Systems and methods for seizure detection and closed-loop neurostimulation

An implantable neurostimulator device, comprising:

• a memory;
• and
• a processor operatively coupled to the memory, the processor configured to be electrically coupled to a plurality of electrodes implanted in a patient, the processor configured to:
• receive brain activity data from the plurality of electrodes;
• detect a precursor activity leading to a seizure, an onset of the seizure, or a presence of the seizure based on the brain activity data;
• identify a pattern in the brain activity data based on mutual information between one or more pairs of electrodes from the plurality of electrodes, the mutual information determined, at least in part, by:
• computing covariances between signals from the one or more pairs of electrodes from the plurality of electrodes,
• generating a first value based on the covariances,
• normalizing the first value to generate a second value rsq, and
• calculating a logarithm of (1−rsq);
• determine a timing with which to deliver current pulses to a brain of the patient to interfere with at least one oscillation in brain activity contributing to the seizure based on the pattern in the brain activity data;
• and
• activate a subset of electrodes from the plurality of electrodes to deliver the current pulses to a target region of the brain of the patient based on the timing.

A method, comprising:

• measuring brain activity data associated with a brain of a patient using a plurality of electrodes implanted in the patient;
• detecting a precursor activity leading to a seizure, an onset of the seizure, or a presence of the seizure based on the brain activity data;
• identifying a pattern in the brain activity data based on mutual information associated with the plurality of electrodes, the mutual information determined, at least in part, by:
• computing covariances between signals associated with the plurality of electrodes,
• generating a first value based on the covariances,
• normalizing the first value to generate a second value rsq, and
• calculating a logarithm of (1−rsq);
• determining, based on the pattern in the brain activity data, a timing with which to deliver electrical stimulation to the brain of the patient to disrupt oscillations in brain activity contributing to the seizure;
• and
• in response to detecting the precursor activity leading to a seizure, the onset of the seizure, or the presence of the seizure, causing delivery of electrical stimulation to the brain of the patient via at least a subset of electrodes from the plurality of electrodes and according to the timing.

A method, comprising:

• calculating one or more features based on brain activity recorded from each electrode of a plurality of electrodes implanted in a patient, the one or more features including mutual information between one or more pairs of electrodes from the plurality of electrodes, the mutual information identified, at least in part, by:
• computing covariances between signals associated with the one or more pairs of electrodes,
• generating a first value based on the covariances,
• normalizing the first value to generate a second value rsq, and
• calculating a logarithm of (1−rsq);
• inputting at least one of (1) the one or more features, or (2) at least a portion of the recorded brain activity into a model trained using brain activity data from at least one of the patient or at least one other patient,
• determining a range of effective prediction probability threshold values and selecting a threshold value from the range of effective prediction probability threshold values;
• comparing an output of the model to the threshold value;
• in response to determining the output crosses the threshold value, determine a timing and a subset of electrodes with which to deliver electrical pulses;
• and
• causing delivery of the electrical pulses to a brain of the patient via at least the subset of electrodes and according to the timing.