Method and apparatus for extracting low SNR transient signals from noise

• US 6,898,582 B2
• Filed: 09/26/2001
• Issued: 05/24/2005
• Est. Priority Date: 12/30/1998
• Status: Expired due to Fees

• Current Assignee: Algodyne Limited
• Original Assignee: Algodyne Limited
• Inventors: Lange, Daniel H.
• Primary Examiner(s): Knight, Anthony
• Assistant Examiner(s): Hirl, Joseph P.
• Application Number:
  US09/965,446
• Publication Number:
  US 20040015894A1
• Time in Patent Office: 1,336 Days
• Field of Search: 706/14, 706/15, 706/30
• US Class Current: 706/14
• 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: G06F 2218/04 Denoising
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

A method and apparatus for processing a composite signal generated by a transient signal generation mechanism to extract a repetitive low SNR transient signal, such as an evoked potential (EP) appearing in an electroencephalogram (EEG) generated in response to sensory stimulation, by: (a) dynamically identifying, via a learning process, the major transient signal types in the composite signal; (b) decomposing the identified major transient signal types into their respective constituent components; (c) synthesizing a parametric model emulating the transient signal generation mechanism; and (d) utilizing the model and the constituent components to identify and extract the low SNR transient signal from the composite signal.

Claims

1. A method of extracting low SNR signals from noise comprising the following steps:

a) dynamically identifying, via a competitive neural network, various transient response types embedded within a recorded ensemble of composite signals;

b) decomposing the response types;

c) constructing a parametric synthesis model that emulates a single-sweep signal generation mechanism.

2. The method of claim 1 further comprising the step of:

taking a single sweep composite signal, identifying its response type, filtering its corresponding decomposed components through the parametric synthesis model'"'"'s filters and summing the resulting filtered components to extract an evoked potential.

3. The method of claim 2 wherein said method extracts an evoked potential appearing in an electroencephalogram.

4. The method of claim 2 wherein said method further comprises the step of displaying said evoked potential.

5. The method according to claim 1, wherein said process (a) is performed by using an adaptive, competitive neural network.

6. The method according to claim 2, wherein said neural network includes a group of artificial neurons divided into sets of inhibitory clusters in which all neurons within a cluster inhibit all other neurons in the cluster, resulting in a competition among the neurons in a cluster to respond to the major transient signal patterns in the composite signal.

7. The method according to claim 2, wherein said process (a) includes:

(i) initializing network weights;

(ii) selecting one input vector at random;

(iii) calculating all neural outputs in selecting a winner neuron;

(iv) updating the winning neuron weights;

(v) repeating the foregoing steps until a stopping criterion is achieved; and

(vi) displaying the identified signal types.

8. The method according to claim 1, wherein said decomposing process (b) is performed by treating peaks of the composite signal as independent components and analyzing said peaks.

9. The method according to claim 5, wherein said process (b) includes:

(i) initializing search limits;

(ii) calculating peak locations;

(iii) optimizing first and second peak moments;

(iv) building a decomposition rule;

(v) executing a decomposition procedure to identify signal types; and

(vi) displaying the identified signal types.

10. The method according to claim 1, wherein said synthesizing process (c) is performed by modeling the EEG via auto regressive (AR) analysis and modeling EP as a sum of finite impulse responses (FIR) filtered EP components.

11. The method according to claim 7, wherein said synthesizing process (c) includes:

(i) assigning transition points;

(ii) building an AR model of the pre-stimulus EEG signal;

(iii) whitening the EEG and EP signals;

(iv) identifying the EP by performing a closed form least square solution of the model; and

(v) extracting the EP signal.

22. A method for extracting an evoked potential comprising the steps of:

obtaining a single sweep composite signal using an adaptive, competitive neural network;

identifying said signal'"'"'s response type;

filtering its corresponding decomposed components through the parametric synthesis model filters based upon the model described in claim 1;

summing the resulting filtered components to extract a transient evoked potential.

23. The method of claim 22 wherein said evoked potential is an evoked potential appearing in an electroencephalogram.

24. An apparatus for extracting an evoked potential comprising:

a competitive neural network;

a transient signal response type identifier;

parametric synthesis model filters based on the model described in claim 1;

a summer.

25. The apparatus of claim 24 wherein said model of claim 1 models an electroencephalogram signal.

12. An apparatus for extracting low SNR signals from noise comprising:

a competitive neural network transient response identifier;

a response type decomposition unit;

a parametric synthesis model that emulates a single-sweep signal generation mechanism.

13. The apparatus of claim 12 further comprising:

an evoked potential extractor including a single sweep composite signal procurer;

a response type identifier;

parametric synthesis model decomposed component filters; and

a filtered components summer.

14. The apparatus according to claim 13, wherein said neural network includes a group of artificial neurons divided into sets of inhibitory clusters in which all neurons within a cluster inhibit all other neurons in the cluster, resulting in a competition among the neurons in a cluster to respond to the major transient signal patterns in the composite signal.

15. The apparatus according to claim 13, wherein said identifier:

(i) initializes network weights;

(ii) selects one input vector at random;

(iii) calculates all neural outputs in selecting a winner neuron;

(iv) updates the winning neuron weights; and

(v) repeats the foregoing steps until a stopping criterion is achieved.

16. The apparatus according to claim 12, wherein said decomposition unit operates by treating peaks of the composite signal as independent components and analyzing said peaks.

17. The apparatus according to claim 12, wherein said decomposition unit:

(i) initializes search limits;

(ii) calculates peak locations;

(iii) optimizes first and second peak moments;

(iv) builds a decomposition rule; and

(v) executes a decomposition procedure to identify signal types.

18. The apparatus according to claim 12, wherein the low SNR transient signal is a transient evoked potential (EP) appearing in an electroencephalogram (EEG) of a living being generated in response to sensory stimulation of the nervous system of the living being.

19. The apparatus according to claim 18, wherein said a parametric synthesis model models the BEG via an AR analysis, and models the EP as a sum of the FIR filtered EP components.

20. The apparatus according to claim 17, wherein said parametric synthesis model:

(i) assigns transition points;

(ii) builds an AR model of the pre-stimulus BEG signal;

(iii) whitens the EEG and EP signals;

(iv) identifies the EP by performing a closed form least square solution of the model; and

(v) extracts the EP signal.

21. The apparatus according to claim 12, wherein said identifier, said decomposition unit, and said model are in the form of computer software.