Co-PIs:

Pascal Frossard (EPFL)

Olivier Michielin (CHUV)

Problem:

Study of immune-tumour interaction

  • Need of expert pathologists to understand complex spatio-temporal patterns
  • Identification of tumour cell evolution patterns, understanding of the disease pathogenesis
  • Understanding of the spatial interaction between cells
  • Design of new interpretable NN architectures
  • Design of new spatio-temporal models
  • Performance evaluation and model validation

Solution:

  • Fitting a graphical model in each layer of the NN
  • Design new NN that will explicitly include priors and constraints
  • Dynamic NN: combination of CNN and RNN
  • Application of machine learning algorithms to mice data and then to tumor response on humans

Impact:

  • Development of new interpretable data-driven algorithms.
  • Development of new interpretable data-driven algorithm to understand the tumor microenvironment