Classification of epileptic and non-epileptic phenotypes from EEG recordings and related closed-loop applications
A detection function is used to detect an event in an electroencephalogram (EEG) signal in order to generate a plurality of detected events. For each detected event in the plurality of detected events, a plurality of features from the EEG signal is measured in order to obtain a measured feature vector, where the plurality of features is defined by a feature space corresponding to the event and a measured feature space includes a plurality of measured feature vectors corresponding to a plurality of labeled points having locations. The EEG signal is classified based at least in part on the locations of the plurality of labeled points in the measured feature space.
1 . A closed-loop feedback system, comprising:
a brainstate estimator that generates a brainstate estimate of a subject's brain, including by:
detecting a peak shape in an electroencephalogram (EEG) signal in order to generate a plurality of detected peaks in the EEG signal;
obtaining a plurality of to-be-measured features that correspond to the peak shape, wherein the plurality of to-be-measured features includes an amplitude feature, a prominence feature, and a duration feature;
measuring the plurality of to-be-measured features from the plurality of detected peaks in the EEG signal in order to obtain a plurality of measured feature vectors, including by:
for a given detected peak in the plurality of detected peaks, analyzing a portion of the EEG signal that includes said given detected peak and excluding a rest of the EEG signal outside of said portion of the EEG signal; and
adding, to the plurality of measured feature vectors, a measured amplitude for said given detected peak, a measured prominence for said given detected peak, and a measured duration for said given detected peak, wherein a multidimensional space includes the plurality of measured feature vectors as a plurality of labeled points having locations within the multidimensional space;
classifying the EEG signal based at least in part on the locations of the plurality of labeled points in the multidimensional space, including by:
obtaining, for the multidimensional space, a plurality of totally disjoint regions that the multidimensional space is divided up into and a plurality of probabilities associated with the plurality of totally disjoint regions, wherein the multidimensional space includes a first axis that corresponds to the amplitude feature, a second axis that corresponds to the prominence feature, and a third axis that corresponds to the duration feature;
determining a combined value based at least in part on the locations of the plurality of labeled points within the multidimensional space, the plurality of totally disjoint regions, and the plurality of probabilities; and
classifying the EEG signal based at least in part on the combined value; and
generating, based at least in part on the classification of the EEG signal, the brainstate estimate;
outputting the brainstate estimate of the subject's brain;
a controller that generates a control signal based at least in part on the brainstate estimate of the subject's brain; and
a stimulus generator that generates a stimulus for the subject's brain based at least in part on the control signal.
2 . The system recited in claim 1 , wherein
the brainstate estimate includes a vector of state probabilities.
3 . The system recited in claim 2 , further including instructions for selecting a second EEG signal to supplement the EEG signal.
4 . The system recited in claim 2 , further including instructions for selecting a second EEG signal to supplement the EEG signal, including by using metadata with timing information, wherein the metadata includes one or more of the following: information associated with a treatment aftereffect or information associated with the subject moving.
5 . The system recited in claim 1 , wherein classifying the EEG signal includes classifying the EEG signal as epileptic or not epileptic.
6 . The system recited in claim 1 , wherein:
those regions in the plurality of totally disjoint regions that have an associated probability define an acceptance region;
those regions in the plurality of totally disjoint regions that do not have an associated probability are outside of the acceptance region; and
determining the combined value includes ignoring those labeled points in the plurality of labeled points with locations that fall outside of the acceptance region.
7 . The system recited in claim 1 , wherein:
the EEG signal is associated with measuring a response of a treatment to the subject of the closed-loop treatment system;
the brainstate estimator further selects a supplemental EEG signal in an event the EEG signal is too short;
the supplemental EEG signal is selected from one or more potential recordings associated with the subject and any potential recording that is associated with a subject other than the subject is ineligible to be the supplemental EEG signal; and
selecting the supplemental EEG signal is based at least in part on a comparison of an aftereffect duration, during which treatment by the closed-loop treatment system lingers in the subject, and time information associated with the one or more potential recordings associated with the subject.
8 . The system recited in claim 1 , wherein:
the EEG signal is associated with measuring a response of a treatment to the subject of the closed-loop treatment system;
the brainstate estimator further selects a supplemental EEG signal in an event the EEG signal is too short;
the supplemental EEG signal is selected from one or more potential recordings associated with the subject and any potential recording that is associated with a subject other than the subject is ineligible to be the supplemental EEG signal; and
selecting the supplemental EEG signal is based at least in part on a comparison of a period of movement, during which the subject was moving, and time information associated with the one or more potential recordings associated with the subject.
9 . A method for operating a closed-loop feedback system, comprising:
generating a brainstate estimate of a subject's brain using a brainstate estimator in the closed-loop feedback system, including by:
detecting a peak shape in an electroencephalogram (EEG) signal in order to generate a plurality of detected peaks in the EEG signal;
obtaining a plurality of to-be-measured features that correspond to the peak shape, wherein the plurality of to-be-measured features includes an amplitude feature, a prominence feature, and a duration feature;
measuring the plurality of to-be-measured features from the plurality of detected peaks in the EEG signal in order to obtain a plurality of measured feature vectors, including by:
for a given detected peak in the plurality of detected peaks, analyzing a portion of the EEG signal that includes said given detected peak and excluding a rest of the EEG signal outside of said portion of the EEG signal; and
adding, to the plurality of measured feature vectors, a measured amplitude for said given detected peak, a measured prominence for said given detected peak, and a measured duration for said given detected peak, wherein a multidimensional space includes the plurality of measured feature vectors as a plurality of labeled points having locations within the multidimensional space;
classifying the EEG signal based at least in part on the locations of the plurality of labeled points in the multidimensional space, including by:
obtaining, for the multidimensional space, a plurality of totally disjoint regions that the multidimensional space is divided up into and a plurality of probabilities associated with the plurality of totally disjoint regions, wherein the multidimensional space includes a first axis that corresponds to the amplitude feature, a second axis that corresponds to the prominence feature, and a third axis that corresponds to the duration feature;
determining a combined value based at least in part on the locations of the plurality of labeled points within the multidimensional space, the plurality of totally disjoint regions, and the plurality of probabilities; and
classifying the EEG signal based at least in part on the combined value;
generating, based at least in part on the classification of the EEG signal, the brainstate estimate;
outputting the brainstate estimate of the subject's brain;
generating a control signal, using a controller in the closed-loop feedback system, based at least in part on the brainstate estimate of the subject's brain; and
generating a stimulus for the subject's brain, using a stimulus generator in the closed-loop feedback system, based at least in part on the control signal.
10 . The method recited in claim 9 , wherein
the brainstate estimate includes a vector of state probabilities; and
the method is performed by a device associated with the closed-loop treatment system.
11 . The method recited in claim 10 , wherein the device and the controller are included in a same edge device.
12 . The method recited in claim 10 , wherein the controller is included in an edge device and the device is included in a different device.
13 . The method recited in claim 10 , further including selecting a second EEG signal to supplement the EEG signal.
14 . The method recited in claim 10 , further including selecting a second EEG signal to supplement the EEG signal, including by using metadata with timing information, wherein the metadata includes one or more of the following: information associated with a treatment aftereffect or information associated with the subject moving.
15 . The method recited in claim 9 , wherein classifying the EEG signal includes classifying the EEG signal as epileptic or not epileptic.
16 . The method recited in claim 9 , wherein:
those regions in the plurality of totally disjoint regions that have an associated probability define an acceptance region;
those regions in the plurality of totally disjoint regions that do not have an associated probability are outside of the acceptance region; and
determining the combined value includes ignoring those labeled points in the plurality of labeled points with locations that fall outside of the acceptance region.
17 . The method recited in claim 9 , wherein:
the EEG signal is associated with measuring a response of a treatment to the subject off the closed-loop treatment system;
the method further includes selecting a supplemental EEG signal in an event the EEG signal is too short;
the supplemental EEG signal is selected from one or more potential recordings associated with the subject and any potential recording that is associated with a subject other than the subject is ineligible to be the supplemental EEG signal; and
selecting the supplemental EEG signal is based at least in part on a comparison of an aftereffect duration, during which treatment by the closed-loop treatment system lingers in the subject, and time information associated with the one or more potential recordings associated with the subject.
18 . The method recited in claim 9 , wherein:
the EEG signal is associated with measuring a response of a treatment to the subject of the closed-loop treatment system;
the method further includes selecting a supplemental EEG signal in an event the EEG signal is too short;
the supplemental EEG signal is selected from one or more potential recordings associated with the subject and any potential recording that is associated with a subject other than the subject is ineligible to be the supplemental EEG signal; and
selecting the supplemental EEG signal is based at least in part on a comparison of a period of movement, during which the subject was moving, and time information associated with the one or more potential recordings associated with the subject.