Methods and systems for optimizing therapy using stimulation mimicking natural seizures
Systems, methods, and devices for automatic generation of a stimulation therapy that mimics electrographic activity in the brain at natural seizure termination define a stimulation therapy to be generated by an implanted component of a medical device system and delivered to a subject through identifying data characterizing a patient's seizures, especially at termination. A machine learning model identifies the seizures or seizure types from which to establish a canonical seizure or seizure type, and an algorithm translates the canonical seizure or seizure type into data that can be used to characterize a stimulation therapy. The systems, methods, and devices, include those configured to deliver the stimulation therapy that emulates the canonical seizure or seizure type when the seizure is detected, with the aim of terminating the seizure sooner than it would terminate without intervention.
1. A method of neurostimulation, the method comprising:
detecting an instance of an electrographic seizure based on a patient's electrographic activity, wherein the electrographic seizure is characterized by an offset; and
responsive to a detection of an instance of the electrographic seizure, delivering a stimulation therapy to the patient, wherein the stimulation therapy emulates a canonical seizure offset constructed from a plurality of seizure offsets in electrographic seizures.
2. The method of claim 1 , wherein detecting an instance of an electrographic seizure based on a patient's electrographic activity comprises applying a machine learning model to the patient's electrographic activity.
3. The method of claim 2 , wherein the electrographic seizure is characterized by an onset, and the machine learning model is trained to detect the onset of the electrographic seizure.
4. The method of claim 2 , wherein:
the machine learning model is trained to identify a seizure type corresponding to the instance of the electrographic seizure; and
the stimulation therapy emulates the canonical seizure offset corresponding to the identified seizure type.
5. The method of claim 2 , wherein the machine learning model comprises one or more of a deep learning model, a clustering algorithm, a recurrent neural network, a hidden Markov model, a long short term memory neural network.
6. The method of claim 1 , further comprising:
creating the stimulation therapy that emulates the canonical seizure offset.
7. The method of claim 6 , wherein creating the stimulation therapy that emulates the canonical seizure offset comprises creating pulse characterization data based on the canonical seizure offset.
8. The method of claim 7 , wherein delivering a stimulation therapy to the patient comprises translating the pulse characterization data into a train of stimulation pulses.
9. The method of claim 1 , wherein constructing the canonical seizure offset comprises applying a machine learning model to the plurality of seizure offsets in the electrographic seizures, wherein the machine learning model is trained to generate the canonical seizure offset based on ingested seizure offsets.
10. An implantable neurostimulator system for delivering a stimulation therapy, the system comprising:
a plurality of electrodes;
a detection module coupled to the plurality of electrodes, the detection module configured to detect an instance of an electrographic seizure based on a patient's electrographic activity, wherein the electrographic seizure is characterized by an offset; and
a stimulation generator configured to deliver a stimulation therapy to the patient in response to a detection of an instance of the electrographic seizure, wherein the stimulation therapy emulates a canonical seizure offset constructed from a plurality of seizure offsets in electrographic seizures.
11. The implantable neurostimulator system of claim 10 , wherein the detection module detects an instance of an electrographic seizure based on a patient's electrographic activity by being further configured to apply a machine learning model to the patient's electrographic activity.
12. The implantable neurostimulator system of claim 11 , wherein the electrographic seizure is characterized by an onset, and the machine learning model is trained to detect the onset of the electrographic seizure.
13. The implantable neurostimulator system of claim 11 , wherein:
the machine learning model is trained to identify a seizure type corresponding to the instance of the electrographic seizure; and
the stimulation therapy emulates a canonical seizure offset corresponding to the identified seizure type.
14. The implantable neurostimulator system of claim 11 , wherein the machine learning model comprises one or more of a deep learning model, a clustering algorithm, a recurrent neural network, a hidden Markov model, a long short term memory neural network.
15. The implantable neurostimulator system of claim 10 , wherein the stimulation therapy that emulates the canonical seizure offset comprises pulse characterization data that is based on the canonical seizure offset.
16. The implantable neurostimulator system of claim 15 , wherein the stimulation generator is configured to translate the pulse characterization data into a train of stimulation pulses.
17. The implantable neurostimulator system of claim 10 , wherein:
the stimulation generator comprises a training module configured to construct the canonical seizure offset based on the plurality of seizure offsets in the electrographic seizures.
18. The implantable neurostimulator system of claim 17 , wherein the training module is configured to construct the canonical seizure offset by being further configured to apply a machine learning model to the plurality of electrographic seizures, wherein the machine learning model is trained to generate the canonical seizure offset based on ingested seizure offsets.
19. The implantable neurostimulator system of claim 18 , wherein the machine learning model comprises one or more of a deep learning model, a clustering algorithm, a recurrent neural network, a hidden Markov model, a long short term memory neural network.