DEVELOPMENT OF FULLY-AUTOMATED CLASSIFIER BUILDERS FOR NEURODIAGNOSTIC APPLICATIONS

• US 20110038515A1
• Filed: 08/14/2009
• Published: 02/17/2011
• Est. Priority Date: 08/14/2009
• Status: Active Grant

• Current Assignee: Brainscope SPV LLC
• Original Assignee: Brainscope Company Inc.
• Inventors: Jacquin, Arnaud, Vodencarevic, Asmir
• Granted Patent: US 10,321,840 B2
• Time in Patent Office: Days
• US Class Current: 382/128
• CPC Class Codes: A61B 5/16 Devices for psychotechnics using teaching or educational appliances G09B1/00 - G09B7/00; Testing reaction times ; Devices for evaluating the psychological state
• CPC Class Codes: A61B 5/369 Electroencephalography [EEG] devices for psychotechnics A61B5/16
• CPC Class Codes: A61B 5/7207 of noise induced by motion artifacts
• CPC Class Codes: A61B 5/726 using Wavelet transforms
• CPC Class Codes: A61B 5/7264 Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems neural networks per se G06N3/00; expert systems per se G06N5/00
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: G06F 18/2111 by using evolutionary computational techniques, e.g. genetic algorithms
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems

Abstract

Methods for constructing classifiers for binary classification of quantitative brain electrical activity data is described. The classifier building methods are based on the application of one or more evolutionary algorithms. In one embodiment, the evolutionary algorithm used is a genetic algorithm. In another embodiment, the evolutionary algorithm used is a modified Random Mutation Hill Climbing algorithm. In yet another embodiment, a combination of a genetic algorithm and a modified Random Mutation Hill Climbing algorithm is used for building a classifier. The classifier building methods are fully automated, and are adapted to generate classifiers (for example, Linear Discriminant Functions) with high sensitivity, specificity and classification accuracy.

Claims

1. A fully-automated method of building a binary classifier for classification of brain function based on electrical brain signals, comprising the steps of:

acquiring quantitative signal features from a reference database of brain electrical activity data;

organizing the quantitative features into hierarchical classes based on one or more quantitative measures indicative of the performance of the features;

selecting at least one set of features at random from the highest class in the hierarchical organization;

encoding the set of features into at least one bit string; and

applying one or more evolutionary algorithms to the at least one bit string to obtain a binary classifier with near-optimal performance.

2. The method of claim 1, wherein the reference database comprises of quantitative features derived from electrical brain activity data recorded from a plurality of individuals in the presence or absence of brain abnormalities.

3. The method of claim 2, wherein each individuals in the reference database is labeled as belonging to one of two diagnostic categories based on a priori information about the individual.

4. The method of claim 1, wherein the quantitative measure used for the hierarchical organization is a measure of the replicability of the features.

5. The method of claim 1, wherein the quantitative measure used for the hierarchical organization is a measure of the discriminative power of the features.

6. The method of claim 1, wherein the evolutionary algorithm applied is a genetic algorithm.

7. The method of claim 1, wherein the evolutionary algorithm applied is a Random Mutation Hill Climbing algorithm.

8. The method of claim 1, wherein the evolutionary algorithm applied is a Modified Random Mutation Hill Climbing algorithm.

9. The method of claim 1, wherein the evolutionary algorithm applied is a combination of a genetic algorithm and a Modified Random Mutation Hill Climbing algorithm.

10. The method of claim 1, wherein the binary classifier is a Linear Discriminant Function.

11. The method of claim 1, wherein the quantitative signal features are derived from electrical brain activity data using diffusion geometric analysis.

12. The method of claim 1, wherein the quantitative signal features are derived from electrical brain activity data using wavelet packet transformation.

13. The method of claim 1, wherein the quantitative signal features are derived from electrical brain activity data using Fast Fourier Transformation.

14. The method of claim 1, wherein an objective function is used to evaluate the performance measure of the at least one bit string.

15. The method of claim 14, wherein the objective function used is Area Under the Receiver Operating Curve of a binary classifier.

16. The method of claim 14, wherein the objective function is modified using a cost function based on the number of features in a classifier.

17. A method of building a Linear Discriminant Function for classification of electrical brain signals, comprising the steps of:

selecting multiple sets of quantitative features from a larger available pool of features derived from electrical brain activity data;

forming an initial population of chromosomes using the selected sets of features; and

applying genetic algorithm operators to the population of chromosomes.

18. The method of claim 17, wherein an individual chromosome in the population of chromosomes is encoded as a binary bit string.

19. The method of claim 17, wherein an objective function is used to evaluate the performance measure of an individual chromosome in the population of chromosomes.

20. The method of claim 19, wherein the objective function used is Area Under the Receiver Operating Curve of a Linear Discriminant Function.

21. The method of claim 17, wherein the population of chromosome evolves through multiple generations.

22. The method of claim 21, wherein the population of chromosome evolves through application of a selection operator.

23. The method of claim 21, wherein the population of chromosome evolves through application of a crossover operator.

24. The method of claim 21, wherein the population of chromosome evolves through application of a mutation operator.

25. The method of claim 17, wherein a Modified Random Mutation Hill Climbing algorithm is applied to a high performing chromosome obtained through the application of the genetic algorithm operators.

26. A method of building a Linear Discriminant Function for classification of electrical brain signals, comprising the steps of:

selecting a set of quantitative features from a larger available pool of features derived from electrical brain activity data;

encoding a chromosome as a binary bit string using the selected set of features;

inverting the value of at least one bit at a random location on the bit string to generate a new bit string; and

computing an objective function value of the new bit string.

27. The method of claim 26, wherein the objective function is used to evaluate the performance measure of a binary bit string.

28. The method of claim 27, wherein the objective function used is Area Under the Receiver Operating Curve of a Linear Discriminant Function.

29. The method of claim 26, wherein the values of at least two bits are changed at two random locations on the bit string.

30. The method of claim 29, wherein the value of a first bit is changed from 1 to 0 and the value of a second bit is changed from 0 to 1.

31. The method of claim 26, wherein the pool of features derived from electrical brain activity data comprises non-linear features.