An adaptive brain-machine interface algorithm for control of burst suppression in medical coma

• Published in: 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society Vol. 2014; pp. 1638 - 1641
• Main Authors: Yuxiao Yang, Shanechi, Maryam M.
• Format: Conference Proceeding Journal Article
• Language: English
• Published: United States IEEE 01.01.2014
• Subjects: Action Potentials; Adaptation models; Algorithms; Bayes Theorem; Brain models; Brain-Computer Interfaces; Coma - physiopathology; Computer Simulation; Drugs; Electroencephalography; Humans; Real-time systems; Steady-state
• ISSN: 1094-687X; 1557-170X
• DOI: 10.1109/EMBC.2014.6943919

Burst suppression is an electroencephalogram (EEG) indicator of profound brain inactivation in which bursts of electrical activity alternate with periods of isoelectricity termed suppression. Specified time-varying levels of burst suppression are targeted in medical coma, a drug-induced brain state used for example to treat uncontrollable seizures. A brain-machine interface (BMI) that observes the EEG could automate the control of drug infusion rate to track a desired target burst suppression trajectory. Such a BMI needs to use models of drug dynamics and burst suppression observations, whose parameters could change with the burst suppression level and the environment over time. Currently, these parameters are fit prior to real-time control, requiring a separate system identification session. Moreover, this approach cannot track parameter variations over time. In addition, small variations in drug infusion rate may be desired at steady state. Here we develop a novel adaptive algorithm for robust control of medical coma in face of unknown and time-varying system parameters. We design an adaptive recursive Bayesian estimator to jointly estimate drug concentrations and system parameters in real time. We construct a controller using the linear-quadratic-regulator strategy that explicitly penalizes large infusion rate variations at steady state and uses the estimates as feedback to generate robust control. Using simulations, we show that the adaptive algorithm achieves precise control of time-varying target levels of burst suppression even when model parameters are initialized randomly, and reduces the infusion rate variation at steady state.