METHOD FOR THE REAL-TIME IDENTIFICATION OF SEIZURES IN AN ELECTROENCEPHALOGRAM (EEG) SIGNAL

• US 20120101401A1
• Filed: 04/07/2010
• Published: 04/26/2012
• Est. Priority Date: 04/07/2009
• Status: Active Grant

• Current Assignee: National University of Ireland Galway
• Original Assignee: National University of Ireland Galway
• Inventors: Faul, Stephen Daniel, Temko, Andrly, Marnane, William Peter, Lightbody, Gordon, Boylan, Geraldine Bernadette
• Granted Patent: US 10,433,752 B2
• Time in Patent Office: Days
• US Class Current: 600/544
• CPC Class Codes: A61B 5/369 Electroencephalography [EEG] devices for psychotechnics A61B5/16
• CPC Class Codes: A61B 5/372 Analysis of electroencephalograms
• CPC Class Codes: A61B 5/4094 Diagnosing or monitoring seizure diseases, e.g. epilepsy
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: G06F 2218/12 Classification; Matching
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

The present invention relates to a method for the real-time identification of seizures in an Electroencephalogram (EEG) signal. The method provides for patient-independent seizure identification by use of a multi-patient trained generic Support Vector Machine (SVM) classifier. The SVM classifier is operates on a large feature vector combining features from a wide variety of signal processing and analysis techniques. The method operates sufficiently accurately to be suitable for use in a clinical environment. The method may also be combined with additional classifiers, such a Gaussian Mixture Model (GMM) classifier, for improved robustness, and one or more dynamic classifiers such as an SVM using sequential kernels for improved temporal analysis of the EEG signal.

Claims

1-50. -50. (canceled)

51. A method for the real-time identification of seizures in an Electroencephalogram (EEG) signal, the steps of the method comprising:

(a) receiving an EEG signal comprising a plurality of channels of EEG data;

(b) for each channel, segmenting the data into sequential epochs, each epoch having an overlap with its neighbouring epochs;

and for an initial epoch performing the following steps comprising;

(c) extracting forty five or more features from each of the constituent channels;

(d) generating a feature vector from the extracted features;

(e) passing the feature vector for each of the constituent channels separately through a multi-patient trained generic Support Vector Machine (SVM) classifier and generating an SVM channel seizure output, in which the multi-patient trained generic support vector machine classifier is trained on EEG data representing all seizure types, over all channels and over all patient types;

(f) fusing the SVM channel seizure outputs for all channels thereby generating an SVM epoch seizure output; and

(g) repeating steps (c) to (f) for each subsequent epoch thereby generating a sequence of SVM channel seizure outputs and SVM epoch seizure outputs.

52. The method as claimed in claim 51 comprising the additional steps of:

generating a reduced feature vector;

passing the reduced feature vector through a multi-patient trained Gaussian Mixture Model (GMM) classifier; and

generating a GMM channel seizure output.

53. The method as claimed in claim 52 comprising the additional step of combining the SVM channel seizure output and GMM channel seizure output for each epoch thereby generating a combined channel seizure output.

54. The method as claimed in claim 53 comprising the additional step of fusing the combined channel seizure outputs for an epoch across all channels thereby generating a combined epoch seizure output for each epoch.

55. The method as claimed in claim 51 comprising the additional step of converting each channel seizure output into a binary channel seizure decision.

56. The method as claimed in claim 51 comprising the additional step of converting each channel seizure output to a normalised probabilistic value between 0 and 1.

57. The method as claimed in claim 51 in which the step of fusing the channel seizure outputs comprises using a logical OR operator.

58. The method as claimed in claim 51 comprising the additional step of applying a moving average filter to the sequence of epoch seizure outputs.

59. The method as claimed in claim 51 comprising the additional step of applying a collar technique to the sequence of epoch seizure outputs

60. The method as claimed in claim 51 in which the step of extracting features from the constituent channels comprises extracting features using speech recognition analysis techniques.

61. The method as claimed in claim 51 in which the step of extracting features from the constituent channels comprises extracting features using time domain analysis techniques, frequency domain analysis techniques and information theory analysis techniques.

62. The method as claimed in claim 51 in which the method comprises the additional step of passing the feature vector through a Support Vector Machine (SVM) with sequential kernels and generating a dynamic SVM channel seizure output.

63. The method as claimed in claim 62 in which the method comprises the additional steps of:

combining the dynamic SVM channel seizure output for an epoch with any other channel seizure outputs for that epoch, thereby generating a dynamic SVM combined seizure output; and

fusing the dynamic SVM combined seizure outputs for that epoch over all channels thereby generating a dynamic SVM combined epoch seizure output.

64. The method as claimed in claim 51 further comprising passing the feature vector through a multi-patient trained generic SVM classifier and generating an SVM channel artefact output.

65. A computer program product having computer program instructions for causing a computing device to perform the method of claim 51.

66. An apparatus for the real-time identification of seizures in an Electroencephalogram (EEG) signal, the apparatus comprising:

means for receiving an EEG signal comprising a plurality of channels of EEG data;

means for segmenting the data of each channel into a plurality of sequential epochs, each epoch having an overlap with its neighbouring epochs;

means for extracting forty five or more features from the constituent channels;

means for generating a feature vector from the extracted features;

a multi-patient trained generic Support Vector Machine (SVM) classifier adapted to process the feature vector so as to generate an SVM channel seizure output, the multi-patient trained generic SVM classifier is trained on EEG data representing all seizure types, over all channels and over all patient types; and

means for fusing the SVM channel seizure outputs for all channels thereby generating an SVM epoch seizure output.

67. The apparatus as claimed in claim 66 comprising:

means for generating a reduced feature vector;

means for passing the reduced feature vector through a multi-patient trained Gaussian Mixture Model (GMM) classifier and generating a GMM channel seizure output;

means for combining the SVM channel seizure output and GMM channel seizure output for each epoch thereby generating a combined channel seizure output; and

means for fusing the combined channel seizure outputs for an epoch across all channels thereby generating a combined epoch seizure output for each epoch.

68. The apparatus as claimed in claim 66 comprising means to convert each channel seizure output into a binary channel seizure decision.

69. The apparatus as claimed in claim 66 comprising means to apply a moving average filter to the sequence of epoch seizure outputs.

70. The apparatus as claimed in claim 66 comprising means to apply a collar technique to the sequence of epoch seizure outputs.