Reduction of classification error rates and monitoring system using an artificial class
View Patent ↗Systems and methods for enhancing the accuracy of classifying a measurement by providing an artificial class. Seizure prediction systems may employ a classification system including an artificial class and a user interface for signaling uncertainty in classification when a measurement is classified in the artificial class.
1. A method of classification of a subject's neurological condition, the method comprising:
identifying at least one class associated with an identified neurological condition; introducing an other class, the other class not being associated with an identified neurological condition;
classifying a feature vector from a subject into an identified class or the other class; and
generating a signal in response to classifying the feature vector.
2. The method as recited in claim 1 , wherein identifying at least one class comprises processing feature vectors from a training data set to associate each processed feature vector with an identified class.
3. The method as recited in claim 1 , wherein identifying at least one class comprises processing feature vectors from a training data set using unsupervised learning to identify at least one class.
4. The method as recited in claim 2 , wherein introducing the other class comprises:
generating artificial feature vectors; and
associating the generated artificial feature vectors with the other class.
5. The method as recited in claim 4 , wherein generating artificial feature vectors comprises randomly generating artificial feature vectors.
6. The method as recited in claim 5 , wherein randomly generating artificial feature vectors comprises randomly generating artificial feature vectors according to a specified probability distribution.
7. The method as recited in claim 6 , wherein the specified probability distribution has a mean and standard deviation substantially equal to the mean and standard deviation of the population of the feature vectors associated with the identified classes.
8. The method as recited in claim 1 , wherein the other class is characterized by a probability density function.
9. The method as recited in claim 8 , wherein identifying at least one class associated with an identified neurological condition comprises associating a probability density function with a class associated with an identified neurological condition.
10. The method as recited in claim 8 , wherein identifying at least one class comprises processing feature vectors from a training data set to associate each processed feature vector with an identified class.
11. The method as recited in claim 8 , wherein the probability density function represents a radial basis function.
12. The method as recited in claim 11 , wherein the radial basis function represents a normal distribution.
13. The method as recited in claim 11 , wherein the radial basis function represents a multiquadratic.
14. The method as recited in claim 11 , wherein the radial basis function represents a thin plate spline.
15. The method as recited in claim 8 , wherein the probability density function represents a uniform distribution.
16. The method as recited in claim 1 , wherein the feature vector from a subject comprises at least one feature extracted from brain activity of the subject.
17. The method as recited in claim 1 , wherein generating a signal in response to classifying a feature vector comprises providing an output to the subject indicative of a classification into the other class in response to a classification into the other class.
18. The method as recited in claim 1 , wherein the at least one class comprises a class associated with a neurological condition selected from the group consisting of a pre-ictal condition and a pro-ictal condition.
19. The method as recited in claim 1 , wherein the at least one class comprises a class associated with a contra-ictal condition.
20. The method as recited in claim 1 , wherein the at least one class comprises a class associated with an inter-ictal condition.
21. The method as recited in claim 18 , wherein the at least one class further comprises a second class associated with a contra-ictal condition.