Systems and methods for seizure detection with a statistical analysis and an artificial intelligence analysis
A seizure detection system including one or more circuits configured to receive an electroencephalogram (EEG) signal generated based on electrical brain activity of a patient. The one or more circuits are configured to identify candidate seizures with a statistical analysis that identifies the candidate seizures based on changes in non-linear features of the EEG signal, determine to switch from identifying the candidate seizures with the statistical analysis to an artificial intelligence model, and switch from identifying the candidate seizures with the statistical analysis to identifying the candidate seizures based on the artificial intelligence model with the EEG signal.
1 . A seizure detection system comprising one or more circuits, wherein the one or more circuits are configured to:
receive an electroencephalogram signal generated based on electrical brain activity of a patient;
identify candidate seizures with a statistical analysis from the electroencephalogram signal;
cause a display device to display the candidate seizures identified by the statistical analysis;
receive, via the display device, indications of sections of the electroencephalogram signal identified by a user as being associated with the displayed candidate seizures;
train, while the display device displays the candidate seizures identified by the statistical analysis, using the sections identified by the user as being associated with the displayed candidate seizures, an artificial intelligence model to classify the electroencephalogram signal as a candidate seizure or a non-seizure;
identify candidate seizures from the electroencephalogram signal with the artificial intelligence model;
determine to switch from displaying the candidate seizures identified with the statistical analysis to displaying the candidate seizures identified with the artificial intelligence model; and
switch from displaying the candidate seizures identified with the statistical analysis to displaying the candidate seizures identified with the artificial intelligence model.
2 . The seizure detection system of claim 1 , wherein the one or more circuits are configured to:
identify the candidate seizures with the statistical analysis based on changes in non-linear features of the electroencephalogram signal.
3 . The seizure detection system of claim 1 , wherein the one or more circuits are configured to:
train the artificial intelligence model to classify the candidate seizures based on training data comprising the electroencephalogram signal.
4 . The seizure detection system of claim 1 , wherein the determination to switch from displaying the candidate seizures identified with the statistical analysis to displaying the candidate seizures identified with the artificial intelligence model is made responsive to a determination that a performance of the artificial intelligence model to classify the candidate seizures satisfies a threshold.
5 . The seizure detection system of claim 1 , wherein the one or more circuits are configured to:
determine a first performance level of the statistical analysis; and
determine a second performance level of the artificial intelligence model;
wherein the determination to switch from displaying the candidate seizures identified with the statistical analysis to displaying the candidate seizures identified with the artificial intelligence model is made in response to determining that the second performance level is greater than the first performance level.
6 . The seizure detection system of claim 1 , wherein the determination to switch from displaying the candidate seizures identified with the statistical analysis to displaying the candidate seizures identified with the artificial intelligence model is made by determining that at least a predefined amount of detections of the candidate seizures by the statistical analysis are also detected by the artificial intelligence model.
7 . The seizure detection system of claim 1 , wherein the one or more circuits are configured to identify the candidate seizures with the statistical analysis by:
determining a plurality of metrics based on the electroencephalogram signal, the plurality of metrics indicating non-linear features of the electroencephalogram signal; and
determining that the electroencephalogram signal indicates the candidate seizures by determining, based at least in part on the plurality of metrics, changes in the non-linear features of the electroencephalogram signal over time, wherein the changes in the non-linear features indicate physiological forces that give rise to the candidate seizures.
8 . The seizure detection system of claim 7 , wherein determining, based at least in part on the plurality of metrics, the changes in the non-linear features of the electroencephalogram signal comprises determining an increase in the non-linear features over time.
9 . The seizure detection system of claim 7 , wherein the plurality of metrics comprise at least one of dimensionality, synchrony, Lyapunov exponents, one or more forms of entropy, one or more forms of eigenvalues, global non-linearity, distance differences between recurrence trajectories, higher order spectra, loss of complexity, a surrogate test, or self-similarity.
10 . The seizure detection system of claim 7 , wherein the one or more circuits are configured to determine that the electroencephalogram signal indicates the candidate seizures by:
determining, based at least in part on the plurality of metrics, the changes in the non-linear features of the electroencephalogram signal over time by:
performing a preliminary analysis with one metric of the plurality of metrics, wherein the preliminary analysis indicates that the electroencephalogram signal indicates the candidate seizure or that the electroencephalogram signal includes noise; and
performing a secondary analysis with one or more metrics of the plurality of metrics other than the one metric to determine whether the electroencephalogram signal indicates the candidate seizure or that the electroencephalogram signal includes the noise.
11 . The seizure detection system of claim 1 , wherein the one or more circuits are configured to:
operate in a first operating phase where the one or more circuits identify the candidate seizures with the statistical analysis; and
operate in a second operating phase where the one or more circuits identify the candidate seizures with the artificial intelligence model.
12 . The seizure detection system of claim 11 , wherein the one or more circuits are configured to operate in the first operating phase by:
generating seizure alerts indicating that the electroencephalogram signal indicates the candidate seizures identified by the statistical analysis; and
causing the display device to display the seizure alerts.
13 . The seizure detection system of claim 1 wherein the one or more circuits are configured to operate in a first operating phase by:
training the artificial intelligence model based on training data while the candidate seizures are identified by the statistical analysis.
14 . The seizure detection system of claim 13 , wherein the one or more circuits are configured to operate in the first operating phase by training the artificial intelligence model based on the electroencephalogram signal and the candidate seizures identified by the statistical analysis.
15 . The seizure detection system of claim 13 , wherein the one or more circuits are configured to operate in the first operating phase by:
generating seizure alerts indicating that the electroencephalogram signal indicates the candidate seizures identified by the statistical analysis;
causing the display device to display the seizure alerts;
receiving labels of the candidate seizures and the electroencephalogram signal from the user via the display device; and
training the artificial intelligence model based on the labels and the electroencephalogram signal.
16 . A method, comprising:
receiving, by one or more processing circuits, an electroencephalogram signal generated based on electrical brain activity of a patient;
identifying, by the one or more processing circuits, candidate seizures with a statistical analysis from the electroencephalogram signal;
causing, by the one or more processing circuits, a display device to display the candidate seizures identified by the statistical analysis;
receiving, by the one or more processing circuits, via the display device, indications of sections of the electroencephalogram signal identified by a user as being associated with the displayed candidate seizures;
training, by the one or more processing circuits, while the display device displays the candidate seizures identified by the statistical analysis, using the sections identified by the user as being associated with the displayed candidate seizures, an artificial intelligence model to classify the electroencephalogram signal as a candidate seizure or a non-seizure;
identifying, by the one or more processing circuits, candidate seizures from the electroencephalogram signal with the artificial intelligence model;
determining, by the one or more processing circuits, to switch from displaying the candidate seizures identified with the statistical analysis to displaying the candidate seizures identified with the artificial intelligence model; and
switching, by the one or more processing circuits, from displaying the candidate seizures identified with the statistical analysis to displaying the candidate seizures identified with the artificial intelligence model.
17 . The method of claim 16 , comprising:
determining, by the one or more processing circuits, a first performance level of the statistical analysis; and
determining, by the one or more processing circuits, a second performance level of the artificial intelligence model;
wherein the determination to switch from displaying the candidate seizures identified with the statistical analysis to displaying the candidate seizures identified with the artificial intelligence model is made in response to determining that the second performance level is greater than the first performance level.
18 . The method of claim 16 , wherein the determination to switch from displaying the candidate seizures identified with the statistical analysis to displaying the candidate seizures identified the artificial intelligence model is made by determining that at least a predefined amount of detections of the candidate seizures by the statistical analysis are also detected by the artificial intelligence model.
19 . The method of claim 16 , comprising:
identifying, by the one or more processing circuits, the candidate seizures with the statistical analysis by:
determining a plurality of metrics based on the electroencephalogram signal, the plurality of metrics indicating non-linear features of the electroencephalogram signal; and
determining that the electroencephalogram signal indicates the candidate seizures by determining, based at least in part on the plurality of metrics, changes in the non-linear features of the electroencephalogram signal over time, wherein the changes in the non-linear features indicate physiological forces that give rise to the candidate seizures.
20 . One or more non-transitory storage medium, storing instructions thereon, that, when executed by one or more processors, cause the one or more processors to:
receive an electroencephalogram signal generated based on electrical brain activity of a patient;
identify candidate seizures with a statistical analysis from the electroencephalogram signal;
cause a display device to display the candidate seizures identified by the statistical analysis;
receive, via the display device, indications of sections of the electroencephalogram signal identified by a user as being associated with the displayed candidate seizures;
train, while the display device displays the candidate seizures identified by the statistical analysis, using the sections identified by the user as being associated with the displayed candidate seizures, an artificial intelligence model to classify the electroencephalogram signal as a candidate seizure or a non-seizure;
identify candidate seizures from the electroencephalogram signal with the artificial intelligence model;
determine to switch from displaying the candidate seizures identified with the statistical analysis to displaying the candidate seizures identified with the artificial intelligence model; and
switch from displaying the candidate seizures identified with the statistical analysis to displaying the candidate seizures identified with the artificial intelligence model.