Authors :
Presenting Author: Christina Sheckler, BS – University of Michigan
Kathryn Kish, PhD – University of Michigan
Maria Luisa Saggio, PhD – Aix Marseille Universite
Matthew Szuromi, BS – Boston University
William Stacey, MD, PhD – University of Michigan
Rationale:
Rationale: Epileptic seizures involve the brain transitioning from a resting state to an abnormal state of synchronized bursting, akin to a bifurcation in dynamical systems where a parameter shift triggers a sudden change in behavior. The advent of AI algorithms has made it increasingly important to generate large amounts of realistic seizure data, with a wide range of dynamical signatures, for training.
Methods:
Methods: We developed a comprehensive model that uses dynamical equations capable of simulating 16 “dynamotypes” that span the full range of theoretical seizure dynamics. This model contains methods to add realistic pink noise to both the acquisition data and underlying dynamics. It also adds the effect of electrode filtering. The result is an open-source method of generating a vast library of seizures that look like human seizures for training seizure detection/analysis algorithms.Results:
Results: We developed several tools with this model. First, we have a dynamical atlas of all 16 possible onset-offset bifurcation combinations, each characterized by distinct features in the simulated EEG-like recordings. The atlas demonstrates the effects of noise on the output. We developed a Matlab primer the explains the dynamical background and generates different seizures in real time. Finally, there is a GUI that allows direct user feedback to see how different parameters affect the dynamical trajectories and generate time series data. This user-friendly guide has been designed to be adapted by other labs for educational purposes as well as for generating diverse datasets of simulated seizure recordings that have a strong resemblance to human EEG data.
Conclusions:
Conclusions: This toolbox produces large numbers of diverse seizure patterns that have similar noise and filtering characteristics as human EEG, which can aid in training seizure detection algorithms, understanding brain dynamical behavior for clinicians and researchers, and exploring the impact of noise on EEG recordings.
Funding: Michigan Medicine Lucas Family Fund