Method and apparatus for detecting nonlinearity in an electrocardiographic signal

• US 5,792,062 A
• Filed: 05/14/1996
• Issued: 08/11/1998
• Est. Priority Date: 05/14/1996
• Status: Expired due to Term

• Current Assignee: Massachusetts Institute of Technology
• Original Assignee: Massachusetts Institute of Technology
• Inventors: Barahona, Mauricio, Merrill, Christopher K., Poon, Chi-Sang
• Primary Examiner(s): Kamm, William E.
• Application Number:
  US08/645,793
• Time in Patent Office: 819 Days
• Field of Search: 128/696, 128/705, 128/706
• US Class Current: 600/509
• CPC Class Codes: G06F 17/18 for evaluating statistical data , e.g. average values, frequency distributions, probability functions, regression analysis forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04
• CPC Class Codes: G06F 2218/00 Aspects of pattern recognition specially adapted for signal processing

Abstract

A method and apparatus for detecting the presence of a nonlinear characteristic in a data sequence is provided.

Claims

1. An apparatus for analyzing an electrocardiograph signal to determine whether the electrocardiograph signal includes a nonlinear signal component the apparatus comprising:

(a) a modeling system comprising;

a modeling processor for receiving the electrocardiograph signal and for representing at least a portion of the electrocardiograph signal with a first linear model having a first set of parameters and a first set of coefficients and for representing at least a portion of the electrocardiograph signal with a second model corresponding to a first nonlinear model having a second set of parameters and a second set of coefficients; and

a performance processor for computing a first performance measure of the first model and for computing a second performance measure of the second model; and

(b) a performance measurement system coupled to said modeling system said performance measurement system comprising;

a comparison processor for receiving the first and second performance measures and for comparing the first performance measure to the second performance measure; and

a selection processor for receiving an indication from said comparison processor and for identifying which of the first and second performance measures is a preferred performance measure and for providing an output signal indicating whether the electrocardiograph signal includes a nonlinear signal component.

2. The apparatus of claim 1 wherein said performance measurement system further comprises:

a residual sequence generator having an input port coupled to receive a sequence of values generated from a linear model and having an output port, the residual sequence generator for providing a residual data sequence at the output port thereof; and

a residual modeling system having an input port coupled to the output port of said residual sequence generator and having an output port, the residual modeling system for generating a linear model and a nonlinear model of the residual data sequence fed received at the input port thereof and for comparing the nonlinear model to the linear model.

3. The apparatus of claim 1 wherein said modeling system further comprises at least one sensor having an input port and having an output port coupled to provide data to said modeling processor.

4. The apparatus of claim 1 wherein said modeling system further comprises a data storage device coupled to said modeling processor and having the electrocardiograph signal stored therein.

5. A method for detecting a nonlinear pattern in a bio-signal, the method comprising the steps of:

representing at least a portion of the bio-signal with a linear model having a first set of parameters;

representing at least a portion of the bio signal with a nonlinear model, the first nonlinear model having a second set of parameters;

computing a first performance measure of the first model;

computing a second performance measure of the second model;

comparing the first performance measure to the second performance measure;

identifying which of the first and second performance measures is a preferred performance measure; and

providing an output signal indicating whether the bio-signal includes a nonlinear signal component.

6. The method of claim 5 further comprising the steps of:

selecting a first one of the first and second models having the preferred performance measure to represent the bio-signal.

7. The method of claim 6 wherein the comparing step includes the step of:

determining a difference value between the first performance measure and the second performance measure;

comparing the difference value to a predetermined threshold difference value;

in response to the difference value being equal to or greater than the threshold difference value, identifying a respective one of the performance measures as the preferred performance measure.

8. The method of claim 6 further comprising the steps of:

generating a surrogate data set;

representing the surrogate data set with a second linear model having a third set of parameters and a third set of coefficients;

representing the surrogate data set with a second nonlinear model having a fourth set of parameters and a fourth set of coefficients;

comparing a performance measure of the second linear model to a performance measure of the second nonlinear model

identifying which of the performance measures is a preferred performance measure;

selecting a first one of the second linear model and second nonlinear model having the preferred performance measure; and

comparing the first one of the second linear model and second nonlinear model to the first one of the linear model and nonlinear model.

9. The method of claim 8 wherein the performance measure corresponds to a first one of:

an Akiake information criteria;

a minimum description length;

a log likelihood measure; and

a residual sum of the square error.

10. The method of claim 6 wherein:

the first set of parameters and first set of coefficients of the linear model correspond to parameters and coefficients which provide an optimum linear model; and

the second set of parameters and second set of coefficients of the nonlinear model correspond to parameters and coefficients which provide an optimum nonlinear model.

11. The method of claim 10 wherein the steps of representing the bio-signal with a nonlinear model and a nonlinear model includes the step of generating linear and nonlinear models corresponding to Volterra-Wiener models.

12. The method of claim 11 wherein the step of comparing the performance measure of the linear model to the performance measure of the nonlinear model comprises the step of statistically comparing a first value of the performance measure of the linear model to a first value of the performance measure of the nonlinear model.

13. A method for detecting nonlinearity in a data set generated by a biological system, the method comprising the steps of:

representing the data set with a linear model having a first set of parameters and a first set of coefficients;

representing the data set with a non-linear model having a first set of parameters and a first set of coefficients;

computing a performance measure of the linear model;

computing a performance measure of the nonlinear model;

comparing the performance measure of the linear model to the performance measure of the nonlinear model;

identifying which of the performance measures of the linear model and the nonlinear model is a preferred performance measure;

selecting a first one of the linear and nonlinear models having the preferred performance measure;

and

providing an output signal indicating whether the data set generated by the biological system includes a nonlinear signal component.

14. The method of claim 13 wherein the step of comparing the performance measure of the linear model to the performance measure of the nonlinear model includes the step of statistically comparing the performance measure of the linear model to the performance measure of the nonlinear model.

15. The method of claim 14 wherein the linear model and nonlinear model are not the optimum models.

16. The method of claim 14 wherein:

the linear model corresponds to an optimum linear model; and

the nonlinear model correspond to an optimum nonlinear model.

17. A method for analyzing a hemodynamic signal to determine the presence of a chaotic signal component within the hemodynamic signal, the method comprising the steps of:

dividing the hemodynamic signal into a plurality of time segments, each of the plurality of time segments having a discrete value;

representing each of the plurality of time segments of the hemodynamic signal with a first model corresponding to a first linear model;

representing each of the plurality of time segments of the hemodynamic signal with a second model corresponding to a first non-linear model;

computing a performance measure of the first lineary model,

computing a performance measure of the first non-linear model; and

comparing the performance measure of the first model to the performance measure of the second model.

18. The method of claim 17 further comprising the steps of:

identifying which of the first and second performance measures is a preferred performance measure; and

selecting a first one of the first and second models having the preferred performance measure.

19. The method of claim 18 wherein:

the hemodynamic signal corresponds to an electrocardiograph signal; and

the time segments correspond to RR intervals of the electrocardiograph signal.

20. The method of claim 17 wherein the step of comparing the performance measure of the first model to the performance measure of the second model includes the step of statistically comparing the performance measure of the first model to the performance measure of the second model.

21. A method for analyzing an electrocardiograph signal to detect the presence of a chaotic signal component in the electrocardiograph signal, the method comprising the steps of:

generating a linear model to represent a first sequence of data values corresponding to measured RR intervals of the electrocardiograph signal, the linear model having a first predetermined number of terms; and

generating a non-linear model to represent the first sequence of data values corresponding to the series of measured RR intervals of the electrocardiograph signal, the non-linear model having a second predetermined number of terms,

computing a performance measure of the linear model;

computing a performance measure of the nonlinear model; and

comparing the performance measure of the linear model to the performance measure of the nonlinear model.

22. The method of claim 21 further comprising the steps of:

identifying one of the performance measure of the linear model and the performance measure of the nonlinear model as a preferred performance measure;

in response to the preferred performance measure corresponding to the performance measure of the linear model performing the steps of;

generating a predicted sequence of data values using the linear model;

subtracting the predicted linear sequence of data values from the first sequence of data values to generate a residual sequence of data values;

generating a linear model to represent the residual sequence of data values;

generating a nonlinear model to represent the residual sequence of data values;

computing a performance measure of the linear model which represents the residual sequence of data values;

computing a performance measure of the nonlinear model which represents the residual sequence of data values; and

comparing the performance measure of the linear model which represents the residual sequence of data values to the performance measure of the nonlinear model which represents the residual sequence of data values.

23. The method of claim 22 further comprising the step of identifying one of the performance measures of the linear and nonlinear models which represent the residual sequence of data values having a preferred value.

24. The method of claim 23 further comprising the steps of:

providing a first medical diagnosis in response to the performance measure of the linear model which represents the residual sequence of data values having the preferred value; and

providing a second medical diagnosis in response to the performance measure of the nonlinear model which represents the residual sequence of data values having the preferred value.

25. A method for selecting samples from a continuous parameter signal, the method comprising the steps of:

(a) dividing the continuous parameter signal into a first plurality of discrete time intervals each of the discrete time intervals including a plurality of discrete data points;

(b) selecting a first plurality of discrete data points from predetermined ones of the first plurality of discrete time intervals, each of the first plurality of discrete data points spaced by a first predetermined time interval;

(c) representing the continuous parameter signal using the first plurality of discrete data points with a linear model;

(d) representing the continuous parameter signal using the first plurality of discrete data points with a nonlinear model;

(e) determining how well the linear model represents the continuous parameter signal by computing a performance measure of the linear model;

(f) determining how well the nonlinear model represents the continuous parameter signal by computing a performance measure of the non-linear model;

(g) comparing the performance measure of the linear model to the performance measure of the non-linear model; and

(h) in response to the performance measure of the linear model being preferred over the performance measure of the nonlinear model, then performing the steps of;

(1) selecting a next plurality of discrete data points from predetermined ones of the first plurality of discrete time intervals, each of the next plurality of discrete data points spaced by a next predetermined time interval; and

(2) repeating steps (c)-(h) until a first one of the following conditions occurs;

(i) the performance measure of the non-linear sequence is preferred over the performance measure of the linear sequence; and

(ii) a maximum time interval is reached.

26. The method of claim 25 further comprising the steps of:

selecting a sequence of data points used to represent the continuous parameter signal, each of the data points spaced by the time interval at which the performance measure of the non-linear sequence is preferred over the performance measure of the linear sequence.