker containers.
This infrastructure is currently used to support a large-scale workflow aimed at estimating the average glandular dose (AGD) in x-ray breast imaging. This study leverages advanced anthropomorphic digital breast phantoms, generated from clinical data through machine learning techniques, to overcome the limitations of traditional breast models.
The workflow is based on extensive Monte Carlo simulations implemented with the Geant4 toolkit, requiring up to 10^10 primary photons per run in voxelized geometries. To cover the full parameter space (spectra, geometry, and patient variability), up to 126,000 simulations are needed.
The availability of cutting-edge hardware has allowed to reduce the computation time by a factor of three with respect to a previously adopted cluster, enabling the execution of this large-scale campaign. Preliminary results based on Generalized Additive Models (GAM) show unbiased estimates of patient dose.