Introduction - Penn Engineering



PENN BIOENGINEERING

LIE DETECTOR

-ENIAC @ 11:45pm 12/6/98

BE 309

FINAL PROJECT

Fall 1998

Group T3

Anastasios Argeros

Eric Brahin

Mike Dolan

Jenny Li

Bryan Wells

ABSTRACT

The experimental goal of building a “lie detector”; an experimental apparatus consisting of a variety of sensors that would monitor the physiological changes in an individual when he/she is lying was accomplished. The capability of distinguishing between a resting vs. nervous; heart rate, galvanic skin response, degree of fidgeting and respiratory capacity of 102.1 vs. 76.9 bepm (beats per minute), |1.421| Volts vs. |0.172| Volts, |0.3| Volts vs. |0.06| Volts, and |1.113| Volts vs. |0.366| Volts, respectively is indicative of the realization of the first goal (see figures in appendix). A secondary goal of developing a formula-based approach of analyzing sensor data in an attempt to state with statistical accuracy the probability that an individual lied was not accomplished. It was concluded that the more practical and feasible manner of analyzing the data by direct observation of the EEG readouts is better suited for the purposes of this experiment.

Background

Biological Background

Many psychological as well as medical studies have been done on physiological responses to lying. These mental responses are products of the response of the central nervous system to stress and anxiety caused when a lie is told. It is proposed that when an individual under duress must lie, that his heart and respiratory rates increase while respiratory capacity and skin conductance vary. Specifically, skin conductance is considered a function of the sweat gland activity as controlled by the sympathetic nervous system. When a subject is startled or experiences anxiety, there will be a fast increase in the skin's conductance (a period of seconds) due to increased activity in the sweat glands (unless the glands are saturated with sweat.) After a startle, the skin's conductance will decrease naturally due to reabsorption. Sweat gland activity increases the skin's capacity to conduct the current passing through it and therefore changes in the skin conductance reflect changes in the level of arousal in the sympathetic nervous system.

Another proposed physiological response to lying is a slight variance in body temperature. Under stress muscles tense and blood flow is restricted to the main body, away from the body’s extremities. A final physiological response that might be indicative of an individual’s tendency to lie is a function of the person’s overall tension. If a person tenses up when lying he/she may relieve such tension through simple “fidgeting”. Measuring the tendency and degree with which a person moves around when answering questions is therefore another means of lie detection.

Methods of Data Analysis

The signals recorded by the various sensors fall into two categories: those to be analyzed by their frequency or amplitude. For example signals recorded by respiration rate and heart rate will be analyzed by changes in frequency whereas signals recorded by the temperature change, galvanic skin response, and the two respiratory rate sensors will be analyzed by changes in amplitude. It should be noted that if an individual is apt to sit still under questioning then the movement signal should be analyzed in terms of amplitude (i.e. the occurrence of movement is taken to be a deviation from the norm and therefore indicative of tension). If the individual tends to move constantly then changes in the frequency with which the individual moves will be analyzed.

A mean response value for each sensor will be obtained by averaging the response values (i.e. frequency or amplitude) from the responses to the "baseline" questions. The standard deviation from the mean will be determined. As a general rule the standard deviation will be the criterion used to determine the accuracy of the sensors. For example if for a known lie the signal varies by at least 2 standard deviations from the baseline value then the sensor would be deemed accurate. In order to determine the overall accuracy of each sensor a series of trials where the subject is told to lie in response to a set number of baseline questions (e.g. 2 out of 10) is conducted. The signal outputs for each sensor will be checked to see if the "lies" are indicated by 2 standard deviations. The subject will then reveal which responses were lies. Over the series of trials the accuracy with which a sensor indicates a lie indicated by a percentage. The total accuracy with which the experimenters can claim the subject is lying is then given by the following formula.

Probability of a lie = Sum (ni*ki) / N (1)

Where,

-n is the accuracy, determined experimentally, of the given sensor.

-k is either 1 or 0; if the sensor signal indicates a signal deviation from

of 2 or greater from the mean value then k = 1 otherwise k = 0.

-N is the total number of sensors (7).

-i indicates the given sensor, i = 1 through 7.

METHODS and MATERIALS

The following is a description of the sensors used in lie detection.

1) Sensor 1 measures: Heart Rate

-This sensor will measure the heart rate of the subject.

-By using the procedure from the EEG lab (experiment 1) the subject’s heart rate was recorded (3). Any changes in the heart rate, which may have been attributed to stress, were recorded.

Sensor 1 Materials:

-2 EEG “Red dots”

-EEG leads from Red dots into the patient isolator

-Conductive gel

2 & 3) Sensors 2 and 3 measure: Respiratory Capacity.

-The respiratory capacity circuit set-up consisted of the following circuit (see Figure #1). Two strain gauges were mounted on separate pieces of flexible metal, which in turn were fixed relative to one another with another piece of metal. The entire apparatus was mounted onto the subject by means of two belts of thick rubber bands that circled the subject’s chest and stomach (see Figure #2).

-When the subject breathes in (i.e. the chest cavity or the abdomen expands) this will cause a stretching of the strain gauge and therefore a change in the strain gauge resistance which will in turn vary the voltage output.

Figure #1 (1)

Figure #2

Sensor 2 & 3 Materials:

-620 AD OpAmp chip

-10K ( and 100K ( variable resistors

-100 ( resistor

-3 120 ( resistors

-Copper wire & tape

4) Sensor 4 measures: Galvanic Skin Response.

-Electrodes with conductive gel (i.e. as used in EEG experiment) were placed on the subject’s palm in close proximity to one another (approx. 2 cm apart). The Wheatstone bridge shown below (Figure #3) was used in the galvanic skin response to magnify the voltage change across the skin. A small 1.5 V battery supplied the voltage.

Figure #3 (3)

-In theory, the resistance of the skin should change as the subject ‘lies’ (see background). The subject will perspire and this will lead to a decrease in the resistance of the skin, due to the presence of salts in sweat.

-This voltage signal was processed by sending two inputs (points 2 and 4 on Figure #3) to the EEG. The output signal will ‘spike’ when a change occurs in the resistance of the skin (i.e. the subject tells a lie).

Sensor 4 Materials:

- EEG leads from Red dots on palm to patient isolator

- 3 M( resistors

- Bench copper wire

- 2 “Red dot” EEG skin pads

- 1.5 Volt battery

5) Sensor 5 measures: Body Temperature

-A thermistor records changes in the temperature of the subject. The body temperature sensor circuit is shown in Figure #4. The thermistor circuit includes a voltage source (1.5 V battery) and relays to the EEG machine.

-The thermistor was placed in the subject’s ear.

Figure #4 (4)

Sensor 5 Materials:

-Thermistor

-Bench copper wire

-10 Kilo-ohm resistor

-1.5 Volt DC battery

-Tape (lots of it needed to keep thermistor in the subject’s ear)

6) Sensor 6 measures: Respiratory Rate

-A circuit with a condenser microphone was designed as shown in Figure #5. The microphone was placed in the subject’s nasal cavity and the output signal was sent through the EEG machine. Every time the patient breathed, the resistance of the microphone (a variable resistor) changes and the output voltage sent to the EEG machine also changes.

Figure #5

Sensor 6 Materials:

-Resistors (as specified in Figure #5)

-0.1(F Capacitor

-Microphone

-Bench copper wire

-1.5 V DC battery

-Again, lots of tape

7) Sensor 7 measures: Movement/Fidgeting

-Using a foam pad connected to a battery (i.e. variable resistor), this sensor measured pressure changes. The pressure-to-voltage circuit is shown below in Figure #6.

-While the subject is sitting on the pad (see Figure #7), if he should start to ‘fidget’, the sensor is able to record those changes in body position as changes in voltage, which will be sent and displayed on the EEG readouts.

Figure #6 (4)

Figure #7

Sensor 7 Materials:

-Pressure pad

-Resistor as specified in Figure #6

-1.5 Volt DC battery

General Experimental Materials Needed

-EEG machine (see Figure #8)

-Input/output coax cables (running in and out of EEG machine)

-Solderless circuit boards

-LabView (data logger)

Figure #8

PROCEEDURE

Week one

The first lab period was devoted to building all of the circuits and deciding how each sensor was to be attached to the subject. After the circuits were completed and tweaked, they were tested separately using multimeters (i.e. not attached to a human subject) to verify that the sensors themselves produced varying output signals in response to different conditions (e.g. the voltage output of the movement/fidget sensor changed when pressure was exerted on the pressure plate).

Week two

Labeling the seven sensors and wiring them into the EEG machine assembled the PENN Bioengineering Lie Detector. Next the group connected the outputs of the EEG machine to the inputs of Labview in order to collect data. One of the group members then volunteered to be “wired” with the sensors. Every sensor that involved touching the skin of the subject was wired through the patient isolator for safety. Once wired, each sensor was systematically calibrated in order to ensure that each signal could be completely recorded on the graph paper. This entailed having the subject make their own physiological responses such as breathing deeply, moving in the seat, etc. The proper gains for each circuit were set according to the sensitivities of each sensor and the proper sensitivity and filters on the EEG machine were selected. After each sensor was individually set, they were all tested simultaneously to make sure adjacent pens on the EEG machine functioned properly and a “readout” of all the signal outputs was produced.

Week 3

Week three involved the actually testing of the lie detector. Five “tests” were conducted and the ink pens on EEG machine and LabView recorded the output signals. The five tests were conducted following the establishment of a resting state.

-Test #0 - Establishment of Resting State Responses

-This portion of the lab consisted having the subject sitting in the chair for a few minutes while the resting state physiological responses were recorded.

-Tests #1 & #2

-These two tests entailed asking the subject 10 simple yes/no questions. The subject was told to “lie” to any two of the questions. The questions were very basic and straightforward (e.g. “Is your name ---- “ and “Do you go the U. of Penn”, etc.,)

-The interrogator was careful to allow 10-15 seconds between questions so that the time of question and response could be clearly recorded on the EEG readout.

-Test #3 – Establishment of Excited State Responses

-In order to prove all sensors were operating properly (i.e. actually recording distinct changes in physiology), the subject was asked to get physically active (i.e. do push-ups and jumping jacks).

-The physiologically active subject was then wired to the machine and the various sensors recorded his responses.

-Test #4 & #5 - Shocking and Unexpected Questions

-These two tests resemble tests #1 & 2. The difference is that the nature of these questions was left unknown to the subject. The questions were explicit and personal in nature.

-Test #4 consisted of mildly “shocking” questions whereas in Test #5 the most shocking of questions were asked.

RESULTS

The output signals corresponding to the 6 measured physiological characteristics of the test subject in a state of rest are displayed below in Figure #9. The average frequencies and amplitudes of the signals are also listed.

Figure #9

The value for heart rate of 1.066 Hz is equivalent to ( 64 bepm (beats per minute) which is very close to the rough estimate made of 62 bepm on the initial EEG readout (please see page 1 of the presentation handout). The respiratory rate of 0.343 Hz corresponds to ( 20.6 brpm (breaths per minute) which also validates the initial estimate of 17 brpm on the readout (again see 1st page of the presentation handout). The discrepancies in these values are attributed to the required approximation of the time interval from the readout. In Excel the time interval over which a certain number of peaks occur can be determined exactly.

It should be noted that the average voltage peak values, as listed in the legend of Figure #9, where determined by manually sifting through the data in order to find the max/min voltage values. While it may have been a goal of this experiment to quantify the data in such a manner that max/min amplitude values were obtained as a result of a formula applied to the Excel data, this proved impractical (see discussion).

The output signals corresponding to the 6 measured physiological characteristics of the test subject in a state of excitement are displayed below in Figure #10. Again, the average frequencies and amplitudes of the signals are also listed.

Figure #10

The excited heart rate as calculated from Figure #10 is 1.55 Hz or 93 bepm. The respiratory frequency is 0.383 Hz or equivalently 23 brpm. Again these values are more exact then the rough calculations (of 92 bepm and 28 brpm respectively) made directly from the EEG readout during the time of the experiment because of the accuracy with which the time domain can be measured in both cases. It should be noted that the excited frequencies and average amplitudes are larger as compared to the resting results.

The remaining results correspond to those obtained during our 5th round of questioning the subject. (See the methods section as to the nature of this round of questioning and the discussion session as to the rationale behind it.) Figures #11 and #13 in the appendix display the heart rate and galvanic skin response signals, respectively. The time domain of these graphs corresponds to just prior to and immediately following the subject being asked the first of ten questions. This period surrounding question #1 of test #5 has been identified as a period of anxiousness (see discussion). Figures #12, #14 and #15 display the signal outputs of the galvanic skin, movement, and respiratory rate sensors respectively. The time domain in these latter graphs corresponds to the asking of questions #5 and #6; a period that has been identified as one of shock (see discussion).

DISCUSSION

Discussion of Experimental Development

A goal of this experiment was to develop sensors that would monitor discreet changes in a subject’s physiology and. The manner in which this goal was realized is outlined in the discussion that follows.

As discussed in the procedure, the subject was initially asked 10 straightforward questions and told to lie in answering 2 of the questions. Two trials of this form of questioning were completed (i.e. Tests #1 & #2) and the readouts from the EEG were analyzed immediately. The readouts showed that there was no noticeable difference in the output signals (in terms of frequency or amplitude) during the period that the subject was lying and telling the truth. This lack of differentiation in signal responses was a result of a) sensor incapability in detecting changes in physiology or b) a lack of physiological change in the subject when he lies or tells the truth.

In order to test the sensor capabilities the experimenters decided to monitor the subject in an excited physical state (i.e. Test #3). The sensors picked up increased breathing and heart beat frequencies as well as greater respiratory capacity (compare Figures #9 & #10). The other physiological responses (except body temperature) also showed increased activity (see Figure #10). From these observations the experimenters concluded that the sensors were capable of monitoring changes in physiology and that the lack of change in the signals was in fact due to a lack of physiological change in the test subject. It was concluded that the circumstance under which the subject was being interrogated was not adequate for producing noticeable changes in the subject’s physiology.

Throughout the experiment it has been assumed that the changes in physiology that occur when a person lies are due to the individual’s concern with lying. Since the subject in the lab knew that there would be no consequence to lying, that he was being encouraged to lie for the sake of the experiment, the experimenters were wrong in assuming that noticeable physiological changes might be detected. The problem was that a tense atmosphere under which the subject might feel pressure or anxiety in telling the truth could not be duplicated in the laboratory setting. It was proposed that the relevant physiological responses such as “sweaty palms”, changes in breathing rate and capacity, changes in heart rate, etc., could be induced by shocking, surprising, or embarrassing the subject with the questions being asked.

The initial trial involving mildly explicit questions (i.e. Test #4) revealed that the subject was not a terribly sensitive person; meaning that the questions were not “shocking” enough and that the results of Test #4 mirrored those of Tests #1 & #2. Due to time constraints only one more trial was conducted. In Test #5 the questions asked were sufficiently “shocking” (refer to the Appendix for graphed results, Figures #11 through #15) (see page 4 of the presentation handout).

In anticipation of the ‘shocking’ questions that he was to be asked the subject tensed prior to question #1. This tensing was due to anticipated embarrassment induced by the nature of the questions to be asked. Figure #11 and Figure #13 refer to the time period just before and just following the asking of question #1. Prior to the question being asked the subject’s heart rate was 102.1 bepm (more precise then the rough calculation of 75 bepm made on page 4 of the presentation handout). The substantial peak displayed in Figure #5 is a result of the increased conductivity of the subject’s skin, interpreted as an indicator of the increased tension the subject was putting himself under (see page 4 of the presentation handout). It should be noted that the first question happened to be “Is your name _ _ _ _?” While the subject was anticipating a very personal question the rather mundane nature of the question caused him to partially relax. The subject’s relaxation can be seen again in Figures #11 & #13, where the pulse rate significantly decreases to 76.9 bepm and the galvanic skin response resumes a more constant signal pattern. This anxiousness or nervousness that the sensors signals seem to indicate during this time period were confirmed by the subject when he was asked to describe (following the question period) what he was feeling at that time.

The signal outputs during the time period corresponding to the asking of questions #5 and #6 of Test #5 were analyzed in the same manner as above (also refer to page 4 of the presentation handout throughout the following argument). The galvanic skin response shows a pattern of greater variance in skin conductance following the asking of question 5 (see Figure #12). The movement signal also shows that the subject (who was normally prone to sitting still) fidgeted significantly after having been asked question #5 (see Figure #14). Finally, the subject is shown to have taken a deep breath just following question #6 (see Figure #15). Questions #5 and #6 proved to be extremely shocking in nature. Figures #12 & #14 can be interpreted as evidence that the subject was caught off guard by question #5; the subject’s hands begin to sweat and he fidgets ever so slightly while responding. Figure #15 can be taken to show that in an effort to compose himself slight in answering a second shocking question (namely question #6) the subject takes a slightly deeper breath.

Further evidence that the sensors were fully capable of monitoring slight changes in physiology due to induced states of shock or surprise came rather unexpectedly following Test #1. While the subject was resting (sleeping), the group was discussing ways to get the subject nervous. Kidding around, a lab member said that a gun should be taped to the subject’s head, and if he lies, the gun would discharge. Even though the subject was not awake, there is a sharp increase in the lung capacity for the breath exactly corresponding to the time of the question (see page #2 of the presentation handout). There was also a quick and sharp temperature peak, indicating a change in the temperature of the subject. In all of the tests performed, this was the only time a temperature change was recorded. It is also interesting to note that this happened when the subject was in a semi-conscious state.

Test #5 showed that the sensors could, indeed monitor slight changes in physiology as a result of an induced type of stress (i.e. embarrassment, shock, anxiousness, etc.,). Since it was proposed that these physiological changes are closely related to those an individual might feel when mindful of lying, the experimenters felt that the developed experimental set-up could be used as a “lie-detector” given the proper circumstances.

Discussion of Data Analysis

A second goal of the experiment was to develop a sound (meaning numerical and formula-based) method of analyzing the recorded data so that the likelihood of a “lie” could be stated with statistical accuracy (see background). This goal unfortunately was not accomplished, since the sensor signals, recorded as a series of discreet data values, are difficult to quantify. The experimenters were forced to rely on the observational approach (as detailed in the above discussion section) as the most feasible means of determining whether or not a person is lying. The reason for this decision is outlined in the discussion that follows.

Initially the experimenters believed that the signals corresponding to galvanic skin response for example would be mostly constant except when a voltage spike occurred. This spike would correspond to a significant increase in the subject’s skin conductance (and therefore thought to result from the telling of a lie) if the amplitude of this spike was at least 2 standard deviations greater then the mean signal amplitude. (The mean signal amplitude corresponds to the state during which the subject is telling the truth.) In Figures #9 & #10 it is plainly obvious that the galvanic signal actually consists of many small voltage peaks that follow a general trend in forming larger peaks. It is these larger peaks that, for analysis purposes, the experimenters are interested in. The smaller peaks are considered insignificant or due to noise. Unfortunately, with Excel it is difficult to distinguish between the large and small voltage peak values. This distinction can be made manually by plotting the points (see Figures #9 and #10) but this is impractical when a file of 20,000+ data points must be analyzed. In any case this type of analysis was attempted where the averages and standard deviations of the amplitudes (frequency intervals for heart rate and breathing rate) were determined. The reliability of each sensor was then determined by checking that the sensor produced a signal in the expected amplitude (or frequency) range when it was supposed to (i.e. during a time when the subject was told to lie or tell the truth). A sample table of the sensor reliabilities from Test #4 is shown below as Table #1.

Table #1

The reliability of each sensor was averaged in order to determine the overall reliability with which the system can predict a lie. Ultimately the reliability of each sensor is supposed to be worked into Formula #1 (see background) and in looking at the output signal data for each sensor the probability that the subject lied in answering a question can be determined. The low reliability values are attributed to the fact that the mean and standard deviation values (i.e. the criterion for measuring reliability) are not representative of the true values since the smaller (noise peaks) are averaged with the larger (true signal peaks). Since with the available data analysis techniques a sensor reliability of no better then 50% could be established (might as well flip a coin to determine whether or not the subject lied) it was determined that a statistical approach to analyzing the data was not practical. It is the observational approach to analyzing the EEG readouts that the experimenters deemed most effective in indicating the physiological state of the subject at any given time during the questioning.

It should be noted however that the frequency-dependent signals (heart and respiratory rates) have a substantially higher reliability factor. This is because peak amplitude is not important, only the presence of repeating peaks over a given time period. The loss of 20% reliability in both of these two sensors is attributed to the poor circumstances under which questioning was conducted throughout the experiment. During Test #4 there was little difference in the sensor signals during the two supposed periods of “lying” versus the period of “truth telling” because of the poor question content and lack of consequence to lying (see Experimental Development section of this discussion).

As a final note, due to the apparent lack of noticeable variation in body temp (no matter what the circumstances) the experimenters conclude that the body temperature sensor (i.e. thermistor) should be removed from the experimental set-up.

APPENDIX

Figure #11

Figure #12

Figure #13

Figure #14

Figure #15

REFERENCES

1) Be 209 Bioengineering Laboratory I, Laboratory Manual. Fall 1997 ed.

2) Be 210 Bioengineering Laboratory II, Laboratory Manual. Spring 1998 ed.

3) Be 309 Bioengineering Laboratory III, Laboratory Manual. Fall 1998 ed.

4) Radio-Shack Booklet, Sensor Projects. As provided by Al Giandomenico.

-----------------------

[pic]

[pic]

[pic]

[pic]

[pic]

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download