Detection of patient seizures for wearable devices

Patent metadata

Field	Value
Publication number	US-2024366142-A1
Application number	US-202218289223-A
Country	US
Kind code	A1
Filing date	Apr 29, 2022
Priority date	May 1, 2021
Publication date	Nov 7, 2024
Grant date	—

Abstract

Systems and method for seizure detection. A seizure detection device utilizes frequency discrimination, time series features, and machine learning clustering algorithms to distinguish between the EEG signal of those experiencing a seizure compared to those who are not experiencing a seizure.

First claim

1 . A seizure detection device comprising:
sensing circuitry comprising at least one electrode configured to generate electroencephalogram (EEG) signal data; and processing circuitry configured to:
receive the EEG signal data according to a collection window epoch,
generate a power spectral density estimation of the EEG signal data,
determine whether a shift in power spectrum has occurred for a selected band based on the power spectral density estimation,
when the shift in power spectrum has occurred for the selected band:
apply a machine-learning clustering to the EEG signal data to generate a clustered data set based on at least a plurality of time-series features and a plurality of machine learning-defined features,
apply a cluster differentiator to the clustered data set to separate the EEG signal data,
indicate a seizure state when the EEG signal data is greater than a differentiator threshold, and
indicate a non-seizure state when the EEG signal data is less than the differentiator threshold.

2 . The seizure detection device of claim 1 , wherein the collection window epoch is 1 second.

3 . The seizure detection device of claim 1 , wherein the EEG signal data is sampled at a rate lower than 1000 Hz.

4 . The seizure detection device of claim 1 , wherein the EEG signal data is sampled at a rate of 256 Hz.

5 . The seizure detection device of claim 1 , wherein the selected band is 0.1 to 0.3 pi radians per sample and the shift is a 10% upshift in dB.

6 . The seizure detection device of claim 1 , wherein the power spectral density estimation includes an n-1 filter to approximate the first derivative of a time domain signal of the EEG signal data.

7 . The seizure detection device of claim 1 , wherein the cluster differentiator is a hyperplane.

8 . The seizure detection device of claim 1 , wherein the processing circuitry is further configured to:
when the seizure state is indicated, increment a seizure state counter, compare the seizure state counter against a seizure sensitivity threshold, and indicate a seizure present state when the seizure state counter is greater than the seizure sensitivity threshold, wherein the seizure detection device further comprises alerting circuitry configured to alert a user including when the processing circuitry indicates the seizure present state.

9 . The seizure detection device of claim 1 , wherein the processing circuitry is further configured to:
when the non-seizure state is indicated, increment a non-seizure state counter, compare the non-seizure state counter against a non-seizure sensitivity threshold, and indicate a non-seizure present state when the non-seizure state counter is greater than the non-seizure sensitivity threshold, wherein the seizure detection device further comprises alerting circuitry configured to alert a user including when the processing circuitry indicates the non-seizure present state.

10 . The seizure detection device of claim 1 , wherein the seizure detection device is wearable by a patient, and wherein the at least one electrode is a wearable sensor.

11 . The seizure detection device of claim 10 , wherein the at least one electrode is operably coupleable to skin of the patient by an adhesive element, and wherein the seizure detection device further comprises a wearable band to further secure the at least one electrode to the skin of the patient.

12 . A method of detecting a seizure with a seizure detection device, the seizure detection device including sensing circuitry comprising at least one electrode configured to generate electroencephalogram (EEG) signal data, the method comprising:
receiving the EEG signal data according to a collection window epoch; generating a power spectral density estimation of the EEG signal data; determining whether a shift in power spectrum has occurred for a selected band based on the power spectral density estimation; when the shift in power spectrum has occurred for the selected band:
applying a machine-learning clustering to the EEG signal data to generate a clustered data set based on at least a plurality of time-series features and a plurality of machine learning-defined features,
applying a cluster differentiator to the clustered data set to separate the EEG signal data,
indicating a seizure state when the EEG signal data is greater than a differentiator threshold, and
indicating a non-seizure state when the EEG signal data is less than the differentiator threshold.

13 . The method of claim 12 , wherein the collection window epoch is 1 second.

14 . The method of claim 12 , wherein the selected band is 0.1 to 0.3 pi radians per sample and the shift is a 10% upshift in dB.

15 . The method of claim 12 , wherein the power spectral density estimation includes an n-1 filter to approximate the first derivative of a time domain signal of the EEG signal data.

16 . (canceled)

17 . (canceled)

18 . The method of claim 12 , wherein the cluster differentiator is a hyperplane.

19 . The method of claim 12 , further comprising:
when the seizure state is indicated, incrementing a seizure state counter, comparing the seizure state counter against a seizure sensitivity threshold, and indicating a seizure present state when the seizure state counter is greater than the seizure sensitivity threshold, alerting a user including when the seizure present state is indicated.

20 . The method of claim 12 , further comprising:
when the non-seizure state is indicated, incrementing a non-seizure state counter, comparing the non-seizure state counter against a non-seizure sensitivity threshold, and indicating a non-seizure present state when the non-seizure state counter is greater than the non-seizure sensitivity threshold, alerting a user including when the non-seizure present state is indicated.

21 . The method of claim 12 , wherein the seizure detection device is wearable by a patient, and wherein the at least one electrode is a wearable sensor.

22 . The method of claim 21 , wherein the at least one electrode is operably coupleable to skin of the patient by an adhesive element, and wherein the method further comprises using a wearable band to further secure the at least one electrode to the skin of the patient.

23 - 28 . (canceled)

Assignees

• Medtronic Inc

Inventors

• Shrivastav Maneesh
• Ithapu Pavankumar

Classifications

• A61B5/746

  Alarms related to a physiological condition, e.g. details of setting alarm thresholds or avoiding false alarms · CPC title

• A61B5/6833

  Adhesive patches · CPC title

• A61B5/6831

  Straps, bands or harnesses · CPC title

• A61B5/1117

  Fall detection · CPC title

• A61B5/374

  Detecting the frequency distribution of signals, e.g. detecting delta, theta, alpha, beta or gamma waves · CPC title