• Academic projects,
  • Research,
  • Engineering for health,

UNGRAM : Unfolded Non-Gaussian Reconstruction Algorithms with Applications to Mammograms

Published on November 17, 2026 – Updated on November 17, 2026
The UNGRAM project aims to develop accurate breast cancer risk assessment tools which are robust to changes in the acquisition and clinical setups and to population shifts.

This represents a high-impact scientific challenge as breast cancer is the most represented cancer among women worldwide and a major cause of mortality. Early detection is key to the patient’s survival, motivating a regular screening of the populations. UNGRAM focuses on breast images, obtained from projectional X-ray mammography. A promising lever to improve breast cancer prognosis consists in an upstream assessment of cancer risk. This would enable to prescribe more frequent screenings to women at higher risk. Yet, reading X-ray images is very difficult, even for experienced radiologists as the visual appearance of healthy dense tissues and disrupted tissues might be very similar, impairing their precise discrimination by eye. This motivated the design of computed-aided tools, and in particular of a very promising breast cancer risk assessment methodology based on a characterization of mammographic textures through their fractal attributes. Nevertheless, the generalization of this risk assessment strategy into an effective clinical tool is significantly hindered by the fact that mammograms are corrupted by deformations and non-Gaussian noise, which strongly depend on the imaging apparatus and on the patients.

The objective of UNGRAM is to design robust reconstruction algorithms removing the blur and the non-Gaussian noise in mammograms while preserving the fractal properties of the underlying textures to permit accurate cancer risk evaluation. To design efficient and versatile schemes while remaining robust and explainable, model-based approaches will be intertwined with state-of-the-art deep learning techniques. Model-based iterative algorithmes will be unfolded into neural networks, trained on carefully designed synthetic textures and then fine-tuned on real data in a self-supervised way. 

JCJC Call for project

Barbara Pascal was laureate of the JCJC (Jeunes Chercheuses et Jeunes Chercheurs - Young Researchers) call for project, organized by the ANR (French National Research Agency). This initiative aims to support research projects led by young scientists and to promote the development of original research.

Portrait of Barbara Pascal, porteuse du projet UNGRAM

Since 2022, Barbara Pascal has been a CNRS Research Fellow at the Laboratory of Digital Sciences of Nantes (LS2N) as part of the Signal, Image and Sound (SIMS) team. Barbara Pascal is an expert in solving inverse problems using methods based on modelling, deep learning, or a combination of both. Her research covers both signal and image processing, with a strong focus on applications: biomedical imaging, non-linear physics and epidemiology, in particular.
 

From 2020 to 2022 she was a postdoctoral researcher at CRIStAL, working at the interplay between time-frequency analysis and stochastic geometry. She received the Ph.D degree in Signal & Image Processing from the ENSL in 2020, for her work at the Laboratoire de Physique, under the supervision of P. Abry and N. Pustelnik. She ccompleted a M.S. degree in Physics from the École Normale Supérieure de Lyon, France, in 2017 and accessed the status of Professeur-Agrégé of Mathematics in 2016.


Published on November 17, 2026 Updated on November 17, 2026