METHOD AND SYSTEM FOR CONCENTRATION DETECTION

• US 20110191350A1
• Filed: 04/28/2008
• Published: 08/04/2011
• Est. Priority Date: 04/28/2008
• Status: Active Grant

• Current Assignee: Agency For Science Technology and Research
• Original Assignee: Agency For Science Technology and Research
• Inventors: Guan, Cuntai, Zhang, Haihong, Hamadi Charef, Brahim Ahmed Salah, Phua, Kok Soon, Wang, Chuanchu
• Granted Patent: US 8,862,581 B2
• Time in Patent Office: Days
• US Class Current: 707/743
• CPC Class Codes: A61B 5/165 Evaluating the state of mind, e.g. depression, anxiety
• CPC Class Codes: A61B 5/168 Evaluating attention deficit, hyperactivity
• CPC Class Codes: A61B 5/374 Detecting the frequency distribution of signals, e.g. detecting delta, theta, alpha, beta or gamma waves
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: G06F 2218/08 Feature extraction
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

A method and system for concentration detection. The method for concentration detection comprises the steps of extracting temporal features from brain signals; classifying the extracted temporal features using a classifier to give a score x₁; extracting spectral-spatial features from brain signals; selecting spectral-spatial features containing discriminative information between concentration and non-concentration states from the set of extracted spectral-spatial features; classifying the selected spectral-spatial features using a classifier to give a score x₂; combining the scores x₁ and x₂ to give a single score and determining if the subject is in a concentration state based on the single score.

Claims

1. A method for concentration detection, the method comprising the steps of:

extracting temporal features from brain signals;

classifying the extracted temporal features using a classifier to give a score x₁;

extracting spectral-spatial features from brain signals;

selecting spectral-spatial features containing discriminative information between concentration and non-concentration states from the set of extracted spectral-spatial features;

classifying the selected spectral-spatial features using a classifier to give a score x₂;

combining the scores x₁ and x₂ to give a single score; and

determining if the subject is in a concentration state based on the single score.

2. The method as claimed in any of the preceding claims, wherein the step of extracting temporal features from brain signals further comprises the steps of:

computing statistics of brain waveforms in each of a plurality of electrode channels; and

concatenating the statistics into a joint feature vector.

3. The method as claimed in claim 2, wherein the statistics of the brain waveforms are standard deviations.

4. The method as claimed in any of the preceding claims, wherein the step of extracting spectral-spatial features of brain signals further comprises the steps of:

extracting respective brain signal components in discrete frequency windows using filter banks to obtain spectral features of brain signals; and

applying a CSP algorithm to each of the spectral features using a CSP array to obtain the spectral-spatial features of brain signals.

5. The method as claimed in claim 4, wherein said filter banks comprise low-order bandpass Chebyshev Type II filters with a pass-band width of 4 Hz.

6. The method as claimed in any of the preceding claims, wherein the step of selecting spectral-spatial features containing discriminative information between concentration and non-concentration states from the set of extracted spectral-spatial features further comprises the step of selecting spectral-spatial features based on the mutual dependence of the features with respect to the concentration and non-concentration states.

7. The method as claimed in any of the preceding claims, wherein the step of combining the scores x₁ and x₂ to give a single score further comprises the steps of:

normalizing the scores x₁ and x₂ according to an equation (x−

m_x)/s_x whereby m_x and s_x are the mean and standard deviation of outputs from the classifiers using training samples to give x_1n and x_2n respectively;

assigning weights w₁ and w₂ to normalized scores x_1n and x_2n respectively; and

combining the scores x_1n and x_2n according to an equation x_1n*w₁+x_2n*w₂ to give a single score.

8. The method as claimed in claim 7 wherein said weights w₁ and w₂ are calculated according to the equation w_i=(y_i)^p where y_i is the classification accuracy in classifying the extracted temporal features if i=1 and in classifying the extracted spectral-spatial features if i=2 and p (p>

0) controls the power of w_i in the calculation of the single score.

9. The method as claimed in any of the preceding claims wherein the step of determining if the subject is in a concentration state based on the single score further comprises determining that the subject is in a concentration state if the single score is higher than a threshold and that the subject is not in a concentration state if the single score is lower than a threshold.

10. The method as claimed in any of the preceding claims, wherein said classifier comprises one or more of a group consisting of a Linear Discriminant Analysis classifier, Neural Networks, Support Vector Machines, Fuzzy Inference System, Tree-based classifiers, Fuzzy Type 2 and Relevance Vector Machine.

11. The method as claimed in any of the preceding claims, the method further comprises the step of using training data to generate parameters for classifying the extracted temporal features using a classifier, for extracting spectral-spatial features from brain signals, for selecting spectral-spatial features containing discriminative information between the concentration and non-concentration states from the set of extracted spectral-spatial features and for classifying the selected spectral-spatial features using a classifier.

12. The method as claimed in claim 11, wherein said parameters comprise one or more of a group consisting of projection matrices of CSPs for the CSP algorithm, parameters for selecting spectral-spatial features based on mutual information and a model for the classifiers.

13. The method as claimed in claim 11 or 12, wherein said step of using training data to generate parameters further comprises the steps of:

collecting training data from subjects performing a set of tasks; and

determining said parameters via machine learning methods.

14. The method as claimed in claim 13, wherein said set of tasks comprises one or more of a group consisting of reading a technical paper, performing mental arithmetic with closed eyes, relaxing and looking around, and resting with closed eyes.

15. A system for concentration detection, the system comprising:

a temporal feature extracting unit for extracting temporal features from brain signals;

a temporal feature classifying unit for classifying the extracted temporal features using a classifier to give a score x₁;

a spectral-spatial feature extracting unit for extracting spectral-spatial features from brain signals;

a spectral-spatial feature selecting unit for selecting spectral-spatial features containing discriminative information between the concentration and non-concentration states from the set of extracted spectral-spatial features;

a spectral-spatial feature classifying unit for classifying the selected spectral-spatial features using a classifier to give a score x₂; and

a processing unit coupled to said temporal feature classifying unit and said spectral-spatial feature classifying unit for combining the scores x₁ and x₂ to give a single score and for determining if the subject is in a concentration state based on the single score.

16. The system as claimed in claim 15, wherein the system further comprises:

filter banks to extract respective brain signal components in discrete frequency windows to obtain spectral features of brain signals; and

a CSP array to apply a CSP algorithm to each of the spectral features to obtain the spectral-spatial features of brain signals;

17. The system as claimed in claim 16, wherein said filter banks comprise low-order bandpass Chebyshev Type II filters with a pass-band width of 4 Hz.

18. A data storage medium having stored thereon computer code means for instructing a computer system to execute a method for concentration detection, the method comprising the steps of:

extracting temporal features from brain signals;

classifying the extracted temporal features using a classifier to give a score x₁;

extracting spectral-spatial features from brain signals;

selecting spectral-spatial features containing discriminative information between the concentration and non-concentration states from the set of extracted spectral-spatial features;

classifying the selected spectral-spatial features using a classifier to give a score x₂;

combining the scores x₁ and x₂ to give a single score; and

determining if the subject is in a concentration state based on the single score.

19. A method for concentration detection, the method comprising the steps of:

extracting features from brain signals;

selecting features containing discriminative information between concentration and non-concentration states from the set of extracted features;

classifying the selected features using a classifier to give a score;

wherein subject dependant training data is used to generate parameters for extracting the features from the brain signals, for selecting the features containing discriminative information between the concentration and non-concentration states from the set of extracted features and for classifying the selected features using a classifier; and

determining if the subject is in a concentration state based on the score.

20. A system for concentration detection, the system comprising:

a feature extracting unit for extracting features from brain signals;

a feature selecting unit for selecting features containing discriminative information between concentration and non-concentration states from the set of extracted features;

a feature classifying unit for classifying the selected features using a classifier to give a score;

wherein subject dependant training data is used to generate parameters for extracting the features from the brain signals, for selecting the features containing discriminative information between the concentration and non-concentration states from the set of extracted features and for classifying the selected features using a classifier; and

a processing unit for determining if the subject is in a concentration state based on the score.

21. A data storage medium having stored thereon computer code means for instructing a computer system to execute a method for concentration detection, the method comprising the steps of:

extracting features from brain signals;

selecting features containing discriminative information between concentration and non-concentration states from the set of extracted features;

classifying the selected features using a classifier to give a score;

wherein subject dependant training data is used to generate parameters for extracting the features from the brain signals, for selecting the features containing discriminative information between the concentration and non-concentration states from the set of extracted features and for classifying the selected features using a classifier; and

determining if the subject is in a concentration state based on the score.