We present a noninvasive method to measure the venous oxygen saturation (SvO2) in tissues using near-infrared spectroscopy (NIRS). This method is based on the respiration-induced oscillations of the near-infrared absorption in tissues, and we call it spiroximetry (the prefix spiro means respiration). We have tested this method in three piglets (hind leg) and in eight human subjects (vastus medialis and vastus lateralis muscles). In the piglet study, we compared our NIRS measurements of the SvO2(SvO2-NIRSresp) with the SvO2 of blood samples. SvO2-NIRSresp and SvO2 of blood samples agreed well over the whole range of SvO2considered (20–95%). The two measurements showed an average difference of 1.0% and a standard deviation of the difference of 5.8%. In the human study, we found a good agreement between SvO2-NIRSresp and the SvO2 values measured with the NIRS venous occlusion method. Finally, in a preliminary test involving muscle exercise, SvO2-NIRSresp showed an expected postexercise decrease from the initial baseline value and a subsequent recovery to baseline.
- tissue spectroscopy
- pulse oximetry
- hemoglobin saturation
the possibility of using light to measure the oxygen saturation of hemoglobin in vivo has been explored since the 1940s (37). The feasibility of optical blood oximetry stems from the oxygenation dependence of the optical spectrum of hemoglobin. This is illustrated in Fig.1, which shows the absorption spectra of 100 μM hemoglobin for oxygen saturation values of 0, 20, 40, 60, 80, and 100%. The spectra of Fig. 1 were calculated from published values of the molar extinction coefficients of oxyhemoglobin (HbO2) and deoxyhemoglobin (Hb) (43, 53).
Oxygen saturation of the pulmonary capillary blood in rabbits has been measured by using dynamic invasive techniques (48). Near-infrared light in the wavelength range from 700 to 900 nm results in a sufficient penetration depth for the noninvasive optical monitoring of skeletal muscle, cerebral gray matter, and breast tissue. As a result, near-infrared techniques allow a noninvasive assessment of hemoglobin saturation for a wide range of applications, such as the study of muscle metabolism (7, 9, 12, 29, 45), the diagnosis of vascular disorders (2, 20, 32, 33, 44, 49), functional brain imaging (3, 10, 24, 30, 35, 50), and breast cancer detection (23, 28, 40, 42, 46).
If near-infrared light is highly sensitive to the oxygen saturation of hemoglobin, then its large penetration depth inside tissues implies that the arterial, venous, and capillary compartments all contribute to the optical signal. The average hemoglobin oxygenation measured with near-infrared spectroscopy (NIRS) (19, 34, 41) is usually referred to as tissue oxygen saturation (StO2). StO2 values are assumed to be in between arterial and local venous saturation values (SaO2 and SvO2, respectively). A number of research studies have investigated the relationship between the near-infrared (noninvasive) measurement of StO2 and the values of SaO2and local SvO2 measured invasively from drawn blood samples (31, 51). The contribution of the arterial compartment to the noninvasive optical signal can be isolated because of its unique temporal dynamics associated with the systolic-diastolic blood pressure variation at the heartbeat frequency (1). The pulsatile component of the optical signals at two or more wavelengths at the heartbeat frequency is used by conventional (1, 36) or self-calibrated (21) pulse oximeters to measure the SaO2. SaO2is a parameter that provides information about the ventilation and the oxygen exchange in the lungs. In contrast, SvO2 is a parameter that reflects the local balance between blood flow and oxygen consumption. The noninvasive optical measurement of SvO2 is complicated by the fact that the isolation of the contribution of the venous compartment to the noninvasive optical signal is not straightforward. There are no clinical devices presently capable of monitoring SvO2 noninvasively.
A number of experimental approaches have been proposed to measure SvO2 from induced local changes in the venous blood volume. For instance, proposed approaches involve a venous occlusion in a limb (13, 39, 55, 56), tilting the patient's head down by 15 degrees (47), a partial jugular vein occlusion (15, 54), or mechanical ventilation (52). In all these approaches, SvO2 is optically measured as the ratio between the increases in the HbO2 concentration ([HbO2]) and the total hemoglobin concentration (equal to [HbO2] + [Hb], where [Hb] is deoxyhemoglobin concentration) induced by the local increase in venous blood volume. To overcome the limitations of these methods, which can either be applied only to the limbs (venous occlusion method) or require an external perturbation (partial jugular vein occlusion, mechanical ventilation, and tilting methods), we propose an alternative approach that is an extension of the method of Wolf et al. (52). This approach involves no external perturbations and is applicable to subjects who are breathing either spontaneously or synchronously with a metronome set at their average respiratory frequency. Furthermore, this method can provide continuous and real-time monitoring of SvO2. The basic idea is to measure SvO2 from the amplitude of the optically measured [HbO2] and [Hb] oscillations at the respiratory frequency. The basic hypothesis, originally formulated in this context and tested on the brain of mechanically ventilated infants by Wolf et al., is that the oscillatory components of [HbO2] and [Hb] at the breathing rate are mostly representative of the venous compartment. Because the venous compliance is ∼20 times as large as the arterial compliance (4), a given change in the blood pressure in the veins causes a venous volume change ∼20 times as large as the arterial volume change corresponding to the same pressure change in the arteries. During normal breathing, the inspiration phase involves a decrease in the intrathoracic pressure and an increased pressure gradient between the peripheral venous system and the intrathoracic veins. This causes blood to be drawn from the extrathoracic veins into the intrathoracic vessels and heart (26). Because of the vein valves, venous return is increased more by inspiration than it is decreased by expiration (38). The net effect is the so-called respiratory pump that facilitates the venous return from the periphery by the respiration-induced periodic fluctuations in the central venous pressure (38). As a result of the respiratory pump, the peripheral venous blood volume oscillates at the respiratory frequency, decreasing during inspiration and increasing during expiration.
It is on this oscillatory component at the respiratory frequency that we base our near-infrared measurement of the SvO2. We coin the term spiroximeter to indicate an instrument for measuring the SvO2 from respiration-induced oscillations in the venous blood pressure and in the venous volume fraction in tissues. It must be observed that respiration may also induce perturbations to the heart rate (respiratory sinus arrhythmia) and consequently to the cardiac output and arterial blood pressure. As a result, the arterial compartment volume may, in general, also oscillate at the respiratory frequency; thus near-infrared spiroximetry data must be carefully examined to guarantee a reliable reading of SvO2.
We report a validation study conducted on the hind leg of three piglets, in which we compared the near-infrared measurements of SvO2 (SvO2-NIRS) with the SvO2 values obtained by the gas analysis of venous blood samples (SvO2-blood). To show the applicability of spiroximetry to human subjects, we also conducted a preliminary test on the vastus medialis and vastus lateralis muscles of healthy volunteers at rest and postexercise.
MATERIALS AND METHODS
The near-infrared measurements were performed with a frequency-domain tissue spectrometer (model 96208, ISS, Champaign, IL) (18,25). This instrument uses two parallel photomultiplier tube detectors that are time shared by eight multiplexed laser diodes emitting at 636, 675, 691, 752, 780, 788, 830, and 840 nm, respectively. The frequency of intensity modulation is 110 MHz, and heterodyne detection is performed with a cross-correlation frequency of 5 kHz. The multiplexing rate, i.e., the frequency of sequential laser switching, is 100 Hz. As a result, 50 cross-correlation periods are acquired during the on time of each laser diode, and a complete acquisition cycle over the eight wavelengths is completed every 80 ms. The laser diodes and the photomultiplier tubes are all coupled to fiber optics. The eight individual illumination fibers, each 400 μm in internal diameter, are arranged into a fiber bundle having a rectangular cross-section of 3.5 × 2.0 mm2. The collecting circular fiber bundles are 3.0 mm in internal diameter. The optical fibers are placed in contact with the skin by means of a flexible plastic probe. The optical probe arranges the tips of the illuminating and collecting fiber bundles along a line, with the two collecting fiber bundles at distances of 1.0 and 2.0 cm from the single illuminating bundle. In some cases, we have used a second tissue spectrometer to perform simultaneous measurements on both legs (piglets 2 and 3) or at different locations on the same leg (human subjects). In the second tissue spectrometer (which used the optical probes PL and HVL defined below), the 840-nm laser diode was replaced by a laser diode emitting at 814 nm.
Measurements on piglets.
We performed measurements on three piglets that were 15 ± 1 days old and weighed 5 ± 1 kg. The experimental arrangement for the piglet measurements is schematically illustrated in Fig.2. The piglets were anesthetized by inhalation of 3–4% isoflurane administered by means of a breathing mask applied to the piglet's snout. The animals were not mechanically ventilated, and they breathed freely throughout the experiment. A strain-gauge belt (Sleepmate/Newlife Technologies, Resp-EZ) was placed around the piglet's thorax to continuously monitor the respiratory excursion. A pulse oximeter (Nellcor, N-200) continuously recorded the heart rate at the foot of the right hind leg. The analog outputs from the strain gauge and the pulse oximeter were fed to the auxiliary input ports of the tissue spectrometer for continuous coregistration of optical and physiological data. A femoral cutdown was performed into the left inferior femoral vein to insert a catheter for periodic blood sampling. The femoral venous blood samples were run through a commercial blood-gas analyzer (Instrumentation Laboratory, model 1304 pH/blood-gas analyzer) to obtain invasive readings of SvO2-blood. One optical probe (identified as probe PR) was always located on the right (noncatheterized) hind leg. In piglets 2 and 3, a second probe (probe PL) was placed on the catheterized (left) leg. The protocol consisted of varying the femoral SvO2 over the approximate range of 20–95% by modulating the volume fraction of oxygen inspired by the piglet (Fi O2) over the range of 10–100%. The oxygenation cycles performed on the three piglets are illustrated in Fig. 3. Each cycle consisted of varying the Fi O2 approximately every 4–6 min through the values of ∼40, 15, 10, and 100% (piglets 1 and 2) or ∼40, 20, 17.5, 15, 12.5, 10, and 100% (piglet 3). We performed two Fi O2 cycles on piglet 1, four on piglet 2, and three on piglet 3. For each specific value of Fi O2, we acquired about 3,000 optical data points [4 min × (60 s/min)/(80 ms/data point)] or more. During cycles C and D onpiglet 2, the optical probe PR was slightly moved with respect to the location examined during cycles A andB, to collect data on two different muscle volumes during the two cycle pairs A-B and C-D. Opticalprobe PR always collected data on the right hind leg, whereas probe PL was placed on the left hind leg duringcycles A and B of piglet 2 andcycles A and B of piglet 3 (we did not collect data with the optical probe PL on piglet 1, during cycles C-D on piglet 2, and duringcycle C on piglet 3). In all three piglets, the invasive measurement of SvO2 from a femoral vein blood sample was performed at the end of each Fi O2 interval, as shown in Fig. 3. Motion artifacts were minimized in the optical data by securing the piglet's legs to the operating table. The protocol was approved by the Institutional Review Board of the Massachusetts General Hospital, where the piglet experiments were performed.
Measurements on human subjects.
We performed measurements on eight healthy human subjects (6 men and 2 women; mean age of 24.5 yr, age range of 20–35 yr). The subjects sat on a comfortable chair and rested for 10–15 min before the experimental protocol was started. A pneumatic cuff was placed around the right thigh of the subject to later induce a venous occlusion by inflating the cuff to a pressure of 70 mmHg. A pulse oximeter probe (Nellcor, N-200) was placed on the index finger of the left hand. A strain-gauge belt (Sleepmate/Newlife Technologies, Resp-EZ) was placed around the subject's upper abdomen to monitor the respiratory excursion. As in the piglet experiment, we used the analog outputs of the pulse oximeter and strain gauge for continuous coregistration of the physiological and near-infrared data. Two optical probes were placed on the right thigh; the first probe (probe HVM) was positioned on top of a visible superficial vein of the vastus medialis muscle, and the second probe (probe HVL) was placed on the vastus lateralis muscle, far from visible superficial veins. During the measurements, we asked the subject to breathe regularly, following a metronome whose frequency was set to the average breathing rate of the subject at rest (typically 14–15 breaths/min). During the whole experiment, the subject was asked to breathe at the same frequency as the metronome pace. No subjects experienced any discomfort or difficulties with this procedure. The measurement protocol consisted of 2 min of baseline (we acquired 1,535 optical data points at 80 ms/point), followed by 40 s of venous occlusion, and a final recovery period of a few minutes. A few subjects performed an additional exercise routine to test the effect of exercise on the measured value of SvO2-NIRSresp on the muscle. The exercise consisted of raising the right foot, voluntarily contracting the leg muscles (isometric contraction), until the subject felt tired. The human study was approved by the Institutional Review Board of Tufts University, where the human experiments were performed; all subjects gave their written, informed consent.
Near-infrared data processing for the measurement of SvO2.
We used a modified Beer-Lambert law approach (14) to translate the temporal intensity ratio collected at each wavelength [I(λ, t)/I(λ, 0), whereI is intensity, λ is wavelength, and t is time] at a distance of 1.0 cm from the illumination point, into a time variation in the tissue absorption [Δμa(λ, t), where μa is the tissue absorption coefficient]. This approach was implemented by applying the following equation (14) Equation 1where L eff is the effective optical pathlength from the illuminating point to the light collection point. We measured L eff by quantifying μa and the reduced scattering coefficient (μ′s) using the frequency-domain multidistance method (17). The diffusion-theory relationship that gives L eff in terms of μa, μ ′, and the source-detector separation (r) in a semi-infinite turbid medium (where the illumination and collection points are at the boundary of the turbid medium) is the following (17) Equation 2More details on this hybrid frequency-domain [to measureL eff(λ)] and continuous wave (modified Beer-Lambert law) approach are given in Refs. 14,17, 21, and 22. Equation 2 shows that for typical values of the near-infrared μa and μ′s, say μa = 0.1 cm−1 and μ′s = 10 cm−1, the value of L eff is ∼5.5 cm for r= 1 cm. The multi-distance scheme was implemented by considering the data collected by the two fiber bundles located at two different distances (1.0 and 2.0 cm) from the source fiber bundle. At these source detector distances, the diffusion regime of light propagation in tissues is already established (18). As an alternative to the diffusion equation model to describe the spatial dependence of the optical signal, empirical approaches have been proposed (6). The different sensitivity of the two detector channels was accounted for by a preliminary calibration measurement on a synthetic tissue-like sample. The applicability of the initial calibration to the whole data set was verified at the end of each measurement session by repositioning the optical probe on the calibration sample. We typically reproduced the calibration values of the block optical coefficients to within 10%. In the piglet experiments, we updated the measured values ofL eff at each wavelength every time the Fi O2 was changed. Specifically,L eff was computed, according to Eq.2 , from average measurements of μa and μ′s over the last 80 s of each period corresponding to a specific Fi O2value. For the measurements on human subjects, we computed an initial value of L eff (at each wavelength) over the first 2 min of baseline, and we used this value for the analysis of the data over the whole measurement session. The long integration time for the mean pathlength measurements (80 s in the piglet experiment, 120 s in the human subjects measurements) realized a low-pass filter that minimized the time-varying contributions from the Hb oscillations caused by the arterial pulsation and breathing. Furthermore, the two-distance measurement scheme for the mean pathlength measurement also provided some level of spatial averaging. In contrast, the optical data for the measurement of SvO2 were acquired with an 80-ms temporal resolution and with the use of a single source-detector distance (1 cm).
To measure the SvO2, we followed a two-step procedure. First, we computed the amplitude of the absorption oscillations at the respiratory frequency at each of the eight wavelengths considered. Second, we fit the spectrum of the experimental absorption amplitude with the hemoglobin absorption spectrum. We have used two alternative methods to quantify the absorption oscillations at the respiratory frequency. The first method is based on the fast Fourier transform (FFT) of Δμa(t). The sum of the amplitudes of the FFT of Δμa over the respiratory frequency band yields a measure of the amplitude of the respiration-induced absorption oscillations. This method assumes that the Fourier spectrum of Δμa clearly shows a discernable peak at the respiratory frequency. The second method is based on a band-pass (BP) filter of Δμa(t) and on a modeling algorithm (MA) (sine-wave fit). The BP filter serves the purpose of isolating the absorption oscillations at the respiratory frequency by suppressing higher and lower frequency components in Δμa(t). The MA consists of fitting a sine wave to Δμa(BP) over each respiratory cycle. The amplitude of the fitted sine wave gives an estimate of the absorption oscillation amplitude at the respiratory frequency. As a result, the second method (BP + MA) achieves a reading of SvO2 from each individual respiration cycle, whereas the first method (FFT) requires multiple respiration cycles to produce a SvO2 reading. Both methods provide phase readings that can be used to verify that the respiration-induced absorption oscillations at different wavelengths are in phase with each other. We indicate the SvO2 measurement according to the FFT and BP + MA methods with SvO2-NIRSresp(FFT) and SvO2-NIRSresp(BP), respectively.
In the piglet experiments, we evaluated the FFT of Δμa over 256 data points, corresponding to a time trace of 20.5 s, to achieve reliable spectra from a number of breathing periods (typically 13–16). Furthermore, we averaged about 800 successive FFTs (each computed from a data set shifted by one data point with respect to the previous one), so that the total number of data points resulting in a single SvO2 reading was on the order of 1,000, corresponding to a train of data 80 s long. This 80-s-long data set was chosen to be at the end of each Fi O2 period, and it coincides with the 80-s period over which we measured L eff. In the human subject experiment, we used 512 points for the FFT because the breathing frequency was lower (0.22–0.26 Hz) than that of the piglets (0.6–0.9 Hz) and we wanted to have a similar number of breathing periods. As in the piglets experiment, we averaged the results from multiple (500–1,000) successive FFTs.
The spectrum of the amplitude of the absorption oscillations at the respiratory frequency [Δμ (λi)] was fitted with a linear combination of the HbO2 and Hb extinction spectra, ɛ (λi)Δ[HbO2]resp+ ɛHb(λi)Δ[Hb]resp, where ɛ (λi) and ɛHb(λi) are the extinction coefficients of HbO2 and Hb, respectively (43,53). The fitting parameters were the amplitudes of the oscillatory concentration of oxyhemoglobin (Δ[HbO2]resp) and deoxyhemoglobin (Δ[Hb]resp) at the respiratory frequency. The minimization of the sum of the squares of the residuals, i.e., Σi[Δμ (λi) − Δμ (λi)]2, yields a linear system whose solution gives the following best fit concentrations of amplitude of the oscillatory [HbO2] and [Hb] (11) Equation 3 Equation 4The oxygen saturation of the hemoglobin compartment oscillating synchronously with respiration (SvO2-NIRSresp) is then given by Equation 5It is important to note that for the determination of SvO2-NIRSresp one only needs to knowL eff to within a wavelength-independent factor. In fact, Eq. 5 shows that a common, wavelength-independent multiplicative factor in Δμa(λi) cancels out in the expression for SvO2-NIRSresp. In contrast, the wavelength dependence of L eff is important for the measurement of SvO2 with our method, and this is why we have opted to measure L eff at each wavelength using the multidistance, frequency-domain technique. It is also important to observe that our method requires 1) oscillations of μa at the respiratory frequency to be reliably attributed to hemoglobin (and not, for instance, to motion artifacts), 2) the hemoglobin concentration fluctuations to result from the volume oscillation of a hemoglobin compartment rather than from periodic fluctuations in the blood flow, and 3) the fluctuating hemoglobin compartment responsible for the measured Δμa to be mainly the venous compartment. In our measurements, we have considered each one of the three above points. The assignment of the absorption oscillations to hemoglobin (point 1) was done by requiring that the hemoglobin spectrum fits the absorption data relatively well. To this aim, we requested that the average absolute value of the relative residuals, defined as ɛfit = 1/NΣ ‖ Δμ (λi) − Δμ (λi) ‖ / Δμ (λi), where N is the number of wavelengths considered, be at most twice the experimental percent error in Δμ . We also used the standard deviation of the SvO2-NIRSresp(FFT) values obtained with the 800 (piglet experiment) or 500–1,000 (human experiment) successive FFTs to estimate the error in SvO2-NIRSresp(FFT). We discarded the cases having a standard deviation error in SvO2-NIRSresp greater than 15%. The assignment of the absorption oscillations to volume rather than blood flow fluctuations (point 2) is achieved by verifying that the absorption oscillations at the eight wavelengths are in phase. In fact, blood flow fluctuations induce out-of-phase oscillations in the [HbO2] and [Hb] (because of the increased rates of inflow of HbO2 and washout of Hb), as opposed to the in-phase oscillations of HbO2 and Hb that result from volume pulsations. The third point, namely the requirement that the absorption oscillations at the respiratory frequency are representative of venous blood, is investigated by 1) comparing the SvO2-NIRS from the respiratory hemoglobin oscillations (SvO2-NIRSresp) with the corresponding values measured by gas analysis of SvO2-blood (piglet experiments) or by the NIRS venous occlusion method (SvO2-NIRSvo) (human subject experiments),2) studying the effect on the [Hb] and [HbO2] oscillations at the respiratory frequency of a venous occlusion induced between the lungs and the peripheral measurement area (the thigh muscles in human subject experiments), and3) by recording the effect of muscle exercise on the near-infrared measurements of SvO2[SvO2-NIRSresp(BP)] in human subjects.
Figure 4 reports average spectra ofL eff measured for a source-detector separation of 1 cm. Figure 4 A refers to piglet measurements conducted at two different values of Fi O2, whereas Fig. 4 B refers to human measurements with probes HVM and HVL. The error bars in Fig. 4 represent the standard deviations over multiple measurements (multiple Fi O2 cycles and piglets for Fig.4 A and multiple subjects for Fig. 4 B).
In the piglet experiment, we discarded 11 (from a total of 67) SvO2-NIRSresp(FFT) measurements because the standard deviation over 800 FFTs exceeded 15%. These discarded SvO2-NIRSresp(FFT) readings occurred as follows: one (of 8) in piglet 1, two (of 26) in piglet 2, and eight (of 33) in piglet 3. One discarded reading was assigned to motion artifacts, whereas the other ten discarded measurements all occurred at low-Fi O2values (10–17.5%) corresponding to SvO2-blood values of 20–50%. We were not able to apply the BP method topiglet 2 and to the Fi O2 cycles A and B of piglet 3 because of irregular absorption oscillation waveforms that were not reliably processed by the BP + MA approach.
Figure 5 shows typical temporal traces of the relative [HbO2] and [Hb] measured on the piglet's leg (with optical probe PR) (Fig. 5 A) and on the human vastus medialis muscle at rest (Fig. 5 B) and during venous occlusion on the upper thigh (optical probe HVM) (Fig. 5 C). The temporal traces of [Hb] and [HbO2] are obtained by fitting the measured spectrum of Δμa(λ,t) (whose value at each wavelength was obtained from Eq. 2 ) with a linear combination of the HbO2 and Hb extinction spectra. This procedure results in the application of Eqs. 3 and 4 without the superscript “resp” on Δμa, Δ[HbO2], and Δ[Hb]. Two oscillatory components are clearly visible in the relative [HbO2] and [Hb] traces of Fig. 5 A: the first one, associated with the heartbeat (as shown by the pulse oximeter data; top trace in Fig. 5) is at a frequency of ∼2.5 Hz, whereas the second one, associated with respiration (as shown by the strain gauge signal; second trace from the top in Fig. 5), is at a frequency of ∼0.65 Hz. Only the latter oscillatory component (at a frequency of ∼0.23 Hz in human subjects) is clearly visible in Fig. 5 B, whereas neither is present in Fig. 5 C. Figure 5,B and C, shows additional low-frequency oscillations associated with changes in blood pressure and heart rate. We observe that the strain-gauge signal (second trace from the top in Fig. 5) increases during inspiration and decreases during expiration. The BP filter described in the previous section aims at isolating the oscillatory component at the respiratory frequency by filtering out higher and lower frequency components. The relative [HbO2] and [Hb] traces after BP filtering are shown in Fig. 5, bottom. In the case reported in Fig. 5, which is representative of the results reported in this article for SvO2-NIRSresp, the oscillatory components of [HbO2] and [Hb] at the respiratory frequency are in phase with each other and disappear during venous occlusion.
Figure 6 illustrates representative Δμ spectra measured on the piglet's leg (probe PR) (Fig. 6 A) and on the human vastus medialis muscle at rest (Fig. 6 B) and during venous occlusion on the upper thigh (probe HVM) (Fig.6 C). The y-axis of each panel of Fig. 6 refers to the values of Δμ obtained with the BP filter method. The values of Δμ computed with the FFT method are normalized by a wavelength-independent factor to match the BP value of Δμ at 636 nm. The relatively high value of εfitduring venous occlusion (Fig. 6 C) is an indication of the poor fit, which in turn results from the lack of hemoglobin oscillations at the respiratory frequency (see Fig. 5 C,bottom, and the y-axis values of Fig.5 C compared with those of Fig. 5 B). Figure 6 also shows the best fit of the hemoglobin absorption spectrum to the BP Δμ and to the FFT Δμ . The best-fit hemoglobin spectra represent the oxygen saturation of hemoglobin, as illustrated in Fig.1. The value of SvO2-NIRSresp is given byEq. 5 .
Figure 7 compares the measurements of SvO2-NIRSresp(BP), SvO2-NIRSresp(FFT), and SvO2-blood during cycle A of piglet 1 and during cycle C of piglet 3. The SvO2-NIRSresp(BP) traces reported in Fig.7 were obtained by performing a running average of the breath-to-breath values obtained with the BP method. In Fig. 7, the averaging procedure consists of a 5-point (in Fig. 7 A) or 15-point (in Fig.7 B) running average. The assessment of the agreement between the measurements of SvO2-NIRSresp(FFT) and SvO2-blood in the full piglet study is carried out according to the procedure described by Bland and Altman (5). Figure 8 Aplots the results of the NIRS method based on the respiratory oscillations of the tissue absorption against the invasive measurement of SvO2-blood. The shape of the symbols in Fig.8 A indicates the piglet number, whereas the type of fill indicates the location of the NIRS measurement. The range of SvO2-blood values considered in this study is ∼20–95%. The error bars in Fig. 8 A are the standard deviations (SD) computed from the results of ∼800 successive FFTs (as described in materials and methods). Figure 8 Bdisplays the difference between the two readings vs. their average, and it quantifies the discrepancy between the two methods and the possible dependence of such a difference on the level of SvO2. The mean difference between SvO2-NIRSrespand SvO2-blood over the full oxygenation range considered in this study is 1.0% (a measurement of the bias of the SvO2-NIRSresp measurement), and the SD of the difference is 5.8%. Figure 8 B does not show any striking dependence of the difference on the mean. If we take the values of mean difference ± 2 SD as the limits of agreement of the two methods (5), we get an estimate of the maximal discrepancies between SvO2-NIRSresp(FFT) and SvO2-blood of −10.6% and +12.6%.
In the human experiment, we found that the NIRS values of SvO2 measured with probe HVM (placed on top of a visible vein) were typically smaller than those measured withprobe HVL (placed far from any visible vein). Furthermore, the amplitude of the oscillatory absorption at the respiration (heartbeat) frequency was typically greater (smaller) for the data collected with probe HVM than with probe HVL. Of the 16 SvO2-NIRSresp measurements (8 subjects, 2 locations), we discarded only 2 measurements (because of a value of Σfit greater than twice the error in Δμ ), both collected withprobe HVL. Figure 9 compares the SvO2-NIRSresp values measured in the human subjects at rest in which the FFT method and the BP filtering approach were used. Figure 9 A shows the good agreement of the two measurements, and Fig. 9 B quantifies the average difference (0.9%) and the maximum discrepancies of −5.1 and +6.9%, as given by the mean ± 2 SD of the differences. Figure10 reports a similar comparison between SvO2-NIRSresp(FFT) and SvO2-NIRSvo. As described by Yoxall and Weindling (56), under the assumption that a venous occlusion induces an initial increase in the venous blood volume, SvO2-NIRSvo is given by [Hb˙O2]0/[Hb˙ −T]0, where the dots indicate a time derivative and the subscript 0 indicates the initial time that immediately follows the onset of venous occlusion. The agreement between SvO2-NIRSresp(FFT) and SvO2-NIRSvo is good, with an average deviation of 0.8% and maximum discrepancies of −4.2 and +5.8%. Two horizontal lines in Figs. 8 B and 9 B indicate the range given by the mean difference ± 2 SD. The maximum discrepancy among SvO2-NIRSresp(FFT), SvO2-NIRSresp(BP), and SvO2-NIRSvo is less than the maximum deviation between SvO2-NIRSresp(FFT) and SvO2-blood found in piglets (see Fig. 8 B).
The effect of muscle exercise on the measurement of SvO2-NIRSresp(BP) on top of a visible superficial vein (probe HVM) is illustrated in Fig.11. Although SaO2(measured with a pulse oximeter) is unaffected by the exercise, SvO2-NIRSresp(BP) shows a significant postexercise decrease from a baseline value of 75–78% down to a minimum value of ∼54%. The recovery to the baseline value of SvO2-NIRSresp(BP) occurs after ∼30 s. By using the BP approach, we could monitor SvO2-NIRSresp at every breathing period, i.e., every ∼5 s, thus achieving a real-time monitoring of SvO2. We observe that we could not obtain meaningful measurements of SvO2-NIRSresp during exercise because of motion artifacts.
Various methods for measuring SvO2.
The method presented in this article to measure SvO2from the near-infrared absorption oscillations at the respiratory frequency (spiroximetry) can be implemented by using a FFT or a digital BP filter in conjunction with a MA. We have indicated the measurements of SvO2 obtained with these two approaches with the notations SvO2-NIRSresp(FFT) and SvO2-NIRSresp(BP), respectively. An alternative method for measuring SvO2 with NIRS is based on a previously described venous occlusion protocol (13,39, 55, 56). We have identified the results of this measurement procedure with the notation SvO2-NIRSvo. In the human study, the NIRS measurements were conducted at two locations on the thigh. One location was on top of a visible superficial vein of the vastus medialis muscle (probe HVM), and the second location was far from visible superficial veins on the vastus lateralis muscle (probe HVL). Finally, the invasive measurement of SvO2 performed by the gas analysis of venous blood samples is indicated with SvO2-blood. In this section, we discuss the different features of these measurements of SvO2, and the comparison of their results, as reported in Figs. 7-9.
The FFT and BP filter approaches to near-infrared spiroximetry.
The major advantage of the BP approach is that it allows for a real-time measurement of SvO2 by providing a reading of SvO2-NIRSresp(BP) at every respiration cycle. Consequently, this method is particularly effective during transients, as illustrated by the recovery of the SvO2-NIRSresp(BP) traces corresponding to the sudden increase of Fi O2 to 100% in piglets (see Fig. 6, A and B), or to the end of the exercise period in human subjects (see Fig. 11). On the other hand, the BP filter + MA method is susceptible to fluctuations in the respiratory frequency and to irregular respiration patterns. This accounts for the fact that we did not get reliable readings of SvO2-NIRSresp(BP) in piglet 2and in Fi O2 cycles A andB of piglet 3. The FFT method was more robust, producing reliable readings in 56 of 67 cases (84%) in the piglet study and in 14 of 16 cases (87%) in the human study. It is important to observe that 10 of the 11 discarded readings in piglets occurred at low-SvO2-NIRS-blood values (20–50%), and one was assigned to motion artifacts. Both discarded readings in the human study were collected with probe HVL, which was placed far from visible veins. Therefore, we have found indications that the measurement of SvO2-NIRSresp(FFT) is particularly robust at SvO2 values >50% (in piglets) and when the optical probe is placed on top of a visible superficial vein (in human subjects). Although the FFT method, which is based on a measurement of the integrated peak at the respiratory frequency, is less sensitive than the BP method to irregular respiration patterns, it is not applicable during transients. In fact, we did not obtain reliable readings of SvO2 when the time frame used to compute SvO2-NIRSresp(FFT) (80 s in piglets, 80–120 s in human subjects) included significant changes in the SvO2. When both the FFT and the BP methods can be applied, they provide SvO2-NIRSrespmeasurements that are in excellent agreement, as shown in Figs. 6 and9. The differences between the two measurements (SD of 3.0%) are comparable with measurement errors and significantly less than the maximum deviation between SvO2-NIRSresp(FFT) and SvO2-blood (approximately ±10%) observed in the piglet study (see Fig. 8 B).
Measurements of SvO2-NIRSresp and SvO2-NIRSvo.
Both of these NIRS methods to measure the SvO2(resp and vo) rely on a change in the volume fraction of venous blood in the tissue. The two major differences between the two methods are as follows. 1) The vo method requires an external perturbation consisting of a pneumatic-cuff-induced venous occlusion, whereas the resp method is only based on the intrinsic blood pressure oscillations induced by normal respiration and can be applied continuously.2) The vo method can be applied only to limbs, whereas the resp method can, in principle, be applied to any tissue and in particular to the brain, as already shown by Wolf et al. (52). However, we stress that it is always important to verify that the [HbO2] and [Hb] oscillate in phase at the respiratory frequency for the resp method to provide reliable measurements of SvO2. For instance, Elwell et al. (16) reported out-of-phase oscillations of [Hb] and [HbO2] in the human brain, which would indicate a blood flow rather than volume oscillations, thus rendering the resp method inapplicable. In our human study, we found an excellent agreement between SvO2-NIRSresp(FFT) and SvO2-NIRSvo, with a maximum deviation on the order of ±4–5% (see Fig. 10).
Optical probes PR, PL, HVM, and HVL.
In the piglet study, we have found no significant difference between the SvO2-NIRSresp(FFT) data collected with probes PR (on the right leg) andPL (on the left leg, where the venous catheter was inserted) (see Fig. 8 A). This result indicates that noninvasive measurements of SvO2 on one leg can be meaningfully compared with invasive measurements of SvO2 on the other leg. In the human study, we found some differences between the SvO2-NIRS measurements with probe HVM(placed on top of a visible superficial vein in the vastus medialis muscle) and with probe HVL (placed far from visible veins on the vastus lateralis muscle). As shown in Figs. 8 and 9, the SvO2-NIRS readings (with both the NIRSrespand NIRSvo method) of probe HVM (see Figs. 8 and9) were typically smaller than the readings of probe HVL(see Figs. 8 and 9). We assign this result to a partial contribution from the capillary and/or arterial compartments picked up byprobe HVL. In fact, although the optical data from probe HVM shown in Fig. 4, B and C, do not show any visible contribution from the arterial pulsation, data from probe HVL (not shown) do contain pulsatile components at the heartbeat frequency. As a result, we believe that the optical probe should be placed on top of visible superficial veins for more accurate readings of SvO2-NIRSresp on human subjects. We believe that the reason that SvO2-NIRSrespreadings in the piglet study were in close agreement with the invasive measurement of SvO2, despite the evident arterial pulsation in Fig. 5 A, is related to the smaller extent of respiratory sinus arrhythmia in piglets with respect to humans. In fact, respiratory sinus arrhythmia is the main origin of the arterial oscillations at the respiratory frequency (38). The larger role played by respiratory sinus arrhythmia in human subjects with respect to piglets will probably require a more careful interpretation of the optical data for spiroximetry. However, the results of Fig. 11show the practical applicability of spiroximetry to human subjects, so that we do not expect respiratory sinus arrhythmia to introduce an intrinsic limitation of the method.
Noninvasive vs. invasive measurements of SvO2.
The comparison between SvO2-NIRSresp(FFT) and SvO2-blood in the piglet study shows a maximum deviation range of −10.6% to +12.6%. The local character of the SvO2 (as opposed to the systemic nature of the SaO2) requires some caution in the comparison of invasive (SvO2-blood) and noninvasive (SvO2-NIRS) measurements of SvO2. In fact, in our piglet study, SvO2-blood was measured from blood samples drawn from the femoral vein, whereas SvO2-NIRSresp was measured with an optical probe placed on the leg muscle. It is likely that the NIRS oscillatory signal (at the respiratory frequency) is not just representative of the femoral vein and may therefore be indicative of the oxygen consumption at different tissue areas than those affecting the femoral vein saturation. This fact may not lead to significant differences under rest conditions, but it may be important under stress. Although we found a good agreement between SvO2-NIRSresp(FFT) and SvO2-blood over the whole range of Fi O2 values considered (see Fig. 8), we observed a meaningfully greater SD of the differences over the 20–55% SvO2-blood range (SD = 7.8%) than in the 55–95% range (SD = 3.6%).
Effect of muscle exercise on SvO2-NIRSresp(BP).
The result reported in Fig. 11 serves the purpose of further illustrating the potential of the SvO2-NIRSresp(BP) measurement approach. In fact, Fig. 11 shows the feasibility of monitoring the SvO2 in real time on a breath-to-breath basis (one data point every 4–5 s). Furthermore, the baseline SvO2-NIRSresp(BP) value of 75–78% and the exercise-induced drop indicate the venous origin of the saturation measurement, since the SaO2 measurement provided by the pulse oximeter stayed constant at 98 ± 1% for the whole measurement period. On the other hand, Fig. 11 reports only one representative case, and more studies are required to quantify the effect of muscle exercise on the measurement of SvO2-NIRSresp.
In conclusion, we have presented a noninvasive approach to measure the SvO2 in tissues from the near-infrared spectrum of the amplitude of respiration-induced absorption oscillations. We have implemented this approach, which we call near-infrared spiroximetry, by processing the optical data with a FFT method or with a digital BP filter in conjunction with a MA. More sophisticated data processing schemes may further improve the effectiveness and the range of applicability of spiroximetry. The SvO2 measurements reported in this article complement previously demonstrated NIRS measurements of StO2 (8, 27) and SaO2 (21). Therefore, our results may lead to the design of a noninvasive optical instrument capable of providing simultaneous and real-time measurements of local SaO2, StO2, and SvO2.
We thank Aradhana Arora, Matthew Hoimes, and Tanya Fridman for technical assistance during the preliminary measurements on human subjects and Dennis Hueber and Valentina Quaresima for helpful discussions. We also thank Enrico Gratton for lending us the eight-wavelength laser board used in this study. We are grateful to the volunteers who participated in this study.
This research is supported by the National Institutes of Health Grants R01-MH-62854 (to M. A. Franceschini) and R29-NS-38842 (to D. A. Boas) and by the US Army Awards DAMD17-99-2-9001 (to D. A. Boas) and DAMD17-99-1-9218 (to S. Fantini). D. A. Boas acknowledges financial support from the Center for Innovative Minimally Invasive Therapies.
The material presented does not necessarily reflect the position or the policy of the U.S. Government, and no official endorsement should be inferred.
Address for reprint requests and other correspondence: M. A. Franceschini, Bioengineering Center, Dept. of Electrical Engineering and Computer Science, Tufts Univ., 4 Colby St., Medford, MA 02155-6013 (E-mail:).
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
- Copyright © 2002 the American Physiological Society