CESM@UCBM

Deep generative modelling for virtual contrast enhancement in spectral mammography, supported by a publicly released in-house CESM dataset.

CESM@UCBM

Clinical Application

CESM@UCBM

Deep generative modelling for virtual contrast enhancement in spectral mammography, supported by a publicly released in-house CESM dataset.

Generative AI Mammography Dataset Curation

Overview

CESM@UCBM studies virtual contrast enhancement in contrast-enhanced spectral mammography. The central technical goal is to generate synthetic recombined images from low-energy acquisitions only, preserving the clinical value of CESM while reducing the burden associated with contrast administration and higher radiation exposure.

The project is also a data resource effort. It released a curated in-house dataset of 1138 CESM images collected at the Fondazione Policlinico Universitario Campus Bio-Medico, with acquisition metadata and lesion-related annotations that support reproducible research on generative models and breast imaging analysis.

Research Directions

  • Virtual contrast enhancement from low-energy images only, reducing reliance on contrast medium and double acquisition.
  • Public release of a 1138-image CESM dataset with imaging and clinical annotations.
  • Comparative evaluation of autoencoder, Pix2Pix, and CycleGAN approaches with radiologist assessment.

Related Publications