Self-learning fuzzy logic control of anaesthetic drug infusions

We are researching into clinical applications of self-learning fuzzy logic control, an intelligent form of direct adaptive control which permits online learning of fuzzy control rules. The adaptive mechanism in this method is based on a performance index, an error from a pre-specified desired trajectory, which modifies the consequent of previously online-generated fuzzy control rules. There is a need to account for the transport delay incurred by the time taken by the bloodstream to transport the drug from the infusion site to its action sites, which is about one minute. There is also a need to give the time for the patient measurement to start to respond to the altered drug infusion rate. For most anaesthetic drug infusions, this means that the generated fuzzy control rules are stored for two minutes before they are modified and incorporated into the fuzzy control rule base. We have clinically demonstrated the effectiveness of this intelligent control technique with muscle relaxation during surgery, starting each session with a completely blank rule base and having the controller learn the rules required by each patient online. We are now investigating the application of this technique to a more generally realistic clinical scenario where multiple drug infusions are managed based on multiple patient measurements: haemodynamic support of intensive care patients in septic shock. This application highlights problems with dimensionality as the number of input and output variables increases and thus challenges its general application to complex processes