CIBM SP CHUV-EPFL SECTION

Computational Medical Imaging & Machine Learning

Section Head: Prof. Jean-Philippe Thiran (EPFL)

The CIBM Signal Processing CHUV -EPFL Computational Medical Imaging & Machine Learning Section gathers researches in (bio)medical imaging reconstruction and analysis. The section investigates novel modeling, reconstruction and segmentation techniques for magnetic resonance imaging and ultrasound imaging. In particular, research projects target the structural brain connectivity estimation from diffusion MRI and its applications in neuroscience. The Section also provides expertise in machine learning, with applications to image reconstruction and analysis.

RESEARCH TOPICS

Tractography validation and structural connectivity

Description: The project aims at the validation of the brain structural connectivity estimated from diffusion MRI tractography. Tractography is an algorithmic method that estimates white matter pathways using the axon orientation sensitive measurements of diffusion MRI. Although many successful applications, tractography suffers from several limitations and biases. The goal of the project is to better characterize those limitations and optimize the reconstruction parameters for a more accurate and reproducible structural connectivity estimation.

Investigator: Gabriel Girard (EPFL)

Collaborators: Jean-Philippe Thiran, Ileana Jelescu, Giorgio Innocenti, Friedhelm Hummel, Elvira Pirondini, Elda Fischi, Marco Pizzolato, Erick Canales-Rodriguez

Computational medical imaging - diffusion MRI

Description: Diffusion MRI measurements are sensitive to the displacement of water molecules and indirectly to the properties of the substrate in which the diffuse. Although the MRI voxel size is the millimeter scale, the measurements reveal the tissue organisation at the micrometer scale, such as the tissue main orientation, cell sizes, and cell density. This project aims at addressing the challenges in estimating the microstructural properties of the tissue from the diffusion MRI signal. This is done through novel signal modeling methods, diffusion MRI protocol designs, and in-silico realistic diffusion MRI simulations.

Investigator: Jean-Philippe Thiran (EPFL)

Collaborators:

Gabriel Girard (EPFL), Ileana Jelescu (EPFL), Giorgio Innocenti (EPFL), Elda Fischi (EPFL), Marco Pizzolato (EPFL & Technical University of Denmark (DTU)), Erick Canales-Rodriguez (EPFL), Muhamed Barakovic (University of Basel), Alessandro Daducci University of Verona, Alonso Ramírez-Manzanares (Centro de Investigación en Matemáticas (CIMAT), Guanajuato, México), Emmanuel Caruyer (IRISA/INRIA, Rennes, France)

us

Computational medical imaging - ultrasound imaging

Description: Coming soon…

Investigator:  Prof. Jean-Philippe Thiran (EPFL)

Collaborators: Coming soon…

Machine Learning/Computer Vision

Description: The emergence of big visual data streams has brought a paradigm shift to many fields of machine learning, medical imaging and computer vision. The ultimate goal of this project is to develop robust deep image representations that would allow a machine to exploit vision as well as our human eye and mind does. Those representations will form the basic building block of downstream tasks in medical imaging such as anomaly detection. To do so, we use advanced deep-learning to address limitations of existing supervised deep learning and improve the generalization ability of machines on data scarcity tasks.

Investigator: Behzad Bozorgtabar (EPFL)

Collaborator: Prof. Jean-Philippe Thiran (EPFL)

Medical Image Analysis

Description: This project aims at complementing classic histo- and molecular pathology to better inform and improve the clinical management of Colorectal Cancer (CRC). Deep learning applied to complex images has the potential to revolutionize pathology, but databases of well-annotated images used to train algorithms on, are extremely rare which hinders development in this field. Our focus is to learn histopathological patterns within cancerous tissue regions through self-supervised learning and using less annotations. As our goal, this project epitomizes personalized medicine by using advanced deep-learning to improve prognostic stratification for colorectal cancer.

Investigator: Behzad Bozorgtabar (EPFL)

Collaborators: Prof. Jean-Philippe Thiran (EPFL), Prof. Inti Zlobec (Institute of Pathology, University of Bern), Prof. Henning Mueller (HES-SO Valais, Sierre), Christian Abbet (EPFL)