Wearable system for real-time detection of epileptic seizures
A wearable system for epileptic seizure detection, comprising an eyeglasses frame, with a left arm and a right arm configured to rest over the ears of an intended person wearing the eyeglasses, a first pair of electrodes located in the left arm, and a second pair of electrodes located in the right arm, the first pair of electrodes and the second pair of electrodes arranged such to be in contact with the skull of the intended person wearing the eyeglasses, and an EEG signal acquiring system integral to the left and right arms, connected to measuring outputs of the respective first pair and second pair of electrodes.
1. A wearable system for real-time detection of epileptic seizure based on electroencephalography, comprising:
an eyeglass frame comprising first and second arms, each arm configured to rest over a respective ear of a user of the eyeglass frame;
first and second head electrodes supported by the first arm, and third and fourth head electrodes supported by the second arm;
wherein the first and second head electrodes are respectively located at arm positions corresponding to head positions T 3 and F 7 , and the third and fourth head electrodes are respectively located at arm positions corresponding to head positions T 4 and F 8 , so that the system only has these four head electrodes, and no additional head electrodes, for detecting epileptic seizures, the first, second, third and fourth head electrodes configured to be in contact with a head of the user for detecting epileptic seizures; and
a processor, comprising processing circuitry, connected to measuring outputs of the first, second, third, and fourth electrodes, and configured to:
detect seizures of the user based on signals from the first, second, third and fourth electrodes processed via a hierarchical system comprising a simple classifier and a full classifier, wherein the simple classifier is configured to consider a determined number (K 1 ) of available features from the electrodes and is computationally efficient, and the full classifier is configured to consider an entire set of the available features from the electrodes and is accurate but computationally complex, and
wherein the full classifier is configured to be invoked when the simple classifier cannot provide sufficient classification results based on a number of agreeing decision trees to reduce the computational complexity and extending the battery lifetime while providing a high classification accuracy for detecting seizures.
2. The wearable system of claim 1 , wherein the processor is configured to use a 24-bit resolution sampling mean configured to measure analog signals from the first, second, third, and fourth electrodes, and to control to emit a signal to a wireless communication receiver via an emitter.
3. The wearable system of claim 1 , wherein the processor is configured to perform feature extraction comprising:
a preprocessing step of discrete wavelet transform on acquired signals from the electrodes;
a first calculation of nonlinear features from the preprocessed acquired signals for different input parameters; and
a second calculation of power features on the acquired signals.
4. The wearable system of claim 1 , wherein the simple classifier and the complex classifier are trained via a random forest algorithm, and are random forest classifiers.
5. The wearable system of claim 4 , wherein the simple classifier and the complex classifier are configured to use the same number of trees within their respective forests.
6. The wearable system of claim 1 , wherein the processor is configured to compare values to determine whether to invoke the simple classifier or the full classifier.
7. A method of using a wearable system for real-time detection of epileptic seizure based on electroencephalography, the method comprising:
having an eyeglass frame comprising first and second arms, each arm configured to rest over a respective ear of a user of the eyeglass frame;
wherein the eyeglass frame comprises first and second head electrodes supported by the first arm, and third and fourth head electrodes supported by the second arm;
wherein the first and second head electrodes are respectively located at arm positions corresponding to head positions T 3 and F 7 , and the third and fourth head electrodes are respectively located at arm positions corresponding to head positions T 4 and F 8 , for detecting epileptic seizures, the first, second, third and fourth head electrodes configured to be in contact with a head of the user for detecting epileptic seizures, so that the system only has these four head electrodes for detecting epileptic seizures; and
measuring and using outputs of the first, second, third, and fourth electrodes to detect seizures via a hierarchical system comprising a simple classifier and a full classifier, wherein the simple classifier considers a determined number (K 1 ) of available features from the electrodes and is computationally efficient, and the full classifier considers an entire set of available features from the electrodes and is accurate but more computationally complex than is the simple classifier, and
comparing values to determine whether to invoke the simple classifier or the full classifier, comprising invoking the full classifier when the simple classifier cannot provide sufficient classification results based on a number of agreeing decision trees to reduce the computational complexity and extending the battery lifetime while providing a high classification accuracy for detecting seizures.
8. The method of claim 7 , further comprising using a 24-bit resolution sampling mean to measure analog signals from the first, second, third, and fourth electrodes, and emitting wireless signals to a wireless communication receiver via an emitter.
9. The method of claim 7 , further comprising training the simple classifier and the complex classifier via a random forest algorithm.
10. The method of claim 9 , wherein the simple classifier and the complex classifier use the same number of trees within their respective forests.
11. The method of claim 7 , further comprising performing feature extraction comprising:
a preprocessing step of discrete wavelet transform on acquired signals from the electrodes;
a first calculation of nonlinear features from the preprocessed acquired signals for different input parameters; and
a second calculation of power features on the acquired signals.