Introduction
A major limitation of current implantable devices is that they do not provide feedback or inform how a host's behavior affects his or her health. In many disorders, such as epilepsy, a system that could teach patients to reduce seizure risk while teaching devices which electrophysiologic patterns are associated with ill health or increased seizure likelihood would greatly improve quality of life.
Methods
We present early progress on the main components of a sentient, implantable, epilepsy management system. We identify a cohort of 12 patients implanted with stereotactic electrodes to localize epileptic seizures. A Hidden Markov Model (HMM) is used to parse intracranial EEG (iEEG) recordings into discrete states based on covariance across channels. An Ecological Momentary Assessment (EMA) delivered via bi-directional text messaging simultaneously samples patient behavior. We develop a BERT-like NLP algorithm to interpret patient texts chronicling activities of daily living and feelings. We are exploring classifiers and learning algorithms to link these streams, so that patient and device can teach each other to optimize health.
Results
We are testing each of the above system components separately for reproducibility, ease of use and computational efficiency in preparation for an inpatient trial of the closed-loop system. Preliminary results suggest such that a trial is feasible, and a finite number of patient-specific states related to behavior and seizure risk can be identified.
Conclusion
We are developing a closed loop system that allows implanted neurodevices and patients to communicate, identify and modulate behavioral and neural state to improve seizure control and quality of life. We plan to initiate a clinical trial to test a first iteration of this system over the next 6 months.
Click Here for Poster