Clinical

Clinical decision support systems in anesthesia are generally limited to simple thresholding or variation detection. Moreover, existing closed-loop control strategies are often poorly adapted to the practical needs of anesthesiologists.

This project aims to exploit the potential of control, optimization and learning methods to propose original approaches for fusing the various measured data and designing advanced detection and control systems that address certain clinical challenges.
A non-exhaustive list of these challenges includes:

• Presence of high uncertainties in biomedical systems: physiological models are often hypothetical and their parameters are described by probability distributions.
• Reliability of measurements: measurements can be affected by drug interactions compromising their interpretability (for example, the influence of ketamine on the bispectral index), or linked to poorly understood clinical phenomena such as analgesia (absence of pain), which has no direct indicator and must often be inferred from several clinical signals.
• Low persistence of excitation in inputs: drug infusion profiles are subject to various clinical constraints, limiting the identifiability of model parameters.
• Critical events impacting control performance: for example, a hemorrhage that can modify drug concentrations in the bloodstream and thus affect system dynamics.

The Clinical project aims to address some of these complexities via learning, probabilistic and stochastic control, and Model Predictive Control approaches. It will exploit both clinical data and physiological simulators to design more realistic, practice-oriented control strategies, taking into account feedback from anesthesiologists.