Thesis of Pierre-Elliott Thiboud

Structure and explicability of neural networks for sepsis prevention

Start date: 03/10/2022
End date (estimated): 03/10/2025

Advisor: Mathieu Lefort


We want to develop new prognostic methods for sepsis based on machine learning, with predictions that can be explained and understood by the hospital staff, in order to improve patient management, and in particular to reduce the length of hospital stay and associated mortality. We will thus target:
- To propose an original learning structure to improve the explicability of existing models in the medical domain and validate it in the context of
context of sepsis prediction,
- To propose a method of restitution of the explanations for the relevant predictions for the hospital staff,
- Evaluate the added value of explicability and its impact on the treatment of sepsis in the hospital.