Development of fully-automated classifier builders for neurodiagnostic applications

• US 10,321,840 B2
• Filed: 08/14/2009
• Issued: 06/18/2019
• Est. Priority Date: 08/14/2009
• Status: Active Grant

• Current Assignee: Brainscope SPV LLC
• Original Assignee: Brainscope Company Inc.
• Inventors: Jacquin, Arnaud, Vodencarevic, Asmir
• Primary Examiner(s): Weston, Tiffany
• Assistant Examiner(s): Tran, Tho Q
• Application Number:
  US12/541,272
• Publication Number:
  US 20110038515A1
• Time in Patent Office: 3,595 Days
• Field of Search: 600544, 600545
• 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 method of building a linear discriminant function for classification of brain electrical activity data into diagnostic categories, the method comprising:

providing a data storage device comprising a reference database of brain electrical activity data and a processor;

acquiring with the processor quantitative brain electrical activity features from the brain electrical activity data in the reference database;

calculating with the processor one or more quantitative measures for each of the acquired quantitative features, the one or more quantitative measures being indicative of classification performance of the acquired quantitative features;

selecting with the processor a reduced pool of brain electrical activity features from the acquired quantitative features based on the one or more quantitative measures associated with each of the acquired quantitative features, wherein calculation of the one or more quantitative measures and selection of the reduced pool of features is performed prior to the application of a genetic algorithm;

selecting with the processor multiple subsets of quantitative brain electrical activity features from the reduced pool of features;

forming with the processor an initial population of chromosomes, each chromosome of the initial population including one of the selected subsets of features and representing a linear discriminant function for classification of brain electrical activity data into diagnostic categories;

applying with the processor genetic algorithm operators to the initial population of chromosomes to derive a new population of chromosomes, wherein each individual chromosome in the new population of chromosomes includes a subset of the brain electrical activity features that is generated from the subsets of features of the initial population of chromosomes;

determining objective function values of the chromosomes in the new population;

repeating the applying step and the determining step until the objective function value of one of the chromosomes in the new population is above a predetermined threshold; and

storing a final linear discriminant function on non-volatile media for use in classification of brain electrical activity data into the diagnostic categories, wherein the stored final linear discriminant function is defined by the subset of brain electrical activity features of the chromosome having the objective function value above the predetermined threshold.

2. The method of claim 1, wherein an individual chromosome in the initial population of chromosomes is encoded as a binary bit string.

3. The method of claim 1, further comprising evaluating with the processor a performance measure of each individual chromosome in the initial population of chromosomes using an objective function.

4. The method of claim 3, wherein the objective function is Area Under the Receiver Operating Curve of a Linear Discriminant Function.

5. The method of claim 1, wherein the initial population of chromosomes evolves through multiple generations through application of the genetic algorithm operators.

6. The method of claim 1, wherein the genetic algorithm operators comprise a selection operator.

7. The method of claim 1, wherein the genetic algorithm operators comprise a crossover operator.

8. The method of claim 1, wherein the genetic algorithm operators comprise a mutation operator.

9. The method of claim 1, wherein a Modified Random Mutation Hill Climbing algorithm is applied to a high performing chromosome obtained through the application of the genetic algorithm operators.

10. The method of claim 1, wherein the final linear discriminant function is a Linear Discriminant Function for classification of brain electrical activity data into diagnostic categories that include at least one of:

(i) normal brain function, (ii) abnormal brain function, (iii) organic brain dysfunction, (iv) functional brain dysfunction, (v) focal brain dysfunction, (vi) diffuse brain dysfunction, (v) (closed-head) traumatic brain injury (TBI), and (vi) mild TBI (concussion).

11. A method of building a linear discriminant function for classification of brain electrical activity data into diagnostic categories, the method comprising:

providing a data storage device comprising a reference database of brain electrical activity data and a processor;

acquiring with the processor quantitative brain electrical activity features from the brain electrical activity data in the reference database;

calculating with the processor one or more quantitative measures for each of the acquired quantitative features, the one or more quantitative measures being indicative of classification performance of the acquired quantitative features;

selecting with the processor a reduced pool of brain electrical activity features from the acquired quantitative features based on the one or more quantitative measures associated with the acquired quantitative features, wherein calculation of the one or more quantitative measures and selection of the reduced pool of features is performed prior to the application of an evolutionary algorithm;

selecting with the processor a set of quantitative brain electrical activity features from the reduced pool of features;

encoding with the processor a chromosome as a binary bit string including a bit corresponding to each feature in the reduced pool of features, wherein the bits in the binary bit string corresponding to the selected set of features are set in an active state and the remaining bits in the binary bit string are set in an inactive state, wherein the binary bit string represents a linear discriminant function for classification of brain electrical activity data into diagnostic categories;

inverting with the processor at least one bit value at a random location on the binary bit string to generate a new binary bit string, wherein inverting the at least one bit value includes switching the at least one bit value from one of the active and inactive states to the other of the active and inactive states;

computing with the processor an objective function value of the new binary bit string;

repeating the steps of inverting the at least one bit value and computing the objective function value until a final binary bit string having an objective function value above a predetermined threshold is obtained; and

storing a final linear discriminant function on non-volatile media for use in classification of brain electrical activity data into the diagnostic categories, wherein the stored final linear discriminant function is defined by the bits in the final binary bit string that are in the active state.

12. The method of claim 11, wherein the objective function is configured to evaluate a performance measure of a binary bit string.

13. The method of claim 11, wherein the objective function is Area Under the Receiver Operating Curve of a Linear Discriminant Function.

14. The method of claim 11, wherein at least two bit values are changed at two random locations on the binary bit string during the inverting step.

15. The method of claim 14, wherein a first bit of the at least two bit values is changed from the active state to the inactive state and a second bit of the at least two bit values is changed from the inactive state to the active state.

16. The method of claim 11, wherein the quantitative features derived from the electrical brain activity data in the reference database include non-linear features.

17. The method of claim 11, wherein the final linear discriminant function is a Linear Discriminant Function for classification of brain electrical activity data into diagnostic categories that include at least one of:

(i) normal brain function, (ii) abnormal brain function, (iii) organic brain dysfunction, (iv) functional brain dysfunction, (v) focal brain dysfunction, (vi) diffuse brain dysfunction, (v) (closed-head) traumatic brain injury (TBI), and (vi) mild TBI (concussion).

18. A method of building a binary classifier for classifying brain electrical activity data into diagnostic categories, the method comprising:

providing a data storage device comprising a reference database of brain electrical activity data, wherein the brain electrical activity data is recorded from a plurality of individuals in the presence or absence of brain abnormalities using one or more neurological electrodes;

providing a signal processing device operatively connected to the data storage device, the signal processing device comprising a processor configured to perform the steps of;

acquiring quantitative brain electrical activity features from the brain electrical activity data in the reference database,

calculating one or more quantitative measures for each of the acquired quantitative features, the one or more quantitative measures being indicative of classification performance of the acquired quantitative features,

selecting with the processor a reduced pool of brain electrical activity features from the acquired quantitative features based on the one or more quantitative measures associated with the acquired quantitative features,

selecting multiple subsets of brain electrical activity features from the reduced pool of features,

constructing an initial population of chromosomes, each chromosome of the initial population including one of the selected subsets of features and corresponding to a binary classifier for classification of brain electrical activity data into diagnostic categories,

applying genetic algorithm operators to the initial population of chromosomes to derive a new population of chromosomes, each chromosome in the new population including a subset of the brain electrical activity features that is generated from the subsets of features of the initial population of chromosomes,

determining the classification performance of the chromosomes in the new population using an objective function, and

repeating the applying and determining steps until the classification performance of one of the chromosomes in the new population is above a predetermined threshold; and

storing a final binary classifier for classification of brain electrical activity data into the diagnostic categories on non-volatile media, wherein the stored final binary classifier is defined by the subset of brain electrical activity features of the chromosome having the classification performance above the predetermined threshold.

19. The method of claim 18, wherein the one or more quantitative measures comprises a measure of the replicability of the features.

20. The method of claim 18, wherein the one or more quantitative measures comprises a measure of the discriminative power of the features.

21. The method of claim 18, wherein the final binary classifier is a Linear Discriminant Function for classification of brain electrical activity data into diagnostic categories that include at least one of:

(i) normal brain function, (ii) abnormal brain function, (iii) organic brain dysfunction, (iv) functional brain dysfunction, (v) focal brain dysfunction, (vi) diffuse brain dysfunction, (v) (closed-head) traumatic brain injury (TBI), and (vi) mild TBI (concussion).