| JOURNAL HOME | CME HOME | THIS MONTH | PAST ISSUES | ETOC | COLLECTIONS |
| AUTHORS | REVIEWERS | EDITORIAL BOARD | FEEDBACK | RSS | HELP |
 
|
|
|
||
|
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Within this setup, we define A as the ratio of the test signal in a motion period to a resting period. This has the effect of normalizing the absolute amplitude of the signal across the population of test subjects. We then use A as an indicator of the intensity of the motion challenge. We hypothesize that there will be some motion conditions where changes in A are small (A close to 1 meaning the motion intensity is not distinguishable from the resting intensity) with the result that all oximeters can read through the motion with little error. Conversely, we hypothesize that there will be some intensities of motion where A is very large and no oximeters will be able to read accurately. In between, we hypothesize that there are levels of activity that we can "request," which will distinguish between the capable oximeters and those that are not motion tolerant (Figs. 2–4). The range of these levels of A define the motion performance of the oximeter for each type of motion. Error was not defined by the ASTM committee—it was left to the developers to find an acceptable level of error, but generally this meant that the observed accuracy of the oximeter on a motion hand was significantly different from some reference value, typically an oximeter on a stationary hand.
|
|
|
Why do we say a different A for each type of motion? We did not want to make assumptions about the relative difficulty of one type of motion versus another. One aspect of this research is to determine which motions have independent characteristics, and thus find the minimum set of test motions necessary. We propose that characterizing the time-frequency behavior of the motion signals can give us evidence about the similarities and help reduce the number of test types or identify new test types not yet considered.
Upper and lower control limits (UCL, LCL) specify the upper and lower boundaries for the test signal being fed back to allow a tolerance band around the desired intensity. A 30% tolerance band was chosen. This produced an UCL defined as:
UCL = A x (1 + 0.3)KRavg
where KRavg is defined as the average amplitude (or whatever function we choose) of the resting IR plethysmogram measured immediately before the motion epoch. The LCL is similarly defined as:
LCL = A x (1 – 0.3)KRavg
Figure 5 shows an example of the test signal during test as an attempt is being made to keep it within the UCL and LCL, as indicated by the dotted red and white lines.
|
RESULTS
As our intent was to develop a complete test method, we started with an application (Fig. 14) that made many assumptions. It became evident after only a few subjects that we had no justification for the values we had chosen, nor could we find literature to support our position. Consultation with manufacturers produced many anecdotal recommendations, but no firm evidence upon which to develop a justifiable test method. We, therefore, began a journey to understand how our system behaved and how to set reasonable values for the parameters of our experiments. This phenomenon, ending with more questions than we started with after each new experimental protocol, repeated itself throughout our studies (and continues). For this reason, much of the work presented is shown empirically, with little group analysis performed, as new questions needed to be answered before the study could be completed. For example, after running our initial application on several subjects, we realized that the factor A was not easy to determine. This terminated the study of the motion testing until we could better understand how A was related to motion type or position within the test. It was observed that the order of motions, the durations of the periods of rest and motion, how long it takes a subject to recover back to the resting baseline, and how well a subject could keep the test signal within the control limits all needed further examination. We then decided to reexamine some basic assumptions, characterizing the electrical response of the instrumentation system and some basic physiological responses of the subject, so that we could move forward on a sound footing. The gain and position studies are reported to represent these findings.
|
Test Instrumentation and Data Collection
A method was required to measure the raw plethysmogram in real-time with sufficient resolution to implement our feedback approach. We were fortunate that two pulse oximeter manufacturers, GE/Ohmeda and Philips, make pulse oximeters available, an Ohmeda TruSat® and a Philips M3®, with a real-time raw plethysmographic output. With this equipment available, we have access to the raw IR plethysmogram (60 Hz for the Ohmeda TruSat and about 50 Hz for the Philips M3) and a computed PI at 1 Hz. It would be even more helpful to have access to both the red and IR plethysmograms, but the manufacturers are reticent to provide access to both wavelengths to prevent their calibration curves from being deduced.
The data collected were part of a Duke University IRB-approved study with all volunteers providing appropriate written consent. The Food and Drug Administration also follows a continuing review of the Duke University IRB process. The test uses a control oximeter on the middle finger of one hand to measure the raw plethysmogram and two other DUT on the index and ring fingers of the same hand. The thumb and fifth finger are not used. The saturations measured by the DUTs are compared with an oximeter on the rest hand. Failure is defined as being more than some amount different from the reference on the rest hand (10). This framework makes several important assumptions:
- Fingers on the same hand see an equivalent motion load
- Changes in the test variable are a function of the amount of motion and not a function of instrumentation
The first assumption can be tested after each trial, provided that we have appropriate tools. It is proposed to use the time-frequency characteristics of the signal. The second assumption can be validated with studies examining the response of the "system" to differing light-emitting diode (LED) excitation currents and to the position of the hand relative to the heart. We want to ensure that the selected motions impose a change in the test signal more than the changes because of relative position.
Testing and Validating Underlying Assumptions
The proposed method of analyzing the recorded signal from one "reference" finger as an indicator of motion intensity of the other fingers relies on the assumption that the plethysmogram is essentially equivalent for the middle three fingers on the same hand. More precisely, for these three fingers, we postulate the oximeters see a raw plethysmogram with comparable intensity and time-frequency characteristics. This relies on an assumption that the sampled tissue bed interacts linearly with light intensity, and that the feedback variable is independent of the level of the fingers with respect to the heart. The same techniques that we propose to use to explore the independence of the motion types can be used to draw conclusions about the similarity of the finger responses. This is the short-term frequency transform (STFT).
Before beginning to see if our test method would be feasible, we wanted to confirm the sensitivity of our analysis methods and ensure the signal detected was a result of a physiologic change and not an instrumentation gain change in the emitter or receiver circuitry. Confirmation of some basic physiologic responses to position was required to determine if these responses had a larger affect on our signal than the proposed motions. Two studies were performed: a gain study and a positional study.
Gain Studies: Response of Test Signal to Differing LED Intensities
With the help of custom software from Ohmeda, we used the TruSats in a mode where the LED emitter outputs could be controlled, thus allowing the transfer functions (output light/input light) to be measured. The subjects were resting with their hands at the level of their heart. The emitter outputs were set at auto-gain (where the oximeter chooses the levels), 8, 30, 60, and 120 mA, and the detected light measured. One study was performed using an opaque bag over only the emitters to determine normal levels of ambient light to distinguish the baseline signal from background electronic noise. The mean and standard deviation for each drive level were compared to see whether the differences in the intensity of the measured raw IR plethysmogram are proportional to the changes in emitter output.
Figure 6 shows an example of the IR plethysmogram and PI as a function of the variation of LED emitter gain as measured using the Ohmeda TruSat oximeter. Ideally, the raw plethsymogram curves should increase linearly, with the 30 mA state generating results 3.75x the 8 mA state. In turn, the 60 mA state should then generate results twice that of 30 mA, and so on. Instead, our experimental data demonstrated only a slight increase in measured light output from 8 to 30 mA. The data from 30 to 120 mA were linear within one significant digit. Additionally, as seen in Figure 6, there is also an unexpected difference in the data from two supposedly identical devices on adjacent fingers. Switching finger–probe-monitor combinations demonstrated that the differences were inherent in the devices but did not affect saturation values.
|
The linear relationship observed in the plethysmographic amplitudes at the higher excitation currents was as expected but at the lower currents, particularly 8 to 30 mA, the plethysmographic response seemed to lose its linear relationship (Fig. 7). We wanted to know if this was a result of the signal getting lost in the noise floor or the effects of ambient light. An opaque covering was applied over the emitters to determine the background signal on the detectors. The covering was intermittently removed and replaced during this test, and the results are illustrated in Figure 8. The background signal with no LED illumination is effectively zero, excluding ambient light as the cause of the nonlinearity.
|
|
A STFT frequency analysis was performed of the data in Figure 6, with spectrograms displayed in Figure 9, to demonstrate that the signals seen by two adjacent fingers were equivalent. The curves look so similar when compared visually that no additional computation was used.
|
There is a jump in the plethysmogram when the TruSat emitters were set to 8 mA from their initial auto-gain setting, thought to be because of the settling in the receiver circuitry. As the emitter gain is then increased, the spectrograms are observed to increase in amplitude, consistent with the increase in emitter drive current. At this juncture, we are unable to precisely determine why the 8 mA emitter setting causes this response, with the most likely theory being the possibility of an inherent nonlinearity in the response of the emitter diodes.
Positional Studies: Response of Test Signal to Varying Hand Position
Are the effects on our test signal of changing hand position larger than the effects we see with some of our exemplar motion types? Varying the position of the finger/probe with respect to the level of the heart provided some interesting data to address this issue. The protocol to investigate positional impact used the PI as calculated by the TruSat or M3 as a function of arm position (down, heart level, head level, up/raised). Position started at heart level and stepped through the following states: down, heart, head, up. After each state, the hand was returned to the heart position as a reference. To simplify our experimental setup, we used the same pulse oximeter on the middle finger to provide the raw IR plethysmogram. This is our control oximeter. The ring and index fingers then provide platforms for testing two additional devices. All the position and drive level studies used dual TruSat oximeters. Additional positional studies used two Philips M3 oximeters to confirm observations made with the TruSats.
The raw IR plethysmogram displayed an increasing trend with amplitude as the hand position was elevated higher toward head level with a concomitant increase in PI, apparently because of the larger AC component of the pulse, but seemed to decrease when the position was raised above the head. A sample of this response is seen in Figure 10. The upper left curve shows the raw plethysmogram acquired from a pair of Ohmeda TruSats as a function of hand position with respect to the heart (referred to as the state). The curve underneath this shows the PI as calculated by the TruSats and two prospective test signals, (max – min)/avg and AC/DC, again as a function of the position. The curve in the upper right shows a 7-s period comparing the time domain wave forms from the index and middle fingers. The curve underneath this shows the correlation between the two plethysmograms as a function of time. Underneath these curves are the two spectrograms (STFT). This response is found in both the TruSats and the M3s, but not consistently with all subjects. Both a normalized correlation coefficient calculation and the STFT spectrogram shown in Figure 10 indicate that the two oximeters on the same hand were seeing the same magnitude of activity in the IR plethysmogram.
|
For some subjects, the PI "followed" the hand position, increasing as the hand was raised, as shown in Figure 11.
|
The middle graph of Figure 11 demonstrates how the AC and DC components respond to the changes in position. From these curves, it appears that the change in the AC component is the greatest contributor to changes in PI (as measured by the TruSat), indicating that our test signal should focus on the AC characteristics. The TruSat oximeters continued to exhibit a small bias between the two monitors shown by the offset between the two signal curves in the top plot of Figure 11 but, importantly, the devices did not demonstrate an appreciable difference in measured saturation.
We continued to explore the question of how similar the finger plethysmograms are by comparing the wideband plethysmograms (the raw signal compensated for by the gain of the LEDs and the photodetector amplifier) from two fingers on the same hand. For both the TruSat and M3s, the motion generated with both fingers was synchronous, via visual assessment of an equivalent spectrogram. Figure 12, performed using the M3s, shows that, for noisy plethysmograms, the STFT is useful for comparing equivalency and that both [max – min]/avg (blue signal in the second trace) and AC/DC (green signal in the second trace) track perfusion though each can be quite noisy. Importantly, we were able to determine when the two fingers did not move synchronously, as can be seen by comparing the spectrogram in Figure 13.
|
|
DISCUSSION
We need to understand our instrumentation and the system that we were exploiting, namely the response of the tissue finger bed to light. It is important to be certain that the changes being measured were from the added motion and not from changes in underlying perfusion, possibly as a result of a change in position with respect to the heart, or a nonlinearity. Although much of what we seek is likely known to corporate research labs and anecdotally, we could not find published evidence of how the tissue bed responds to varying intensity exposure, nor could we find information about the plethysmographic response with respect to position for our particular instrumentation configuration. Rather than make assumptions, several simple experiments were devised to characterize our systems' performance. Two basic experiments were performed: 1) we exposed the tissue bed to a known intensity excitation holding the position constant, and 2) we varied the position of the hand and observed the response. In each of these experiments, we explored whether fingers on the same hand had equivalent plethysmograms, because our proposed test method uses one finger to control the amount of motion on the other two fingers. The visual representation of the STFT was used to compare finger responses with a more elaborate method for assessing similarity, based on the average difference of the point values under development.
In parallel with the continuing basic studies above, we developed an application to prototype a potential test method. Our application allows us to select the order of motions, the duration of the rest and motion periods, the desired A, and the desired control limits. We exercised this application on several volunteers, which raised several questions:
- Does the type of motion that proceeds affect the results? Would the subject fatigue for one type of motion and not another and would this influence the subject's ability to keep the signal within the control limits?
- Does the order of motion types influence the results?
- Might exercise lead to exhaustion and the inability to maintain a high intensity state (not reach the control limits)?
- Does the baseline recover to the same (resting) level? Which is the correct resting baseline to use in the calculation of A: the initial resting period, or the period immediately preceding the motion?
Figure 14 illustrates an example of the LabVIEW® based data collection system. Set-up parameters include length of rest period, length of motion period, the computation to be preformed on the raw plethysmogram (if any), and the control limits for that study. The subject performs the indicated motion, varying the intensity to attempt to maintain the computed feedback variable, K, within the range indicated by the control limits. The screen shot shows the data transition from a rest period to a motion period.
Where Do We Go from Here?
We have demonstrated the tools needed to scientifically and systematically create a motion standard for pulse oximetry. The next step will be selecting values for the parameters, using them to collect data from a representative population of subjects, and developing statistically significant conclusions. By creating a closed-loop control system, it is hoped that testing can be reproducible and will allow devices to be compared using an equivalently challenging test. To accomplish our test method, it was necessary to determine if the plethysmograms are equivalent in the fingers on the same hand, and that the signal observed is a result of the added motion and not an artifact of our measurement system, including the instrumentation and the human subject.
Some work is left to be done to select the computation of K, such that most subjects will be able to keep the test signal within the control limits. The ability of a subject to keep within the control limits is a function of how fast the system responds and the formulation of K. Our system could redraw the screen once per second, leading to a less than smooth feedback signal. Using a longer averaging filter or more points in the calculation will create a feedback variable that reacts more slowly. We instituted the concept of %IN, defined as a measure of time within the control limits compared with the time outside the control limits, when it was realized that many novice subjects had difficulty keeping their test signal within the desired range and it seemed the more experienced subjects had less difficulty. The degree to which the subject can keep the test signal within the control limits may be crucial to demonstrating the feasibility of this method. An acceptable value needs to be found, likely by experimentation and iteration.
Whether there is a critical threshold on %IN, below which the test becomes invalid, remains to be seen and is the focus of future work. It is expected from prior observations that training will improve the attainable %IN, and that %IN will also vary between motion types, as some motions are more difficult to maintain than others. We may use %IN as a quality control measure to discard invalid tests, just as the STFT may identify when the fingers are not acting in synchrony.
Proposed Protocol
Our proposed protocol will have the subject perform the motion for several increasing values of A. Preliminary experiments have shown us that A may not be the same value for all motions. An initial value of A will be set, and the subject will be asked to keep the amplitude of the generated test signal within the UCLs and LCLs for as much of the motion period as possible. At some point, the amount of motion generated with increasing A will cause failure (Af) of the tested device compared with the reference oximeter on the stationary hand. Barker and Shah (8) used the concept of percentage of time when the oximeter error exceeded a specified threshold. As additional subjects are tested for that same motion, additional values for Af will be determined. It remains to be seen how Af behaves over a population of subjects and whether it is reproducible for a subject and across subjects. With these data in hand, a fair threshold for a "pass" or "fail" can be determined. It is expected that some devices will have more difficulty with some motions than others, but a certain minimal level of performance should be mandatory to describe a device as "motion-resistant" or equivalent.
One concept for reporting the performance of each oximeter is a table format with the error measured as a function of motion type (Table 1).
|
Finally, the relative amplitudes and shapes in the STFT spectrograms we believe to be the motion signatures. We postulate that two motions with similar spectrogram signals are not independent, only one is needed in the test method. One of the next dilemmas needing to be addressed is how to validate this postulate. This will allow us to produce an efficient test, one that has a minimum amount of testing yet is still effective in differentiating motion-resistant oximeters from nonresistant ones.
CONCLUSIONS
We have proposed a method and rationale to perform motion testing that we believe will be reproducible and robust with respect to comparing motion-resistant pulse oximeter designs. Tools have been developed to explore and test the validity of the underlying assumptions, including the reliance on adjacent fingers seeing equivalent motion challenges and the independence of motion types. We believe we have identified variables that may have an influence on the ability of the test to be effective. What is left to be done is to perform rigorous, statistically valid protocols to select an appropriate form of the test statistic, optimize the duration of the resting and testing periods, and order the motions in an appropriate way. Our observations have demonstrated that the best chance for success relies on using trained volunteer subjects that can consistently control their motion and the resulting plethysmograms. Toward this end, we will seek additional opportunities to bring our agenda to completion. This work is just beginning.
Footnotes
Accepted for publication May 9, 2007.
Supported by Pooled funds donated by the Committee Members of ASTM committee F1415 on pulse oximetry, Duke University/Department of Anesthesiology fund 399-0046, and US FDA 1 year grant.
REFERENCES
- Shang AB. An investigational study to assess the performance of pulse oximeters under the presence of motion artifact. Anesth Analg 2002;94:S107
- Jopling MW, Mannheimer PD, Bebout DE. Issues in the laboratory evaluation of pulse oximeter performance. Anesth Analg 2002;94:S62–8[Web of Science][Medline]
- Severinghaus JW. History and recent developments in pulse oximetry. Scand J Clin Lab Invest Suppl 1993;214:105–11[Medline]
- Langton JA, Hanning CD. Effect of motion artefact on pulse oximeters: evaluation of four instruments and finger probes. Br J Anaesth 1990;65:564–70
[Abstract/Free Full Text] - Plummer JL, Zakaria AZ, Ilsley AH, Fronsko RR, Owen H. Evaluation of the influence of movement on saturation readings from pulse oximeters. Anaesthesia 1995;50:423–6[Web of Science][Medline]
- Tobin RM, Pologe JA, Batchelder PB. A characterization of motion affecting pulse oximetry in 350 patients. Anesth Analg 2002;94:S54–61[Web of Science][Medline]
- Gehring H, Hornberger C, Matz H, Konecny E, Schmucker P. The effects of motion artifact and low perfusion on the performance of a new generation of pulse oximeters in volunteers undergoing hypoxemia. Respir Care 2002;47:48–60[Medline]
- Barker SJ, Shah NK. The effects of motion on the performance of pulse oximeters in volunteers. Anesthesiology 1997;86:101–8[Web of Science][Medline]
- Barker SJ. "Motion-resistant" pulse oximetry: a comparison of new and old models. Anesth Analg 2002;95:967–72
[Abstract/Free Full Text] - Kästke SW, Konecny E. Determining the artifact sensitivity of recent pulse oximeters during laboratory benchmarking. J Clin Monit Comput 2000;16:509–22[Web of Science][Medline]
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
