Polygraph automated scoring systems

• US 5,327,899 A
• Filed: 01/22/1993
• Issued: 07/12/1994
• Est. Priority Date: 01/22/1993
• Status: Expired due to Term

• Current Assignee: Johns Hopkins University
• Original Assignee: Johns Hopkins University
• Inventors: Olsen, Dale E., Harris, John C.
• Primary Examiner(s): Sykes, Angela D.
• Application Number:
  US08/006,916
• Time in Patent Office: 536 Days
• Field of Search: 128/670-671, 128/668, 128/736, 128/672, 128/677-678, 128/687, 128/716, 128/734, 128/691, 128/694
• US Class Current: 600/484
• CPC Class Codes: A61B 5/0205 Simultaneously evaluating both cardiovascular conditions and different types of body conditions, e.g. heart and respiratory condition
• CPC Class Codes: A61B 5/0533 Measuring galvanic skin response
• CPC Class Codes: A61B 5/7239 using differentiation including higher order derivatives

Abstract

A polygraph automated scoring system in which polygraph signals are input and a probability of deception is output. To begin the process, digitized polygraph signals are transformed into more fundamental signals. These fundamental signals are then subjected to standardization, a critical step. The standardized signals then have certain features extracted at each question. The features for all of the relevant questions are then standardized against the features for all of the control questions. From each of the resulting standardized relevant features, the 80th percentile is taken. Finally, a logistic regression model (logit) converts this set of 80th percentile features into a probability of deception.

Claims

1. A method for automatically scoring by computer a polygraph test comprised of a plurality of relevant and control questions, the polygraph test generating a plurality of digitized physiological signals, the method comprising the steps of:

transforming the digitized signals into a plurality of fundamental signals;

standardizing the fundamental signals;

extracting a plurality of features from the standardized fundamental signals at each question;

standardizing the features for all relevant questions against the features for all control questions;

determining an 80th percentile value of each standardized relevant feature; and

combining the 80th percentile values to produce a probability of deception.

2. The polygraph scoring method as recited in claim 1, wherein the plurality of digitized physiological signals represent galvanic skin response (GSR), blood pressure and respiration.

3. The polygraph scoring method as recited in claim 2, the transforming step comprising the steps of:

detrending the GSR signal;

dividing the blood pressure signal into a high frequency portion representing a pulse and into a low frequency portion representing a blood volume;

baselining the respiration signal; and

measuring the rate of change of the blood volume signal.

4. The polygraph scoring method as recited in claim 3, the detrending step comprising the steps of:

calculating a local mean from 30 seconds of data both preceding and following each point in the GSR signal; and

removing the calculated local mean from each point in the GSR signal.

5. The polygraph scoring method as recited in claim 3, the dividing step comprising the step of passing the blood pressure signal through a Finite Impulse Response filter at four hertz.

6. The polygraph scoring method as recited in claim 3, the baselining step comprising the step of matching each low point of exhalation between breaths to a common level.

7. The polygraph scoring method as recited in claim 3, the measuring step comprising the step of taking the derivative of the blood volume signal.

8. The polygraph scoring method as recited in claim 3, the extracting step comprising the step of computing a characteristic of each standardized, fundamental signal for a specific period of time.

9. The polygraph scoring method as recited in claim 8, the computing step comprising the steps of:

calculating the range for the GSR signal, the range being the maximum amplitude minus the minimum amplitude that the signal attained during the period of time;

calculating a line length for the pulse signal, the line length being the measure of change which is sensitive to both a lot of small changes and a few large ones;

calculating the 55th percentile for the pulse signal, the 55th percentile being the amplitude that the signal was at or below during the period of time for 55% of the time;

calculating the 80th percentile for the upper respiration signal, the 80th percentile being the amplitude that the signal was at or below during the period of time for 80% of the time; and

calculating a 75th percentile for the rate of change of the blood volume signal, the 75th percentile being the amplitude that the signal was at or below during the period of time for 75% of the time.

10. The polygraph scoring method as recited in claim 9, wherein the period of time for calculating the range for the GSR signal is 11 seconds, the period beginning two seconds after the beginning of the question.

11. The polygraph scoring method as recited in claim 9, wherein the period of time for calculating the line length for the pulse signal is 8 seconds, the period beginning two seconds after the beginning of the question.

12. The polygraph scoring method as recited in claim 9, wherein the period of time for calculating the 80th percentile for the upper respiration signal is 16 seconds, the period beginning two seconds after the beginning of the question.

13. The polygraph scoring method as recited in claim 9, wherein the period of time for calculating the 75th percentile for the rate of change of the blood signal is 8 seconds, the period beginning with the beginning of the question.

14. The polygraph scoring method as recited in claim 3, the standardizing the features step comprising the steps of:

calculating the mean of the control questions; and

calculating the joint standard deviation of the control questions and the relevant questions, wherein the joint standard deviation is the variability about the mean of the control questions and the variability about the mean of the relevant questions.

15. The polygraph scoring method as recited in claim 14, wherein the standardization for each feature for the ith relevant question is computed using the formula:

##EQU4## wherein R_i is the feature value for the ith relevant question, μ

_c is the mean of the control questions, σ

_CR is the joint standard deviation of the control and relevant questions, and R_i is the standardized feature value for the ith relevant question.

16. The polygraph scoring method as recited in claim 3, wherein a logistic regression model is used to combine the 80th percentile values of the standardized relevant features.

17. The polygraph scoring method as recited in claim 16, wherein the logistic regression model has the form:

##EQU5## where;

##EQU6##

18. The polygraph scoring method as recited in claim 2, the standardizing the fundamental signals step comprising the step of selecting an interquartile range of the signals, the interquartile range corresponding to a band having a width defined by the 1st and 3rd quartiles or the 25th and 75th percentiles, respectively.

19. A method for automatically scoring by computer a polygraph test, the polygraph test generating a plurality of digitized physiological signals, the method comprising the steps of:

extracting ia plurality of features from the digitized signals; and

using a logistic regression model to compute a probability of deception from the features.

20. A method for automatically scoring by computer a polygraph test comprised of a plurality of relevant and control questions, the polygraph test generating a plurality of digitized physiological signals, the method comprising the steps of:

transforming the digitized signals into a plurality of fundamental signals;

standardizing the fundamental signals;

extracting a plurality of features from the standardized fundamental signals at each question;

using a plurality of logistic regression models to simultaneously compare the relevant questions to the control questions and combine the same feature from all questions to compute a score for each feature; and

using a final logistic regression model to combine the scores of the features to produce a probability of deception.

21. A method for automatically scoring a polygraph test comprised of a plurality of relevant and control questions, the polygraph test generating a plurality of digitized physiological signals, the method comprising the steps of:

transforming the digitized signals into a plurality of fundamental signals;

extracting a plurality of features from the fundamental signals at each question;

standardizing the features for the relevant questions against the features for the control questions;

determining a percentile value of each standardized relevant feature; and

analyzing the percentile values to produce a probability of deception.

22. The method as recited in claim 21, further comprising the step of standardizing the fundamental signals, the standardizing the fundamental signals step following the transforming step.

23. The method as recited in claim 22, wherein the analyzing step comprises the step of combining the percentile values to produce a probability of deception.

24. The method as recited in claim 23, wherein the percentile value of each standardized relevant feature is an 80th percentile.

25. The method as recited in claim 23, wherein the plurality of digitized physiological signals represent galvanic skin response (GSR), blood pressure and respiration.

26. The method as recited in claim 25, the transforming step comprising the steps of:

detrending the GSR signal;

dividing the blood pressure signal into a high frequency portion representing a pulse and into a low frequency portion representing a blood volume;

baselining the respiration signal; and

measuring the rate of change of the blood volume signal.

27. The method as recited in claim 26, the detrending step comprising the steps of:

calculating a local mean from data both preceding and following each point in the GSR signal; and

removing the calculated local mean from each point in the GSR signal.

28. The method as recited in claim 26, the dividing step comprising the step of passing the blood pressure signal through a Finite Impulse Response filter.

29. The method as recited in claim 26, the baselining step comprising the step of matching each low point of exhalation between breaths to a common level.

30. The method as recited in claim 26, the measuring step comprising the step of taking the derivative of the blood volume signal.

31. The method as recited in claim 26, the extracting step comprising the step of computing a characteristic of each standardized, fundamental signal for a specific period of time.

32. The method as recited in claim 31, the computing step comprising the steps of:

calculating the range for the GSR signal, the range being the maximum amplitude minus the minimum amplitude that the signal attached during the period of time;

calculating a line length for the pulse signal, the line length being the measure of change which is sensitive to both a lot of small changes and a few large ones;

calculating a percentile for the pulse signal, the percentile being the amplitude that the signal was at or below during the period of time for a portion of the time equal to the percentile;

calculating a percentile for the upper respiration signal, the percentile being the amplitude that the signal was at or below during the period of time for a portion of the time equal to the percentile; and

calculating a percentile for the rate of change of the blood volume signal, the percentile being the amplitude that the signal was at or below during the period of time for a portion of the time equal to the percentile.

33. The method as recited in claim 23, wherein a logistic regression model is used to combine the percentile values of the standardized relevant features.

34. The method as recited in claim 22, the standardizing the fundamental signals step comprising the step of selecting an interquartile range of the signals.

35. The method as recited in claim 34, wherein the interquartile range corresponds to a band having a width defined by the 1st and 3rd quartiles or the 25th and 75th percentiles, respectively.

36. The method as recited in claim 21, the standardizing the features step comprising the steps of:

calculating the mean of the control questions; and

calculating the joint standard deviation of the control questions and the relevant questions, wherein the joint standard deviation is the variability about the mean of the control questions and the variability about the mean of the relevant questions.