INTRODUCTION

The use of pulse oximetry in the neonatal intensive care unit (NICU) to monitor oxygenation has been a standard of practice for many years, gaining widespread use in the mid-1980s. Studies on the accuracy and reliability of neonatal pulse oximetry have been favorable.^1,2,3,4 Although pulse oximetry oxygen saturation (SpO₂) readings have not been accurate predictors of arterial oxygen tension (PaO₂),^5,6,7,8 pulse oximetry in our NICU and in probably most NICUs remains a critical detector for desaturation and hypoxemia events, and as an oxygenation monitor during oxygen therapy and assisted ventilation.

The issue at hand developed from an index case in July 2001 in our NICU. An enterprising neonatal respiratory therapist presented the last shift's worth of blood gas and hemoximetry slips from a sick, ventilated, preterm infant, which were overwritten with corresponding SpO₂ values from the bedside monitor. The therapist commented, "The saturations really don't match. What do you want me to do about it?" A tabulation of all of the patient's arterial hemoximetry measured saturations, corresponding pulse oximetry saturation readings, and blood gas PaO₂ values demonstrated what appeared to be a systematic saturation-dependent difference between pulse and hemoximeter saturations, worse at low PaO₂ values where SpO₂ seemed to markedly overestimate the arterial measured saturation (SaO₂). We worried that some procedural issue in our use of pulse oximetry might have occurred, or that this patient was illustrating a measurement bias that we had not previously recognized. Hence, we began a systematic review of preanalytic, analytic, and postanalytic process and procedures to try and elucidate the problem. We did not wish to perform a controlled scientific experiment on pulse oximetry function, but rather sought to analyze available data recorded during the process of providing routine care, which might illustrate the operational performance of this technology.

Presented here are our findings from a historical evaluation, a prospective evaluation, a procedural evaluation, a verification evaluation, and an evaluation of corroborating data from NICUs at two other institutions.

METHODS

Saturation Bias Function

Hemoximeter saturation values can be represented as fractional oxyhemoglobin saturation (O₂Hb/THb) or functional saturation (O₂Hb/(O₂Hb+HHb)), where O₂Hb=oxyhemoglobin, HHb=deoxyhemoglobin, THb=total hemoglobin. The difference being that THb=O₂Hb+HHb+(other Hb moieties). In neonates, "other Hb moieties" such as carboxyhemoglobin or methemoglobin are uncommon. In the Verification evaluation presented below, other hemoglobin moieties constituted <3% of THb. Although elevated carboxyhemoglobin, as may occur with extreme hyperbilirubinemia, would render this approximation less accurate, bilirubin levels were not available in the collected data and so no patients were excluded on this basis.

The relation between functional saturation and fractional oxyhemoglobin in neonates can thus be approximated as: functional saturation (fractional oxyhemoglobin/0.97)=1.03 fractional oxyhemoglobin). Pulse oximetry signal processing algorithms produce a value which is equivalent to a "functional" hemoximeter saturation determination. Thus, the appropriate comparison for SpO₂ is SaO₂, the functional hemoximeter saturation. In some data sets examined for this work, the fractional oxyhemoglobin value was the one available and it was "converted" to functional saturation using the above formula.

The relation between SpO₂ and SaO₂ was evaluated by calculating the bias function (SpO₂-SaO₂) after Severinghaus et al⁹ and others,^10,11 and this was plotted against the "gold" standard, measured SaO₂. Precision was indicated by the 10th and 90th data percentiles in the range of the bias.

Pulse Oximeters and Hemoximeters

The pulse oximeters for which SpO₂ data could be provided were the stand-alone or modular units available for use at the individual hospitals, except in the Prospective evaluation, where the Nellcor N-200 was provided by one of the manufacturers. The following sensors were reported as being used with these pulse oximeters: Flex II or Oxytip+ OX-AF with Datex-Ohmeda models 3700 & 3740; Nellcor Oxisensor II, I-20 or N-25 with Nellcor model N200; Nellcor Oxisensor II, I-20 or N-25 with Space Labs; LNOP-Neo with Masimo model 2000. Similarly, the blood gas analyzer or hemoximeter used for SaO₂ measurement was the one available in the NICU for patient care, except in the Verification evaluation where the IL 682 was provided by one of the manufacturers. All sites indicated that their procedure for recording the SpO₂ value was to choose the stable baseline value just prior to initiating the arterial blood gas draw. Blood gas analysis or hemoximetry was conducted in a dedicated NICU blood gas laboratory within a couple of minutes of the blood gas draw. Quality control calibrations were performed every 8 hours. Since a calculated SaO₂ via a blood gas analyzer cannot be used as a point of comparing SpO₂ accuracy, measured, not calculated, SaO₂ was reported with the corresponding pre-blood draw baseline SpO₂. No comparison of pulse oximeter pulse rate performance was undertaken.

Historical Evaluation

An archived respiratory care blood gas data file covering the years 1991 to 1997 was de-identified and queried for all NICU records. As a matter of course, each record represented blood gas analyzer values and also contained ventilator settings, inspired oxygen fraction, SpO₂ reading (Datex-Ohmeda 3700, Madison, WI) at the time of blood gas draw, and hemoximeter (OSM-3, Radiometer, Copenhagen) measured oxyhemoglobin fraction. The queried file was then screened and cleaned for missing values and data entry errors. Functional SaO₂ was derived by converting the available arterial fractional oxyhemoglobin value, then the bias term (SpO₂-SaO₂) was calculated and regressed against SaO₂.

Prospective Evaluation

During July–October 2001, SpO₂ and SaO₂ values were collected and recorded on NICU patients during routine blood gas analysis. Pulse oximeter models included: Datex-Ohmeda 3700 & 3740, Masimo 2000 (Masimo Corp., Irvine, CA), and Nellcor N-200 (Nellcor, Pleasanton, CA). SaO₂ was measured by integrated hemoximeters in the ABL 725 (Radiometer, Copenhagen) and Chiron CIBA-Corning 865 (Bayer Diagnostics, Tarrytown, NY) blood gas analyzers. The bias function was calculated and plotted. Subanalyses were conducted by pulse oximeter type. No infants were receiving nitric oxide, which could have caused increased methemoglobin levels.

Procedural Evaluation

In November 2001, an in-depth procedural evaluation was performed in the NICU to determine whether the methods being used by staff for the selection, placement, attachment and use of pulse oximeter sensors and monitors were appropriate and correct. Observations on the timing of pulse oximeter readings in correspondence to the drawing of blood for blood gases were made. Appropriate processing of blood gas samples and use of the unit's blood gas analyzer and integrated hemoximeter were also checked.

Verification Evaluation

Also in November 2001, an IRB approved study protocol was initiated, which allowed for the drawing of a small number of duplicate arterial blood gas samples in order to verify readings between two different hemoximeters (ABL 725 and IL 682). This was performed in order to rule out a hemoximeter measurement error. The paired t-test was used to compare values between hemoximeters with statistical significance determined at the p0.05 level. The bias function was calculated for simultaneous SpO₂ readings with the Datex-Ohmeda Model 3740 pulse oximeter. No infants were receiving nitric oxide, which could have caused increased methemoglobin levels.

Evaluation of Data from Other Sites

For comparison and corroboration, colleagues at two other institutions (Primary Children's Medical Center, Salt Lake City, UT and Santa Rosa Children's Hospital, San Antonio, TX) agreed to share de-identified data collected from routine neonatal blood gas analysis and hemoximetry. The bias function was calculated and plotted for data from the two sites.

Individual Patient Analyses

Data collected for the Historical and Prospective evaluations included an unlinked patient study number. In patients where sufficient data existed that a calculation seemed feasible, individual correlation coefficients were determined. Sufficient data were defined as at least five data pairs of SpO₂ and SaO₂, and a difference between maximum and minimum SaO₂ of at least 5. For the historical data set, the last 130 patients who met these criteria were used. In the prospective data set, 28 patients met this criterion. Individual correlation coefficients were considered significant at the p0.05 level.

Operational Performance

In order to contrast the neonatal performance of the models of pulse oximeter for which data were available, operational performance was defined as the frequency with which at any given SaO₂ value the corresponding SpO₂ reading was within the technical accuracy specified for neonatal pulse oximetry. In general, technical specifications for pulse oximeters used in neonates suggest an accuracy of 3 digits (1 SD) between 70 and 100% saturation. The accuracy under 70% saturation is unspecified. Operational performance (% of SpO₂ values=SaO₂3) was determined for the Datex-Ohmeda 3700/3740, Masimo 2000, Nellcor N-200, and Spacelabs pulse oximeters using data from the Prospective evaluation and the Evaluation of data from other sites.

RESULTS

Historical Evaluation

There were 31,905 records for analysis with SaO₂ values 57 to 100%. Figure 1 shows the 10th, 50th, and 90th percentiles for the saturation bias function (SpO₂-SaO₂) versus SaO₂ for the 7-year historical data set. Although the spread in data increases with lower SaO₂, the bias function shows a significant negative correlation with SaO₂ (r=-0.67, p<0.001).

Figure 1.

Pulse oximeter bias function: 7-year historical data. Shown are the 10th, 50th, and 90th percentiles for 31,905 NICU records that compare (SpO₂-SaO₂) versus SaO₂ obtained in our NICU between 1991 and 1997. The bias function demonstrates a significant negative correlation with SaO₂, r=-0.67 (p<0.001). The bias function, (SpO₂-SaO₂), and the data variation (distance between 10th and 90th percentiles), that is, precision, worsen as saturation values decrease.

Full figure and legend (27K)

Prospective Evaluation

A somewhat stronger negative correlation (r=-0.86, p<0.001) of (SpO₂-SaO₂) with SaO₂ was seen with the 4-month prospective collected data than with the historical data. There were 566 samples available for analysis on 43 patients (see Figure 2). Correlation coefficients for the three pulse oximeters used during this period were: r=-0.85 (p<0.001) Datex-Ohmeda (Models 3700, 3740), N=278; r=-0.84 (p<0.001) Masimo (Model 2000), N=139; and r=-0.81 (p<0.001) Nellcor (Model N-200), N=149. Pulse oximeters from all three manufacturers appeared to perform similarly.

Figure 2.

Pulse oximeter bias function: 4-month prospective data. Shown is the bias function (SpO₂-SaO₂) data for three pulse oximeters plotted against SaO₂. Correlation coefficients for the pulse oximeters used during this period were: Datex-Ohmeda Models 3700 and 3740, N=278, r=-0.85 (p<0.001); Masimo Model 2000, N=139, r=-0.84 (p<0.001); and Nellcor Model N-200, N=149, r=-0.81 (p=0.001). Pulse oximeters from all three manufacturers appeared to perform in a similar fashion. Overall, 566 samples were available for analysis on 43 patients yielding a negative correlation coefficient of r=-0.86 (p<0.001). The pulse oximetry bias is seen to worsen at both high and low saturation values.

Full figure and legend (69K)

Procedural Evaluation

No systematic variances were identified in the NICU staff's performance in the selection and application of pulse oximetry sensors, in the maintenance and setting of monitors, in the recording of appropriate SpO₂ reading in conjunction with arterial blood gas draws, nor in the processing of the arterial blood sample for analysis and hemoximetry. Observations were made by respiratory therapy supervisors, outside respiratory therapy directors, and industry engineers and scientists.

Verification Evaluation

In total, 52 arterial blood samples were drawn from 10 NICU patients as part of routine blood gas monitoring and these were run simultaneously on ABL 725 and IL 682 blood gas analyzers. All SpO₂ readings were taken using the Datex-Ohmeda 3740 pulse oximeter. The average difference in hemoximeter HbO₂ paired values between the two devices was 0.07, with mean values (95% confidence limits), being 88.3 (86.4, 90.3) and 88.4 (86.9, 89.9) for the ABL 725 and IL 682 devices, respectively. The difference was not statistically significant. (SpO₂-SaO₂) was regressed against SaO₂ for these samples, yielding regression coefficients of -0.90 (p<0.001) and -0.81 (p<0.001) for the ABL 725 and IL 682 devices, respectively.

Evaluation of Data from Other Sites

At one center, NICU "A", 95 samples were provided using data from the Spacelabs pulse oximetry module (Spacelabs Medical, Redmond, WA) and an AVL Omni 1-9 (Roche Diagnostics, Basel, Switzerland) blood gas analyzer/hemoximeter. At the second center, NICU "B", 168 samples were provided using data from Datex-Ohmeda series 3700 pulse oximeters and a Radiometer ABL 725 blood gas analyzer/hemoximeter. These data are presented in Figure 3. The correlation coefficients for pulse oximetry biases in NICU "A" and "B" were: r=-0.92 (p<0.001) and r=-0.72 (p<0.001), respectively.

Figure 3.

Pulse oximeter bias function: other nicus. NICU "A" provided 95 paired SpO₂ (Spacelabs) and SaO₂ values. NICU "B" provided 168 paired values (Datex-Ohmeda 3740). The correlation coefficients for the pulse oximetry bias function for NICU "A" and "B" were: r=-0.92 (p<0.001) and r=-0.72 (p<0.001), respectively. At these sites as well, the pulse oximetry bias is seen to worsen as saturation values decrease.

Full figure and legend (28K)

Individual Patient Analyses

In the subset of the last 130 patients with sufficient data in the Historical data set, 124 (95%) demonstrated significant negative correlations between the bias function and SaO₂. The pulse oximeters used in these patients were the Datex-Ohmeda 3700 and 3740 models. For the 28 patients in the Prospective data set that met the data criteria, 20 (71%) demonstrated significant negative correlations between the bias function and SaO₂. The percentages for the three (3) pulse oximeters used in this data set were: Datex-Ohmeda 3700/3740, 70% (7/10); Masimo 2000, 60% (3/5); and Nellcor N-200, 77% (10/13).

Operational Performance

Shown in Figure 4 is the operational performance of the four models of pulse oximeter that were utilized in the three NICUs. None of the models exhibit consistent operational performance according to specifications across the range of SaO₂ of 70 to 100%. SaO₂ at peak performance varied depending on the device and ranged between an SaO₂ of 92 to 97%. Operational performance declined at SaO₂ both above and below this level. A rapid drop in operational performance was seen in all pulse oximeters as saturations decreased. For SaO₂<88 to 90%, operational performance of the pulse oximeters was less than 50%.

Figure 4.

Pulse oximeter operational performance. Operational performance was defined as the percent of pulse oximetry readings at each SaO₂ value where the pulse oximeter value is within the device-specified accuracy for neonates, that is, SpO₂=SaO₂3. Displayed here are polynomial curves fitted to the operational performance data for the four models of pulse oximeters for which data are available. Operational performance as a percentage is plotted against SaO₂. Since SaO₂3=SaO₂1SD, operational performance for the devices should be at or above 68% (horizontal line) across the range of SaO₂ of values. No pulse oximeter model shows instrument performance which is consistent except within a narrow range. Peak performance occurs between 92 and 97%, depending on device. There is declining performance both above and below this narrow range. Operational performance falls to 50% by an SaO₂ of 88 to 90% in all four models, meaning that one-half the time the SpO₂ reading deviates from the true saturation by more than 3 units of saturation percentage.

Full figure and legend (42K)

DISCUSSION

Data presented here clearly indicate a significant SpO₂ bias when pulse oximetry is used in NICU patients who require arterial access. The bias worsens as true saturation values deviate from a small range (92 to 97%). As an example, at an arterial saturation of 80%, SpO₂ will typically read approximately 90%. The bias does not appear to be procedural, or related to variation in hemoximeters, or related to variation between patients. The bias does appear to be consistent across the pulse oximeter models which were used in the three NICUs, including a new generation model with advanced low perfusion and motion detection algorithms that has recently been clinically evaluated.¹² The poor operational performance of pulse oximetry seen in these data seems at a level inconsistent with their current perceived importance, role and use in the NICU. These results suggest that in the range of SaO₂ approaching 100%, use of the pulse oximeters tested could lead to overtreatment with supplemental oxygen, and in the range of SaO₂ approaching 70%, a neonate could be under treated.

Although there are not a large number of publications that address the topic, pulse oximeter readings at low arterial saturation values have previously been reported in different patient populations to either underestimate, overestimate, or accurately represent true saturation. In healthy adult volunteers breathing sub ambient oxygen concentrations, pulse oximetry underestimated arterial saturation.⁹ For pediatric patients with cyanotic heart disease, pulse oximetry overestimated true saturations at levels below 80%.^13,14 Other reports suggest that infants and children with low saturation levels have SpO₂ values that appeared to adequately represent SaO₂.^15,16 In the neonatal population, a 1988 report by Fanconi¹⁰ analyzed 160 paired values in 20 mostly term/near-term NICU patients who had initial SaO₂< 65% using a first-generation pulse oximeter. He found that SpO₂ overestimated SaO₂ at low saturations and underestimated at high saturations. In addition, the bias did not appear to be correlated to characteristics of the patient sample (i.e., single patient, blood pressure, heart rate, temperature, or medical therapy). Our data would agree with this last observation, as nearly all of the patients who had five or more blood gases demonstrated the poor correlation. A 1989 report by Praud et al.,¹¹ analyzed 112 comparisons (60 neonates and 11 infants) of pulse oximeter and arterial oxygen saturation values where the latter ranged from 80 to 100%.¹¹ SpO₂ overestimated SaO₂ as SaO₂ decreased and the bias function (SpO₂-SaO₂) exhibited a significant negative correlation to SaO₂, r=-0.64 (p<0.01). Although the Praud et al., report does not contain arterial saturation values as low as seen in the Fanconi data or in the current data, the magnitude of the negative correlation is similar. Of concern is that the bias reported in these early papers for first-generation pulse oximeters appears to continue to be present in second- and third-generation devices as seen in our current data.

The etiology of this pulse oximetry bias is unclear. Although our Procedural evaluation did not result in any obvious explanations, there may be some not-yet understood correlation between our procedural methods and the observed bias that will require further study. Bias from patient motion artefact is commonly attributed to signal loss or corruption and often causes SpO₂ to read low rather than high. Hypoperfusion can independently also introduce a pulse oximetry bias.¹⁷ Like motion artefact, hypoperfusion will generally cause the SpO₂ to read low rather than high. Bias from motion artefact and hypoperfusion would occur in a direction opposite to that seen in the current study. Lastly, inconsistencies in readings between hemoximeters have been reported at very low saturation ranges.¹⁸ Since different hemoximeters/blood gas analyzers were used at the three sites for the current data, it is possible that part of the noted bias might be because of this type of systematic bias, and also might help explain why the slope of the bias appears to vary between the various data sets. However, within site, the bias existed regardless of the model of blood analyzer.

Inaccuracy at low saturations may be an inherent problem with the standard dual wavelength pulse oximeter when attempting to measure very low saturations, such as with fetal oxygen saturation monitoring.¹⁹ Adjustments have been proposed in emitter wavelength selection to improve device performance within these low oxygen saturation ranges.²⁰ Similar modifications may be required in order to improve performance of pulse oximetry for neonatal use. Pulse oximeters used in the NICU have historically been calibrated by industry using healthy adults. However, compared to a healthy adult, blood characteristics in peripheral tissue may be quite different in the neonatal patient with cardiopulmonary disease who requires arterial access. These conditions may render the standard calibration inappropriate. In fact, in situ calibration technology may need to be developed in order to provide patient specific calibrations that can change with the disease course. If more than two wavelength emitters become necessary in order to broaden the range of pulse oximetry accuracy, pulse spectrophotometry could become the basis for future generations of devices.²¹

This project by no means was intended to be an exhaustive study of neonatal pulse oximetry, but rather a "snapshot" of clinical performance from available clinical data. Other manufacturers and models of pulse oximeters are routinely used in NICUs and these devices may have operational characteristics different than seen in these data. Pulse oximetry experience in other NICUs may also be different. However, the question of operational performance of pulse oximetry is a very important one, and one that must be examined in much more depth by manufacturers. Isolated calibration and small quality control studies apparently do not correlate with the operational performance that clinicians must rely upon. One would also hope that the accuracy of SpO₂ could be improved to much better than 3 (1 SD). Values within 2 occurring 90% of the time across the entire 70 to 100% saturation range would be a much more clinically appropriate goal for neonatal pulse oximetry monitoring.

In summary, a discrepancy between pulse oximetry and hemoximetry derived arterial saturation measurements in an NICU patient prompted an evaluation of our pulse oximetry use and the accuracy of these readings. All data examined indicate a systematic bias in currently used dual wavelength pulse oximetry devices, with pulse oximetry generally overestimating hemoximetry arterial saturations at low saturations and underestimating at high saturations. The bias increases as saturations decrease. NICU patients who appear to have adequate arterial saturation by pulse oximetry may in fact have actual saturations which are outside of acceptable limits. Until improvements in pulse oximetry monitoring can be made that correct this measurement bias, adjustments to supplemental oxygen and ventilator settings in the NICU patient must be based on and re-evaluated by arterial blood gas analysis and hemoximetry.

References

1. Hay Jr WW, Brockway JM, Eyzaguirre M. Neonatal pulse oximetry: accuracy and reliability. Pediatrics 1989;83:717–722. | PubMed | ISI |
2. Sardesai S, Durand M, McEvoy C, Johnson C, Maarek JM. Pulse oximetry in newborn infants with birth weights of 620 to 4285 grams receiving dopamine and dobutamine. J Perinatol 1996;16:31–34. | PubMed |
3. Rajadurai VS, Walker AM, Yu VY, Oates A. Effect of fetal haemoglobin on the accuracy of pulse oximetry in preterm infants. J Paediatr Child Health 1992;28:43–46. | PubMed |
4. Zubrow AB, Henderson GW, Imaizumi SO, Pleasure JR. Pulse oximetry in sick premature infants and the effects of phototherapy and radiant warmers on the oxygen saturation readout. Am J Perinatol 1990;7:75–78.
5. Walsh MC, Noble LM, Carlo WA, Martin RJ. Relationship of pulse oximetry to arterial oxygen tension in infants. Crit Care Med 1987;15:1102–1105.
6. Thilo EH, Andersen D, Wasserstein ML, Schmidt J, Luckey D. Saturation by pulse oximetry: comparison of the results obtained by instruments of different brands. J Pediatr 1993;122:620–626. | PubMed |
7. Brockway J, Hay Jr WW. Prediction of arterial partial pressure of oxygen with pulse oxygen saturation measurements. J Pediatr 1998;133:63–66. | Article | PubMed | ChemPort |
8. Gupta R, Yoxall CW, Subhedar N, Shaw NJ. Individualised pulse oximetry limits in neonatal intensive care. Arch Dis Child Fetal Neonatal Ed 1999;81:F194–F196.
9. Severinghaus JW, Naifeh KH, Koh SO. Errors in 14 pulse oximeters during profound hypoxia. J Clin Monit 1989;5:72–81.
10. Fanconi S. Reliability of pulse oximetry in hypoxic infants. J Pediatr 1988;112:424–427.
11. Praud JP, Carofilis A, Bridey F, Lacaille F, Dehan M, Gaultier CL. Accuracy of two wavelength pulse oximetry in neonates and infants. Pediatr Pulmonol 1989;6:180–182.
12. Hay Jr WW, Rodden DJ, Collins SM, Melara DL, Hale KA, Fashaw LM. Reliability of conventional and new pulse oximetry in neonatal patients. J Perinatol 2002;22:360–366. | Article | PubMed |
13. Lebecque P, Shango P, Stijns M, Vliers A, Coates AL. Pulse oximetry versus measured arterial oxygen saturation: a comparison of the Nellcor N100 and the Biox III. Pediatr Pulmonol 1991;10:132–135.
14. Schmitt HJ, Schuetz WH, Proeschel PA, Jaklin C. Accuracy of pulse oximetry in children with cyanotic congenital heart disease. J Cardiothorac Vasc Anesth 1993;7:61–65.
15. Boxer RA, Gottesfeld I, Singh S, LaCorte MA, Parnell Jr VA, Walker P. Noninvasive pulse oximetry in children with cyanotic congenital heart disease. Crit Care Med 1987;15:1062–1064.
16. Carter BG, Carlin JB, Tibballs J, Mead H, Hochmann M, Osborne A. Accuracy of two pulse oximeters at low arterial hemoglobin–oxygen saturation. Crit Care Med 1998;26:1128–1133.
17. Gehring H, Hornberger C, Matz H, Konecny E, Schmucker P. The effects of motion artefact and low perfusion on the performance of a new generation of pulse oximeters in volunteers undergoing hypoxemia. Respir Care 2002;47:48–60. | PubMed |
18. Parath M, Sinha P, Dudenhoused JW, Lurrkus AK. Systematic instrumental errors between oxygen saturation analysers in fetal blood during deep hypoxemia. Clin Chim Acta 2001;307:151–157.
19. Mannheimer PD. Design and validation of pulse oximetry for low saturation. Anesth Analg 2002;94(Suppl 1):S21–S25.
20. Mannheimer PD, Casciani JR, Fein ME, Nierlich SL. Wavelength selection for low-saturation pulse oximetry. IEEE Trans Biomed Eng 1997;44:148–158. | Article | PubMed | ISI | ChemPort |
21. Aoyagi T, Miyasaka K. The theory and applications of pulse spectrophotometry. Anesth Analg 2002;94(Suppl 1):S93–S95.