Method and apparatus for the evaluation of EEG data

• US 5,846,208 A
• Filed: 08/22/1997
• Issued: 12/08/1998
• Est. Priority Date: 09/04/1996
• Status: Expired due to Fees

• Current Assignee: Siemens AG
• Original Assignee: Siemens AG
• Inventors: Eckert, Olaf, Pichlmayr, Ina
• Primary Examiner(s): Getzow, Scott M.
• Application Number:
  US08/916,398
• Time in Patent Office: 473 Days
• Field of Search: 600/544, 600/545
• US Class Current: 600/544
• 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 5/412 Detecting or monitoring sepsis
• CPC Class Codes: A61B 5/4821 Determining level or depth of anaesthesia A61B5/1106 takes precedence
• CPC Class Codes: A61B 5/7203 for noise prevention, reduction or removal
• CPC Class Codes: A61B 5/7239 using differentiation including higher order derivatives
• CPC Class Codes: A61B 5/7257 using Fourier transforms
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

A method for the evaluation of EEG data for medical purposes includes the steps of acquiring EEG data, recognizing artifacts and determining an output data value with a neural network. In a training method for a neural network, training vectors are determined to which a respective data value is allocated, the neural network is trained with these training vectors, and an output data value is determined for each neuron that is based on the allocated data values of those training vectors that are contained in the data cluster represented by the neuron. A processing apparatus and an EEG monitor are configured for implementing the evaluation method for EEG data.

Claims

1. A method for evaluating EEG data for medical purposes, comprising the steps of:

(a) acquiring EEG data from a patient and editing said EEG data to produce edited EEG data;

(b) identifying any sections of said EEG data falsified by artifacts and producing artifact information relating thereto;

(c) calculating a feature vector from the edited EEG data and from said artifact information;

(d) determining an initial data value in a neural network by allocating said feature vector to a data cluster represented by a neuron of said neural network, said initial data value being allocated to said data cluster; and

(e) producing an output data value from said initial data value and displaying said output data value.

2. A method as claimed in claim 1 wherein step (a) comprises editing said EEG data by spectrally analyzing said EEG data acquired from said patient.

3. A method as claimed in claim 1 wherein step (b) comprises at least one of the steps of:

comparing parameters of said EEG data to predetermined limit values and identifying a presence of an artifact if at least one of said predetermined limit values is exceeded by said parameters of said EEG data; and

obtaining smoothed parameter values from previous EEG data and comparing said smoothed parameter values to current parameter values of said EEG data, and identifying a presence of an artifact in said EEG data if said current parameter values deviate from said smoothed parameter values by an amount exceeding a predetermined limit value.

4. A method as claimed in claim 3 comprising, in each comparing step, selecting power values of predetermined frequency bands of said EEG data as said parameters.

5. A method as claimed in claim 1 wherein step (c) comprises calculating a feature vector only from edited EEG data which has been identified in step (b) as not being falsified by artifacts.

6. A method as claimed in claim 1 wherein step (c) comprises at least one of the steps of:

calculating components comprising said feature vector indicating a signal power in a predetermined frequency band of the edited EEG data during a measurement interval; and

calculating at least one component indicating an appearance of a burst suppression pattern in said edited EEG data.

7. A method as claimed in claim 1 wherein step (d) comprises initially conducting a coarse classification of said feature vector in a first sub-network of said neural network for determining said data cluster allocated to said feature vector.

8. A method as claimed in claim 7 wherein the step of conducting said coarse classification comprises comparing parameters allocated to said feature vector to predetermined limit values.

9. A method as claimed in claim 1 wherein each neuron of said neural network has a synaptic vector associated therewith and wherein the step of allocating said feature vector to a data cluster represented by a neuron in step (d) comprises allocating a feature vector to a data cluster represented by a neuron of said neural network having a synaptic vector which is most similar to said feature vector.

10. A method as claimed in claim 1 wherein each neuron of said neural network has a parameter indicating an output data value for the data cluster represented by that neuron, and wherein step (e) comprises displaying said output data value for the data cluster represented by the neuron.

11. A method as claimed in claim 1 wherein each neuron of said neural network has a synaptic vector associated therewith, and comprising the additional step of determining parameters of said neural network, including said synaptic vectors, by a self-organizing training method.

12. A method as claimed in claim 11 wherein said neural network has a topology, and wherein the step of determining parameters of said neural network comprises determining parameters of said neural network by a self-organizing training method which is independent of said topology of said neural network.

13. A method as claimed in claim 1 wherein step (e) comprises determining a hypnosis index of said patient as said output data value.

14. A method for training neural network for evaluating EEG data for medical purposes, comprising the steps of:

(a) determining a plurality of training vectors, each training vector having a data value allocated thereto;

(b) determining initial parameters of a neural network;

(c) conducting a training algorithm on said neural network including modifying said parameters of said neural network for allocating one neuron of said neural network to each cluster of training vectors; and

(d) determining an output data value for each neuron dependent on the respective data values for all training vectors that are contained in the cluster of training vectors allocated to that neuron.

15. A method as claimed in claim 14 wherein step (d) is further defined by determining said output data value for a neuron by averaging the respective data values of all of said training vectors contained in the cluster of training vectors represented by that neuron.

16. A method as claimed in claim 14 wherein said neural network is free of any predetermined topology during steps (a), (b), (c) and (d).

17. A method as claimed in claim 16 comprising conducting training steps for each neuron according to the equation

18. A method as claimed in claim 14 comprising initially conducting a coarse classification of said training vectors among a number of coarses categories and respectively allocating different sub-networks of said neural network to each coarse category of training vectors, each sub-network then being trained only with the training vectors contained in the coarse category allocated to that sub-network.

19. A method as claimed in claim 18 comprising splitting a coarse category of training vectors if no satisfactory classification results can be obtained for all training vectors in said coarse category.

20. A processor for training neural network for evaluating EEG data for medical purposes, comprising:

means for determining a plurality of training vectors, each training vector having a data value allocated thereto;

means for determining initial parameters of a neural network;

means for conducting a training algorithm on said neural network including modifying said parameters of said neural network for allocating one neuron of said neural network to each cluster of training vectors; and

means for determining an output data value for each neuron dependent on the respective data values for all training vectors that are contained in the cluster of training vectors allocated to that neuron.

21. An EEG apparatus comprising:

means for acquiring EEG data from a patient and editing said EEG data to produce edited EEG data;

means for identifying any sections of said EEG data falsified by artifacts and producing artifact information relating thereto;

means for calculating a feature vector from the edited EEG data and from said artifact information;

means for determining an initial data value in a neural network by allocating said feature vector to a data cluster represented by a neuron of said neural network, said initial data value being allocated to said data cluster; and

means for producing an output data value from said initial data value and displaying said output data value.

22. An EEG apparatus as claimed in claim 21 wherein said means for displaying comprises means for displaying momentary output data values.

23. An EEG apparatus as claimed in claim 21 wherein said means for displaying comprises means for displaying continuously updated and graduated output data values.

24. An EEG monitor as claimed in claim 21 wherein said means for producing an output value comprises means for producing a hypnosis index and for displaying said hypnosis index.