NMM-Based Patient-Specific Temporally-Adaptive Stimulation Optimization for Seizure Control

We present a framework for the development of closed-loop neurostimulators that can deliver patient-optimized and temporally-adaptive stimulation pulses for epileptic seizure control. While many studies have targeted patient-specific seizure detection, little attention has been given to tailoring stimulation parameters for individual patients. This work prioritizes patient-specific stimulation optimization, assuming the seizure has already been detected. The proposed method employs a model predictive controller (MPC) that can rapidly converge to an optimal set of stimulation parameters for a specific patient thanks to being pre-trained with a neural mass model (NMM) that is fine-tuned to the patient's pre-recorded data. The development and patientspecific customization of the NMM are presented, showing its ability to generate synthetic intracranial electroencephalography (iEEG) with consistently-high spectral correlation to real prerecorded iEEG, for both normal and seizure periods (with a statistically-significant number of samples). We also showcase how these patient-customized NMMs are used along with the MPC to optimize the stimulation parameters for seizure control. Examples from both sub-optimal and optimal stimulation therapies are presented, as well as how the parameter space is navigated efficiently by leveraging historical data, leading to convergence towards an optimal point with minimal iterations.